求助:所有算子都是运行在BPU上,模型也不是很复杂,但是实测出来的模型性能不到2hz

模型转换配置文件如下:

model_parameters:
  onnx_model: '../lraspp_performance_2.onnx'
  march: 'bayes-e'
  output_model_file_prefix: 'lraspp_performance_2'
  working_dir: '../hb_mapper_output_2/lraspp_performance'
  # 实际部署测试必须关掉
  layer_out_dump: False

  node_info: {
    "Resize_220": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
  }

input_parameters:
  input_name: 'input'
  # 当前最新模型真实输入不是纯 RGB,而是 5 通道 featuremap:
  #   0~2: RGB,已做 (x/255 - mean) / std
  #   3  : radar_range,先按 sensor_depth_max 解码到米制,再按 [depth_min, depth_max]
  #        截断并线性归一化到 [0, 1]
  #   4  : radar_valid_mask,取值 {0, 1}
  input_type_train: 'featuremap'
  input_layout_train: 'NCHW'
  input_type_rt: 'featuremap'
  input_layout_rt: 'NCHW'
  input_shape: '1x5x360x640'
  input_batch: 1
  # featuremap 输入要求上板前已经完成全部前处理,因此这里不能再走 rgb/nv12 预处理链
  norm_type: 'no_preprocess'

calibration_parameters:
  # 预处理后的5通道 float32 featuremap 校准数据
  # 数据口径必须与 input_type_train/input_layout_train/input_shape 完全一致
  # 即每个样本都是 row-major 的 float32 二进制,shape = 1x5x360x640 或 5x360x640
  cal_data_dir: '../hb_calibration_data_f32'
  cal_data_type: 'float32'
  preprocess_on: False
  per_channel: True
  # 取值: 'default' 'skip' 'ml'
  calibration_type: 'max'  
  # 取值: 0.99999 0.99995 0.9999 0.9995 0.999
  max_percentile: 0.9999

compiler_parameters:
  compile_mode: 'latency'
  # 部署关闭
  debug: False
  # X5仅支持1
  core_num: 1
  # 部署阶段,为保证最佳运行性能,设置为O3
  optimize_level: 'O3'
  # featuremap 输入通常从 DDR 喂入
  input_source: {'input': 'ddr'}

makertbin.log日志如下:

2026-04-27 18:47:24,661 file: tool_utils.py func: tool_utils line No: 77 log will be stored in /open_explorer/test_lraspp_model_0427/hb_mapper_makertbin.log
2026-04-27 18:47:24,661 file: hb_mapper.py func: hb_mapper line No: 132 Start hb_mapper....
2026-04-27 18:47:24,661 file: hb_mapper.py func: hb_mapper line No: 133 hbdk version 3.49.15
2026-04-27 18:47:24,661 file: hb_mapper.py func: hb_mapper line No: 134 horizon_nn version 1.1.0
2026-04-27 18:47:24,661 file: hb_mapper.py func: hb_mapper line No: 135 hb_mapper version 1.24.3
2026-04-27 18:47:24,661 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 530 Start Model Convert....
2026-04-27 18:47:24,666 file: mapper_conf_parser.py func: mapper_conf_parser line No: 105 validating model_parameters...
2026-04-27 18:47:24,666 file: mapper_conf_parser.py func: mapper_conf_parser line No: 1347 Using abs path /open_explorer/test_lraspp_model_0427/lraspp_performance_2.onnx
2026-04-27 18:47:24,666 file: mapper_conf_parser.py func: mapper_conf_parser line No: 260 Using onnx model file: /open_explorer/test_lraspp_model_0427/lraspp_performance_2.onnx
2026-04-27 18:47:24,681 file: onnx_parser.py func: onnx_parser line No: 39 Model input names: ['input']
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 264 Model has 1 inputs according to model file
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 1347 Using abs path /open_explorer/test_lraspp_model_0427/hb_mapper_output_2/lraspp_performance
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 287 working_dir does not exist. Creating working_dir: /open_explorer/test_lraspp_model_0427/hb_mapper_output_2/lraspp_performance
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 438 node_dict: {self.node_dict}
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 119 validating model_parameters finished
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 123 validating input_parameters...
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 135 validating input_parameters finished
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 139 validating calibration_parameters...
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 1347 Using abs path /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32
2026-04-27 18:47:24,682 file: mapper_conf_parser.py func: mapper_conf_parser line No: 1013 The calibration dir name suffix is the same as the value float32 of the cal_data_type parameter and will be read with the value of cal_data_type.
2026-04-27 18:47:24,683 file: mapper_conf_parser.py func: mapper_conf_parser line No: 155 validating calibration_parameters finished
2026-04-27 18:47:24,683 file: mapper_conf_parser.py func: mapper_conf_parser line No: 159 validating custom_op...
2026-04-27 18:47:24,683 file: mapper_conf_parser.py func: mapper_conf_parser line No: 1076 custom_op does not exist, skipped
2026-04-27 18:47:24,683 file: mapper_conf_parser.py func: mapper_conf_parser line No: 165 validating custom_op finished
2026-04-27 18:47:24,683 file: mapper_conf_parser.py func: mapper_conf_parser line No: 168 validating compiler_parameters...
2026-04-27 18:47:24,683 file: mapper_conf_parser.py func: mapper_conf_parser line No: 183 validating compiler_parameters finished
2026-04-27 18:47:24,683 file: mapper_conf_parser.py func: mapper_conf_parser line No: 187 validating deprecated parameters...
2026-04-27 18:47:24,683 file: mapper_conf_parser.py func: mapper_conf_parser line No: 193 validating deprecated parameters finished
2026-04-27 18:47:24,683 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 54 Dump config:
2026-04-27 18:47:24,683 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 55 calibration_parameters:
  cal_data_dir: ../hb_calibration_data_f32
  cal_data_type: float32
  calibration_type: max
  max_percentile: 0.9999
  per_channel: true
  preprocess_on: false
compiler_parameters:
  compile_mode: latency
  core_num: 1
  debug: false
  input_source:
    input: ddr
  optimize_level: O3
input_parameters:
  input_batch: '1'
  input_layout_rt: NCHW
  input_layout_train: NCHW
  input_name: input
  input_shape: 1x5x360x640
  input_type_rt: featuremap
  input_type_train: featuremap
  norm_type: no_preprocess
model_parameters:
  layer_out_dump: false
  march: bayes-e
  node_info:
    Resize_220:
      InputType: int8
      'ON': BPU
      OutputType: int8
  onnx_model: ../lraspp_performance_2.onnx
  output_model_file_prefix: lraspp_performance_2.onnx
  working_dir: ../hb_mapper_output_2/lraspp_performance

2026-04-27 18:47:24,688 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 60 input 'input' : original model shape: [1, 5, 360, 640]
2026-04-27 18:47:24,689 file: loader.py func: loader line No: 204 *******************************************
2026-04-27 18:47:24,689 file: loader.py func: loader line No: 205 First calibration picture name: scene_001_23.bin
2026-04-27 18:47:24,689 file: loader.py func: loader line No: 207 First calibration picture md5:
2026-04-27 18:47:24,698 file: loader.py func: loader line No: 211 *******************************************
2026-04-27 18:47:24,699 file: loader.py func: loader line No: 282 created RawImageDirLoader of shape:[1, 5, 360, 640]
2026-04-27 18:47:24,699 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_001_23.bin
2026-04-27 18:47:24,702 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_001_38.bin
2026-04-27 18:47:24,704 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_002_31.bin
2026-04-27 18:47:24,706 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_002_33.bin
2026-04-27 18:47:24,710 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_003_43.bin
2026-04-27 18:47:24,712 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_003_57.bin
2026-04-27 18:47:24,714 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_004_44.bin
2026-04-27 18:47:24,716 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_004_47.bin
2026-04-27 18:47:24,718 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_005_11.bin
2026-04-27 18:47:24,720 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_005_28.bin
2026-04-27 18:47:24,723 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_006_00.bin
2026-04-27 18:47:24,725 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_006_17.bin
2026-04-27 18:47:24,727 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_006_48.bin
2026-04-27 18:47:24,729 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_007_14.bin
2026-04-27 18:47:24,730 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_007_24.bin
2026-04-27 18:47:24,732 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_008_32.bin
2026-04-27 18:47:24,734 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_008_39.bin
2026-04-27 18:47:24,736 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_009_15.bin
2026-04-27 18:47:24,738 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_009_50.bin
2026-04-27 18:47:24,739 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_011_12.bin
2026-04-27 18:47:24,741 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_011_17.bin
2026-04-27 18:47:24,743 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_012_00.bin
2026-04-27 18:47:24,744 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_012_05.bin
2026-04-27 18:47:24,746 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_013_08.bin
2026-04-27 18:47:24,748 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_013_10.bin
2026-04-27 18:47:24,749 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_013_46.bin
2026-04-27 18:47:24,751 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_014_36.bin
2026-04-27 18:47:24,753 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_014_43.bin
2026-04-27 18:47:24,755 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_014_51.bin
2026-04-27 18:47:24,756 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_015_13.bin
2026-04-27 18:47:24,758 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_015_28.bin
2026-04-27 18:47:24,760 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_015_47.bin
2026-04-27 18:47:24,762 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_016_42.bin
2026-04-27 18:47:24,765 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_016_53.bin
2026-04-27 18:47:24,767 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_017_27.bin
2026-04-27 18:47:24,769 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_017_51.bin
2026-04-27 18:47:24,772 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_017_56.bin
2026-04-27 18:47:24,774 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_018_20.bin
2026-04-27 18:47:24,776 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_018_32.bin
2026-04-27 18:47:24,777 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_019_19.bin
2026-04-27 18:47:24,779 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_019_42.bin
2026-04-27 18:47:24,781 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_019_44.bin
2026-04-27 18:47:24,783 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_021_46.bin
2026-04-27 18:47:24,784 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_021_56.bin
2026-04-27 18:47:24,786 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_021_58.bin
2026-04-27 18:47:24,787 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_022_25.bin
2026-04-27 18:47:24,789 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_022_27.bin
2026-04-27 18:47:24,791 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_022_40.bin
2026-04-27 18:47:24,793 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_023_13.bin
2026-04-27 18:47:24,795 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_023_32.bin
2026-04-27 18:47:24,797 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_024_21.bin
2026-04-27 18:47:24,799 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_024_27.bin
2026-04-27 18:47:24,801 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_025_08.bin
2026-04-27 18:47:24,802 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_025_46.bin
2026-04-27 18:47:24,805 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_026_28.bin
2026-04-27 18:47:24,806 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_026_30.bin
2026-04-27 18:47:24,808 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_027_00.bin
2026-04-27 18:47:24,811 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_027_24.bin
2026-04-27 18:47:24,813 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_027_36.bin
2026-04-27 18:47:24,815 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_028_17.bin
2026-04-27 18:47:24,817 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_028_31.bin
2026-04-27 18:47:24,818 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_029_07.bin
2026-04-27 18:47:24,820 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_029_50.bin
2026-04-27 18:47:24,822 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_029_51.bin
2026-04-27 18:47:24,824 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_031_39.bin
2026-04-27 18:47:24,825 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_031_47.bin
2026-04-27 18:47:24,827 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_032_35.bin
2026-04-27 18:47:24,828 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_032_38.bin
2026-04-27 18:47:24,830 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_033_08.bin
2026-04-27 18:47:24,832 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_033_26.bin
2026-04-27 18:47:24,833 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_033_58.bin
2026-04-27 18:47:24,835 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_034_16.bin
2026-04-27 18:47:24,837 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_034_30.bin
2026-04-27 18:47:24,838 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_034_36.bin
2026-04-27 18:47:24,840 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_035_35.bin
2026-04-27 18:47:24,841 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_035_49.bin
2026-04-27 18:47:24,843 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_036_22.bin
2026-04-27 18:47:24,845 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_036_43.bin
2026-04-27 18:47:24,846 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_037_03.bin
2026-04-27 18:47:24,848 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_037_28.bin
2026-04-27 18:47:24,849 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_037_31.bin
2026-04-27 18:47:24,851 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_038_08.bin
2026-04-27 18:47:24,853 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_038_59.bin
2026-04-27 18:47:24,854 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_039_30.bin
2026-04-27 18:47:24,856 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_039_44.bin
2026-04-27 18:47:24,858 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_041_33.bin
2026-04-27 18:47:24,859 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_041_49.bin
2026-04-27 18:47:24,861 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_042_14.bin
2026-04-27 18:47:24,863 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_042_33.bin
2026-04-27 18:47:24,864 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_042_51.bin
2026-04-27 18:47:24,866 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_043_01.bin
2026-04-27 18:47:24,867 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_043_13.bin
2026-04-27 18:47:24,869 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_043_34.bin
2026-04-27 18:47:24,870 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_044_00.bin
2026-04-27 18:47:24,872 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_044_13.bin
2026-04-27 18:47:24,874 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_044_47.bin
2026-04-27 18:47:24,876 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_045_05.bin
2026-04-27 18:47:24,877 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_045_11.bin
2026-04-27 18:47:24,879 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_046_05.bin
2026-04-27 18:47:24,881 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_046_28.bin
2026-04-27 18:47:24,882 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_046_47.bin
2026-04-27 18:47:24,883 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_047_16.bin
2026-04-27 18:47:24,885 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_047_50.bin
2026-04-27 18:47:24,887 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_048_23.bin
2026-04-27 18:47:24,889 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_048_59.bin
2026-04-27 18:47:24,890 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_049_24.bin
2026-04-27 18:47:24,892 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_049_37.bin
2026-04-27 18:47:24,893 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_049_39.bin
2026-04-27 18:47:24,895 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_101_33.bin
2026-04-27 18:47:24,896 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_101_43.bin
2026-04-27 18:47:24,898 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_102_22.bin
2026-04-27 18:47:24,899 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_102_31.bin
2026-04-27 18:47:24,901 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_103_44.bin
2026-04-27 18:47:24,903 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_103_47.bin
2026-04-27 18:47:24,904 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_104_33.bin
2026-04-27 18:47:24,906 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_104_38.bin
2026-04-27 18:47:24,907 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_105_50.bin
2026-04-27 18:47:24,909 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_105_58.bin
2026-04-27 18:47:24,910 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_106_19.bin
2026-04-27 18:47:24,912 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_106_38.bin
2026-04-27 18:47:24,913 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_107_25.bin
2026-04-27 18:47:24,915 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_107_30.bin
2026-04-27 18:47:24,916 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_108_41.bin
2026-04-27 18:47:24,918 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_108_44.bin
2026-04-27 18:47:24,919 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_109_23.bin
2026-04-27 18:47:24,921 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_109_33.bin
2026-04-27 18:47:24,922 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_109_41.bin
2026-04-27 18:47:24,924 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_111_04.bin
2026-04-27 18:47:24,925 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_111_20.bin
2026-04-27 18:47:24,927 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_111_36.bin
2026-04-27 18:47:24,928 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_112_08.bin
2026-04-27 18:47:24,930 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_112_32.bin
2026-04-27 18:47:24,932 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_112_51.bin
2026-04-27 18:47:24,933 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_113_13.bin
2026-04-27 18:47:24,935 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_113_16.bin
2026-04-27 18:47:24,937 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_113_56.bin
2026-04-27 18:47:24,938 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_114_00.bin
2026-04-27 18:47:24,940 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_114_24.bin
2026-04-27 18:47:24,941 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_114_43.bin
2026-04-27 18:47:24,944 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_115_21.bin
2026-04-27 18:47:24,946 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_115_58.bin
2026-04-27 18:47:24,948 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_115_59.bin
2026-04-27 18:47:24,950 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_116_09.bin
2026-04-27 18:47:24,952 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_116_21.bin
2026-04-27 18:47:24,953 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_117_41.bin
2026-04-27 18:47:24,955 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_117_56.bin
2026-04-27 18:47:24,957 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_118_11.bin
2026-04-27 18:47:24,958 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_118_15.bin
2026-04-27 18:47:24,960 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_119_12.bin
2026-04-27 18:47:24,963 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_119_24.bin
2026-04-27 18:47:24,965 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_119_49.bin
2026-04-27 18:47:24,966 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_121_01.bin
2026-04-27 18:47:24,967 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_121_16.bin
2026-04-27 18:47:24,969 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_122_00.bin
2026-04-27 18:47:24,971 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_122_54.bin
2026-04-27 18:47:24,972 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_123_21.bin
2026-04-27 18:47:24,974 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_123_37.bin
2026-04-27 18:47:24,976 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_123_58.bin
2026-04-27 18:47:24,978 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_124_10.bin
2026-04-27 18:47:24,980 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_124_33.bin
2026-04-27 18:47:24,982 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_125_08.bin
2026-04-27 18:47:24,983 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_125_39.bin
2026-04-27 18:47:24,985 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_126_16.bin
2026-04-27 18:47:24,987 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_126_55.bin
2026-04-27 18:47:24,989 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_127_27.bin
2026-04-27 18:47:24,991 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_127_32.bin
2026-04-27 18:47:24,993 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_127_59.bin
2026-04-27 18:47:24,994 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_128_14.bin
2026-04-27 18:47:24,996 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_128_29.bin
2026-04-27 18:47:24,997 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_128_37.bin
2026-04-27 18:47:24,999 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_129_41.bin
2026-04-27 18:47:25,000 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_129_56.bin
2026-04-27 18:47:25,002 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_131_19.bin
2026-04-27 18:47:25,004 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_131_39.bin
2026-04-27 18:47:25,005 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_131_50.bin
2026-04-27 18:47:25,007 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_132_40.bin
2026-04-27 18:47:25,008 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_132_42.bin
2026-04-27 18:47:25,010 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_132_43.bin
2026-04-27 18:47:25,012 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_133_23.bin
2026-04-27 18:47:25,014 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_133_25.bin
2026-04-27 18:47:25,016 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_133_46.bin
2026-04-27 18:47:25,018 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_134_00.bin
2026-04-27 18:47:25,020 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_134_13.bin
2026-04-27 18:47:25,022 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_135_11.bin
2026-04-27 18:47:25,024 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_135_57.bin
2026-04-27 18:47:25,026 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_135_59.bin
2026-04-27 18:47:25,029 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_136_07.bin
2026-04-27 18:47:25,030 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_136_39.bin
2026-04-27 18:47:25,033 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_136_41.bin
2026-04-27 18:47:25,035 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_137_20.bin
2026-04-27 18:47:25,037 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_137_48.bin
2026-04-27 18:47:25,039 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_138_22.bin
2026-04-27 18:47:25,041 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_138_46.bin
2026-04-27 18:47:25,042 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_139_12.bin
2026-04-27 18:47:25,045 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_139_13.bin
2026-04-27 18:47:25,047 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_139_29.bin
2026-04-27 18:47:25,049 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_141_22.bin
2026-04-27 18:47:25,051 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_141_27.bin
2026-04-27 18:47:25,054 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_142_19.bin
2026-04-27 18:47:25,056 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_142_58.bin
2026-04-27 18:47:25,059 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_143_01.bin
2026-04-27 18:47:25,062 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_143_02.bin
2026-04-27 18:47:25,064 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_144_31.bin
2026-04-27 18:47:25,067 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_144_41.bin
2026-04-27 18:47:25,069 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_144_53.bin
2026-04-27 18:47:25,071 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_145_01.bin
2026-04-27 18:47:25,074 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_145_32.bin
2026-04-27 18:47:25,076 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_146_11.bin
2026-04-27 18:47:25,078 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_146_22.bin
2026-04-27 18:47:25,080 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_147_13.bin
2026-04-27 18:47:25,083 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_147_15.bin
2026-04-27 18:47:25,086 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_147_16.bin
2026-04-27 18:47:25,088 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_148_25.bin
2026-04-27 18:47:25,090 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_148_33.bin
2026-04-27 18:47:25,092 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_149_15.bin
2026-04-27 18:47:25,094 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_149_28.bin
2026-04-27 18:47:25,096 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_149_54.bin
2026-04-27 18:47:25,097 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_211_21.bin
2026-04-27 18:47:25,098 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_211_56.bin
2026-04-27 18:47:25,100 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_212_22.bin
2026-04-27 18:47:25,102 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_212_44.bin
2026-04-27 18:47:25,104 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_212_53.bin
2026-04-27 18:47:25,106 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_213_17.bin
2026-04-27 18:47:25,109 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_213_37.bin
2026-04-27 18:47:25,111 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_213_59.bin
2026-04-27 18:47:25,112 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_214_01.bin
2026-04-27 18:47:25,114 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_214_19.bin
2026-04-27 18:47:25,115 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_215_15.bin
2026-04-27 18:47:25,117 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_215_26.bin
2026-04-27 18:47:25,118 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_216_09.bin
2026-04-27 18:47:25,120 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_216_20.bin
2026-04-27 18:47:25,121 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_217_27.bin
2026-04-27 18:47:25,122 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_217_57.bin
2026-04-27 18:47:25,124 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_218_42.bin
2026-04-27 18:47:25,125 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_218_47.bin
2026-04-27 18:47:25,127 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_219_15.bin
2026-04-27 18:47:25,128 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_219_22.bin
2026-04-27 18:47:25,129 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_219_23.bin
2026-04-27 18:47:25,131 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_221_16.bin
2026-04-27 18:47:25,132 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_221_32.bin
2026-04-27 18:47:25,134 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_222_05.bin
2026-04-27 18:47:25,135 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_222_12.bin
2026-04-27 18:47:25,136 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_223_29.bin
2026-04-27 18:47:25,138 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_223_50.bin
2026-04-27 18:47:25,139 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_224_18.bin
2026-04-27 18:47:25,141 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_224_28.bin
2026-04-27 18:47:25,142 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_225_08.bin
2026-04-27 18:47:25,143 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_225_41.bin
2026-04-27 18:47:25,145 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_226_03.bin
2026-04-27 18:47:25,146 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_226_07.bin
2026-04-27 18:47:25,147 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_226_19.bin
2026-04-27 18:47:25,149 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_227_19.bin
2026-04-27 18:47:25,150 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_227_48.bin
2026-04-27 18:47:25,151 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_228_36.bin
2026-04-27 18:47:25,153 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_228_46.bin
2026-04-27 18:47:25,154 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_228_49.bin
2026-04-27 18:47:25,156 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_229_39.bin
2026-04-27 18:47:25,158 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_229_55.bin
2026-04-27 18:47:25,159 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_231_19.bin
2026-04-27 18:47:25,162 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_231_28.bin
2026-04-27 18:47:25,164 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_231_44.bin
2026-04-27 18:47:25,166 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_232_17.bin
2026-04-27 18:47:25,169 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_232_20.bin
2026-04-27 18:47:25,170 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_233_17.bin
2026-04-27 18:47:25,172 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_233_26.bin
2026-04-27 18:47:25,174 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_234_17.bin
2026-04-27 18:47:25,176 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_234_25.bin
2026-04-27 18:47:25,179 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_234_48.bin
2026-04-27 18:47:25,182 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_235_34.bin
2026-04-27 18:47:25,193 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_235_39.bin
2026-04-27 18:47:25,205 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_236_37.bin
2026-04-27 18:47:25,215 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_236_43.bin
2026-04-27 18:47:25,217 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_237_48.bin
2026-04-27 18:47:25,218 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_237_53.bin
2026-04-27 18:47:25,220 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_238_00.bin
2026-04-27 18:47:25,223 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_238_17.bin
2026-04-27 18:47:25,226 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_238_32.bin
2026-04-27 18:47:25,228 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_239_04.bin
2026-04-27 18:47:25,229 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_239_30.bin
2026-04-27 18:47:25,232 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_239_55.bin
2026-04-27 18:47:25,234 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_241_39.bin
2026-04-27 18:47:25,236 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_241_54.bin
2026-04-27 18:47:25,238 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_242_11.bin
2026-04-27 18:47:25,240 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_242_57.bin
2026-04-27 18:47:25,242 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_243_00.bin
2026-04-27 18:47:25,245 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_243_31.bin
2026-04-27 18:47:25,247 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_244_00.bin
2026-04-27 18:47:25,249 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_244_14.bin
2026-04-27 18:47:25,251 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_245_07.bin
2026-04-27 18:47:25,253 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_245_14.bin
2026-04-27 18:47:25,256 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_246_10.bin
2026-04-27 18:47:25,259 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_246_49.bin
2026-04-27 18:47:25,260 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_246_57.bin
2026-04-27 18:47:25,263 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_247_06.bin
2026-04-27 18:47:25,266 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_247_32.bin
2026-04-27 18:47:25,269 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_248_25.bin
2026-04-27 18:47:25,270 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_248_48.bin
2026-04-27 18:47:25,273 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_249_04.bin
2026-04-27 18:47:25,275 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_249_14.bin
2026-04-27 18:47:25,278 file: loader.py func: loader line No: 287 Read raw file: /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32/scene_249_57.bin
2026-04-27 18:47:25,279 file: tool_utils.py func: tool_utils line No: 368 num of calibration data: 300
2026-04-27 18:47:25,280 file: tool_utils.py func: tool_utils line No: 369 calibration data shape: (1, 5, 360, 640)
, 'per_channel': True, 'max_percentile': 0.9999}, 'hbdk_dict': {'hbdk_pass_through_params': '--O3 --core-num 1 --fast ', 'input-source': {'input': 'ddr', '_default_value': 'ddr'}}, 'node_dict': {'Resize_220': {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}}, 'check_mode': False}
2026-04-27 18:47:25,452 file: model_builder.py func: model_builder line No: 35 Start to Horizon NN Model Convert.
2026-04-27 18:47:25,470 file: model_debugger.py func: model_debugger line No: 67 Loading horizon_nn debug methods:set()
2026-04-27 18:47:25,470 file: quantization_config.py func: quantization_config line No: 305 The activation calibration parameters:
    calibration_type:     max
    max_percentile:       0.9999
    per_channel:          True
The input of node Resize_220 are set to : int8
The output of node Resize_220 are set to : int8
There are 1 nodes designated to run on the bpu: ['Resize_220'].
2026-04-27 18:47:25,470 file: model_builder.py func: model_builder line No: 197 The specified model compilation architecture: bayes-e.
2026-04-27 18:47:25,470 file: model_builder.py func: model_builder line No: 207 The specified model compilation optimization parameters: [].
2026-04-27 18:47:25,470 file: model_builder.py func: model_builder line No: 35 Start to prepare the onnx model.
2026-04-27 18:47:25,528 file: prepare.py func: prepare line No: 106 Input ONNX Model Information:
ONNX IR version:          7
Opset version:            ['ai.onnx v11', 'horizon v1']
Producer:                 pytorch v1.10
Domain:                   None
Version:                  None
Graph input:
    input:                shape=[1, 5, 360, 640], dtype=FLOAT32
Graph output:
    ...entation:          shape=['Concatsegmentation_dim_0', 'Concatsegmentation_dim_1', 'Concatsegmentation_dim_2', 'Concatsegmentation_dim_3'], dtype=FLOAT32
    depth:                shape=['Sigmoiddepth_dim_0', 1, 'Sigmoiddepth_dim_2', 'Sigmoiddepth_dim_3'], dtype=FLOAT32
2026-04-27 18:47:25,610 file: model_builder.py func: model_builder line No: 38 End to prepare the onnx model.
2026-04-27 18:47:25,637 file: model_builder.py func: model_builder line No: 265 Saving model to: lraspp_performance_2.onnx_original_float_model.onnx.
2026-04-27 18:47:25,637 file: model_builder.py func: model_builder line No: 35 Start to optimize the onnx model.
2026-04-27 18:47:25,932 file: constant_folding.py func: constant_folding line No: 66 Summary info for constant_folding:
2026-04-27 18:47:25,932 file: constant_folding.py func: constant_folding line No: 67   After constant_folding, the number of nodes has changed from 340 to 262.
2026-04-27 18:47:25,932 file: constant_folding.py func: constant_folding line No: 71   After constant_folding, the number of parameters has changed from 3406950 to 3867735.
2026-04-27 18:47:25,932 file: constant_folding.py func: constant_folding line No: 76 Detailed info for constant_folding:
2026-04-27 18:47:25,932 file: constant_folding.py func: constant_folding line No: 88   After folding node (op_name: ConstantOfShape_263, op_type: ConstantOfShape), the number of increased parameters is 230395.
  After folding node (op_name: ConstantOfShape_277, op_type: ConstantOfShape), the number of increased parameters is 230395.
  After folding node (op_name: Concat_166, op_type: Concat), the number of increased parameters is 0.
  After folding node (op_name: Concat_188, op_type: Concat), the number of increased parameters is 0.
  After folding node (op_name: Concat_201, op_type: Concat), the number of increased parameters is 0.
  After folding node (op_name: Concat_217, op_type: Concat), the number of increased parameters is 0.
  After folding node (op_name: Concat_226, op_type: Concat), the number of increased parameters is 0.
  After folding node (op_name: Unsqueeze_259, op_type: Unsqueeze), the number of increased parameters is 0.
  After folding node (op_name: Unsqueeze_260, op_type: Unsqueeze), the number of increased parameters is 0.
  After folding node (op_name: Unsqueeze_261, op_type: Unsqueeze), the number of increased parameters is 0.
  After folding node (op_name: Concat_262, op_type: Concat), the number of increased parameters is 0.
  After folding node (op_name: Unsqueeze_273, op_type: Unsqueeze), the number of increased parameters is 0.
  After folding node (op_name: Unsqueeze_274, op_type: Unsqueeze), the number of increased parameters is 0.
  After folding node (op_name: Unsqueeze_275, op_type: Unsqueeze), the number of increased parameters is 0.
  After folding node (op_name: Concat_276, op_type: Concat), the number of increased parameters is 0.
  After folding node (op_name: Gather_252, op_type: Gather), the number of increased parameters is -4.
  After folding node (op_name: Gather_255, op_type: Gather), the number of increased parameters is -4.
  After folding node (op_name: Gather_258, op_type: Gather), the number of increased parameters is -4.
  After folding node (op_name: Gather_266, op_type: Gather), the number of increased parameters is -4.
  After folding node (op_name: Gather_269, op_type: Gather), the number of increased parameters is -4.
  After folding node (op_name: Gather_272, op_type: Gather), the number of increased parameters is -4.
  After folding node (op_name: Slice_165, op_type: Slice), the number of increased parameters is -5.
  After folding node (op_name: Slice_187, op_type: Slice), the number of increased parameters is -5.
  After folding node (op_name: Slice_200, op_type: Slice), the number of increased parameters is -5.
  After folding node (op_name: Slice_216, op_type: Slice), the number of increased parameters is -5.
  After folding node (op_name: Slice_225, op_type: Slice), the number of increased parameters is -5.
2026-04-27 18:47:26,146 file: model_builder.py func: model_builder line No: 38 End to optimize the onnx model.
2026-04-27 18:47:26,166 file: model_builder.py func: model_builder line No: 265 Saving model to: lraspp_performance_2.onnx_optimized_float_model.onnx.
2026-04-27 18:47:26,166 file: model_builder.py func: model_builder line No: 35 Start to calibrate the model.
2026-04-27 18:47:26,328 file: calibration_data_set.py func: calibration_data_set line No: 111 input name: input,  number_of_samples: 300
2026-04-27 18:47:26,328 file: calibration_data_set.py func: calibration_data_set line No: 123 There are 300 samples in the data set.
2026-04-27 18:47:26,328 file: infer_thresholds.py func: infer_thresholds line No: 84 Run calibration model with max-percentile:percentile=0.9999,per_channel method.
2026-04-27 18:47:26,459 file: base.py func: base line No: 138 Calibration using batch 8
2026-04-27 18:47:28,590 file: ort.py func: ort line No: 207 Reset batch_size=1 and execute forward again...
2026-04-27 18:49:29,939 file: base.py func: base line No: 138 Calibration using batch 8
2026-04-27 18:49:36,637 file: ort.py func: ort line No: 207 Reset batch_size=1 and execute forward again...
2026-04-27 18:55:08,666 file: modelwise_search.py func: modelwise_search line No: 62 Perchannel quantization is enabled.
2026-04-27 18:55:09,176 file: refine_threshold.py func: refine_threshold line No: 39 find illegal threshold in Add_244692_expand_dims_HzCalibration, it will be set default.
2026-04-27 18:55:09,298 file: model_builder.py func: model_builder line No: 38 End to calibrate the model.
2026-04-27 18:55:09,440 file: model_builder.py func: model_builder line No: 265 Saving model to: lraspp_performance_2.onnx_calibrated_model.onnx.
2026-04-27 18:55:09,440 file: model_builder.py func: model_builder line No: 35 Start to quantize the model.
2026-04-27 18:55:11,288 file: constant_folding.py func: constant_folding line No: 66 Summary info for constant_folding:
2026-04-27 18:55:11,288 file: constant_folding.py func: constant_folding line No: 67   After constant_folding, the number of nodes has changed from 220 to 216.
2026-04-27 18:55:11,288 file: constant_folding.py func: constant_folding line No: 71   After constant_folding, the number of parameters has changed from 4353946 to 4353946.
2026-04-27 18:55:11,288 file: constant_folding.py func: constant_folding line No: 76 Detailed info for constant_folding:
2026-04-27 18:55:11,288 file: constant_folding.py func: constant_folding line No: 88   After folding node (op_name: Mul_285_NCHW2NHWC_LayoutConvert_Input0, op_type: Transpose), the number of increased parameters is 0.
  After folding node (op_name: Mul_289_NCHW2NHWC_LayoutConvert_Input0, op_type: Transpose), the number of increased parameters is 0.
  After folding node (op_name: Mul_332_NCHW2NHWC_LayoutConvert_Input0, op_type: Transpose), the number of increased parameters is 0.
  After folding node (op_name: Mul_336_NCHW2NHWC_LayoutConvert_Input0, op_type: Transpose), the number of increased parameters is 0.
2026-04-27 18:55:11,392 file: model_builder.py func: model_builder line No: 38 End to quantize the model.
2026-04-27 18:55:11,497 file: model_builder.py func: model_builder line No: 265 Saving model to: lraspp_performance_2.onnx_quantized_model.onnx.
2026-04-27 18:55:11,498 file: model_builder.py func: model_builder line No: 35 Start to compile the model with march bayes-e.
2026-04-27 18:55:12,110 file: hybrid_build.py func: hybrid_build line No: 111 Compile submodel: torch-jit-export_subgraph_0
2026-04-27 18:55:12,130 file: hbdk_cc.py func: hbdk_cc line No: 126 hbdk-cc parameters:['--O3', '--core-num', '1', '--fast', '--input-layout', 'NHWC', '--output-layout', 'NHWC', '--input-source', 'ddr']
2026-04-27 18:55:12,130 file: hbdk_cc.py func: hbdk_cc line No: 127 hbdk-cc command used:hbdk-cc -f hbir -m /tmp/tmprse_13db/torch-jit-export_subgraph_0.hbir -o /tmp/tmprse_13db/torch-jit-export_subgraph_0.hbm --march bayes-e --progressbar --O3 --core-num 1 --fast --input-layout NHWC --output-layout NHWC --input-source ddr
2026-04-27 19:14:22,182 file: tool_utils.py func: tool_utils line No: 326 consumed time 1150.03
2026-04-27 19:14:22,312 file: tool_utils.py func: tool_utils line No: 326 FPS=19.2, latency = 52071.6 us, DDR = 161155200 bytes   (see torch-jit-export_subgraph_0.html)
2026-04-27 19:14:22,333 file: hybrid_build.py func: hybrid_build line No: 111 Compile submodel: torch-jit-export_subgraph_1
2026-04-27 19:14:22,339 file: hbdk_cc.py func: hbdk_cc line No: 126 hbdk-cc parameters:['--O3', '--core-num', '1', '--fast', '--input-layout', 'NHWC', '--output-layout', 'NHWC', '--input-source', 'ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr']
2026-04-27 19:14:22,340 file: hbdk_cc.py func: hbdk_cc line No: 127 hbdk-cc command used:hbdk-cc -f hbir -m /tmp/tmprse_13db/torch-jit-export_subgraph_1.hbir -o /tmp/tmprse_13db/torch-jit-export_subgraph_1.hbm --march bayes-e --progressbar --O3 --core-num 1 --fast --input-layout NHWC --output-layout NHWC --input-source ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr
2026-04-27 19:38:15,633 file: tool_utils.py func: tool_utils line No: 326 consumed time 1433.28
2026-04-27 19:38:15,735 file: tool_utils.py func: tool_utils line No: 326 FPS=58.92, latency = 16971.5 us, DDR = 121434880 bytes   (see torch-jit-export_subgraph_1.html)
2026-04-27 19:38:15,941 file: model_builder.py func: model_builder line No: 38 End to compile the model with march bayes-e.
2026-04-27 19:38:20,599 file: print_info_dict.py func: print_info_dict line No: 72 The main quantized node information:
=================================================================================================================
Node                       ON   Subgraph  Type                           Cosine Similarity  Threshold  DataType  
-----------------------------------------------------------------------------------------------------------------
Conv_0                     BPU  id(0)     HzSQuantizedConv               0.998274           2.60514    int8      
Mul_2                      BPU  id(0)     HzLut                          0.997032           10.0452    int8      
Conv_3                     BPU  id(0)     HzSQuantizedConv               0.994523           8.42675    int8      
Conv_5                     BPU  id(0)     HzSQuantizedConv               0.993148           10.6772    int8      
Conv_7                     BPU  id(0)     HzSQuantizedConv               0.993031           13.4435    int8      
Conv_9                     BPU  id(0)     HzSQuantizedConv               0.988840           5.09944    int8      
Conv_11                    BPU  id(0)     HzSQuantizedConv               0.989113           7.39392    int8      
Conv_12                    BPU  id(0)     HzSQuantizedConv               0.986031           5.9679     int8      
Conv_14                    BPU  id(0)     HzSQuantizedConv               0.975172           5.84761    int8      
Conv_16                    BPU  id(0)     HzSQuantizedConv               0.968805           6.95885    int8      
Conv_18                    BPU  id(0)     HzSQuantizedConv               0.983533           9.80303    int8      
Conv_20                    BPU  id(0)     HzSQuantizedConv               0.983527           2.53918    int8      
GlobalAveragePool_22       BPU  id(0)     HzSQuantizedGlobalAveragePool  0.998626           6.9441     int8      
Conv_23                    BPU  id(0)     HzSQuantizedConv               0.998960           1.71872    int8      
Conv_25                    BPU  id(0)     HzSQuantizedConv               0.999481           3.41232    int8      
HardSigmoid_26             BPU  id(0)     HzLut                          0.999949           3.0        int8      
Mul_27                     BPU  id(0)     HzSElementwiseMul              0.985715           6.9441     int8      
Conv_28                    BPU  id(0)     HzSQuantizedConv               0.981658           3.87795    int8      
Conv_29                    BPU  id(0)     HzSQuantizedConv               0.981737           8.65345    int8      
Conv_31                    BPU  id(0)     HzSQuantizedConv               0.978815           10.5245    int8      
GlobalAveragePool_33       BPU  id(0)     HzSQuantizedGlobalAveragePool  0.998277           6.56596    int8      
Conv_34                    BPU  id(0)     HzSQuantizedConv               0.997984           1.42848    int8      
Conv_36                    BPU  id(0)     HzSQuantizedConv               0.997904           2.88824    int8      
HardSigmoid_37             BPU  id(0)     HzLut                          0.999931           3.0        int8      
Mul_38                     BPU  id(0)     HzSElementwiseMul              0.978058           6.56596    int8      
Conv_39                    BPU  id(0)     HzSQuantizedConv               0.974428           3.45725    int8      
Conv_41                    BPU  id(0)     HzSQuantizedConv               0.978951           10.1026    int8      
Conv_43                    BPU  id(0)     HzSQuantizedConv               0.973985           5.65957    int8      
GlobalAveragePool_45       BPU  id(0)     HzSQuantizedGlobalAveragePool  0.998186           6.14948    int8      
Conv_46                    BPU  id(0)     HzSQuantizedConv               0.999109           1.35851    int8      
Conv_48                    BPU  id(0)     HzSQuantizedConv               0.998690           2.52155    int8      
HardSigmoid_49             BPU  id(0)     HzLut                          0.999948           3.0        int8      
Mul_50                     BPU  id(0)     HzSElementwiseMul              0.974355           6.14948    int8      
Conv_51                    BPU  id(0)     HzSQuantizedConv               0.967800           3.38589    int8      
Conv_53                    BPU  id(0)     HzSQuantizedConv               0.978041           11.5251    int8      
Mul_55                     BPU  id(0)     HzLut                          0.977002           6.07911    int8      
Conv_56                    BPU  id(0)     HzSQuantizedConv               0.975701           5.44384    int8      
Mul_58                     BPU  id(0)     HzLut                          0.973730           6.6599     int8      
Conv_59                    BPU  id(0)     HzSQuantizedConv               0.975475           6.12633    int8      
Conv_60                    BPU  id(0)     HzSQuantizedConv               0.975861           5.89489    int8      
Mul_62                     BPU  id(0)     HzLut                          0.974721           5.89231    int8      
Conv_63                    BPU  id(0)     HzSQuantizedConv               0.975227           5.3536     int8      
Mul_65                     BPU  id(0)     HzLut                          0.974035           7.1766     int8      
Conv_66                    BPU  id(0)     HzSQuantizedConv               0.975938           6.41371    int8      
Conv_68                    BPU  id(0)     HzSQuantizedConv               0.976914           9.51773    int8      
Mul_70                     BPU  id(0)     HzLut                          0.974127           6.09129    int8      
Conv_71                    BPU  id(0)     HzSQuantizedConv               0.973447           5.47737    int8      
Mul_73                     BPU  id(0)     HzLut                          0.970959           7.56809    int8      
Conv_74                    BPU  id(0)     HzSQuantizedConv               0.970163           6.44408    int8      
Conv_76                    BPU  id(0)     HzSQuantizedConv               0.976085           12.444     int8      
Mul_78                     BPU  id(0)     HzLut                          0.973157           6.06677    int8      
Conv_79                    BPU  id(0)     HzSQuantizedConv               0.971513           5.45808    int8      
Mul_81                     BPU  id(0)     HzLut                          0.969360           7.3182     int8      
Conv_82                    BPU  id(0)     HzSQuantizedConv               0.969047           6.8235     int8      
Conv_84                    BPU  id(0)     HzSQuantizedConv               0.976990           14.6883    int8      
Mul_86                     BPU  id(0)     HzLut                          0.974814           6.45453    int8      
Conv_87                    BPU  id(0)     HzSQuantizedConv               0.979091           5.81083    int8      
Mul_89                     BPU  id(0)     HzLut                          0.977301           6.98648    int8      
GlobalAveragePool_90       BPU  id(0)     HzSQuantizedGlobalAveragePool  0.987749           5.99138    int8      
Conv_91                    BPU  id(0)     HzSQuantizedConv               0.991015           0.768847   int8      
Conv_93                    BPU  id(0)     HzSQuantizedConv               0.985799           1.31595    int8      
HardSigmoid_94             BPU  id(0)     HzLut                          0.999951           3.0        int8      
Mul_95                     BPU  id(0)     HzSElementwiseMul              0.977005           5.99138    int8      
Conv_96                    BPU  id(0)     HzSQuantizedConv               0.981795           3.01097    int8      
Conv_97                    BPU  id(0)     HzSQuantizedConv               0.980276           6.07557    int8      
Mul_99                     BPU  id(0)     HzLut                          0.977407           6.05147    int8      
Conv_100                   BPU  id(0)     HzSQuantizedConv               0.978366           5.29462    int8      
Mul_102                    BPU  id(0)     HzLut                          0.975578           6.41162    int8      
GlobalAveragePool_103      BPU  id(0)     HzSQuantizedGlobalAveragePool  0.986407           5.50975    int8      
Conv_104                   BPU  id(0)     HzSQuantizedConv               0.988208           0.722488   int8      
Conv_106                   BPU  id(0)     HzSQuantizedConv               0.988799           1.25979    int8      
HardSigmoid_107            BPU  id(0)     HzLut                          0.999960           3.0        int8      
Mul_108                    BPU  id(0)     HzSElementwiseMul              0.975142           5.50975    int8      
Conv_109                   BPU  id(0)     HzSQuantizedConv               0.977491           2.76261    int8      
Conv_111                   BPU  id(0)     HzSQuantizedConv               0.982895           9.20412    int8      
Mul_113                    BPU  id(0)     HzLut                          0.980993           5.94812    int8      
Conv_114                   BPU  id(0)     HzSQuantizedConv               0.984251           5.34873    int8      
Mul_116                    BPU  id(0)     HzLut                          0.983348           5.242      int8      
GlobalAveragePool_117      BPU  id(0)     HzSQuantizedGlobalAveragePool  0.991532           4.54412    int8      
Conv_118                   BPU  id(0)     HzSQuantizedConv               0.990856           0.69875    int8      
Conv_120                   BPU  id(0)     HzSQuantizedConv               0.990515           1.42778    int8      
HardSigmoid_121            BPU  id(0)     HzLut                          0.999945           3.0        int8      
Mul_122                    BPU  id(0)     HzSElementwiseMul              0.983214           4.54412    int8      
Conv_123                   BPU  id(0)     HzSQuantizedConv               0.986662           2.30601    int8      
Conv_124                   BPU  id(0)     HzSQuantizedConv               0.984734           3.9707     int8      
Mul_126                    BPU  id(0)     HzLut                          0.981986           4.26747    int8      
Conv_127                   BPU  id(0)     HzSQuantizedConv               0.983515           3.91532    int8      
Mul_129                    BPU  id(0)     HzLut                          0.981507           4.62324    int8      
GlobalAveragePool_130      BPU  id(0)     HzSQuantizedGlobalAveragePool  0.988593           4.02867    int8      
Conv_131                   BPU  id(0)     HzSQuantizedConv               0.987710           0.859272   int8      
Conv_133                   BPU  id(0)     HzSQuantizedConv               0.987758           1.54649    int8      
HardSigmoid_134            BPU  id(0)     HzLut                          0.999947           3.0        int8      
Mul_135                    BPU  id(0)     HzSElementwiseMul              0.981334           4.02867    int8      
Conv_136                   BPU  id(0)     HzSQuantizedConv               0.978587           2.0281     int8      
Conv_138                   BPU  id(0)     HzSQuantizedConv               0.980687           5.58802    int8      
Mul_140                    BPU  id(0)     HzLut                          0.978443           4.09105    int8      
Conv_141                   BPU  id(0)     HzSQuantizedConv               0.979151           3.84357    int8      
Mul_143                    BPU  id(0)     HzLut                          0.976952           4.72488    int8      
GlobalAveragePool_144      BPU  id(0)     HzSQuantizedGlobalAveragePool  0.984766           4.17765    int8      
Conv_145                   BPU  id(0)     HzSQuantizedConv               0.986574           0.940707   int8      
Conv_147                   BPU  id(0)     HzSQuantizedConv               0.983309           1.48093    int8      
HardSigmoid_148            BPU  id(0)     HzLut                          0.999934           3.0        int8      
Mul_149                    BPU  id(0)     HzSElementwiseMul              0.976776           4.17765    int8      
Conv_150                   BPU  id(0)     HzSQuantizedConv               0.968574           2.14135    int8      
Conv_152                   BPU  id(0)     HzSQuantizedConv               0.978111           7.33566    int8      
Mul_154                    BPU  id(0)     HzLut                          0.975831           4.23688    int8      
Conv_155                   BPU  id(0)     HzSQuantizedConv               0.978955           3.96892    int8      
GlobalAveragePool_157      BPU  id(0)     HzSQuantizedGlobalAveragePool  0.990314           3.96892    int8      
Conv_158                   BPU  id(0)     HzSQuantizedConv               0.999803           1.1018     int8      
Sigmoid_159                BPU  id(0)     HzLut                          0.999980           5.49897    int8      
Mul_160                    BPU  id(0)     HzSElementwiseMul              0.983713           6.08526    int8      
Resize_169                 BPU  id(0)     HzQuantizedRoiResize           0.987819           4.2731     int8      
Conv_170                   BPU  id(0)     HzSQuantizedConv               0.987237           5.9679     int8      
Conv_171                   BPU  id(0)     HzSQuantizedConv               0.994769           4.2731     int8      
Conv_173                   BPU  id(0)     HzSQuantizedConv               0.991016           5.9679     int8      
Conv_174                   BPU  id(0)     HzSQuantizedConv               0.994570           4.2731     int8      
Conv_176                   BPU  id(0)     HzSQuantizedConv               0.991359           5.9679     int8      
598_calibrated_Requantize  BPU  id(0)     HzRequantize                   --                 --         int8      
Concat_178                 BPU  id(0)     Concat                         0.988798           4.7036     int8      
Conv_179                   BPU  id(0)     HzSQuantizedConv               0.992879           4.7036     int8      
Conv_181                   BPU  id(0)     HzSQuantizedConv               0.992543           3.59188    int8      
Resize_191                 BPU  id(0)     HzQuantizedRoiResize           0.993917           3.70851    int8      
Conv_192                   BPU  id(0)     HzSQuantizedConv               0.995477           3.70851    int8      
Conv_194                   BPU  id(0)     HzSQuantizedConv               0.993877           3.43902    int8      
Resize_204                 BPU  id(0)     HzQuantizedRoiResize           0.994091           3.52803    int8      
Conv_205                   BPU  id(0)     HzSQuantizedConv               0.995181           3.52803    int8      
Conv_207                   BPU  id(0)     HzSQuantizedConv               0.992899           2.87636    int8      
Conv_209                   BPU  id(0)     HzSQuantizedConv               0.994420           3.27891    int8      
Conv_211                   BPU  id(0)     HzSQuantizedConv               0.991340           3.93734    int8      
Resize_220                 BPU  id(0)     HzQuantizedRoiResize           0.994853           18.308     int8      
Resize_229                 BPU  id(0)     HzQuantizedRoiResize           0.993870           20.0555    int8      
Sigmoid_230                BPU  id(0)     HzLut2Layer                    0.999839           11.0903    int16     
Slice_235                  BPU  id(0)     Slice                          0.995382           --         int8      
ReduceMax_236              BPU  id(0)     HzQuantizedReduceMax           0.994704           18.308     int8      
Slice_241                  BPU  id(0)     Slice                          0.994350           --         int8      
ReduceMax_242              BPU  id(0)     HzQuantizedReduceMax           0.996857           20.0555    int8      
Add_244                    BPU  id(0)     HzSElementwiseAdd              0.996858           20.0555    int8      
Less_245                   CPU  --        Less                           0.987903           23.7686    float     
Not_246                    CPU  --        Not                            0.995166           --         bool      
Cast_247                   CPU  --        Cast                           0.995166           --         bool      
Sub_249                    BPU  id(1)     HzSElementwiseSub              0.987903           1.0        int8      
Slice_278                  BPU  id(0)     Slice                          0.994026           --         int8      
Slice_279                  BPU  id(0)     Slice                          0.993472           --         int8      
Mul_280                    BPU  id(1)     HzSElementwiseMul              0.946916           20.0555    int8      
Mul_281                    BPU  id(1)     HzSElementwiseMul              0.994196           18.308     int8      
Add_282                    BPU  id(1)     HzSElementwiseAdd              0.964029           18.3494    int8      
Slice_283                  BPU  id(0)     Slice                          0.947599           --         int8      
Mul_284                    BPU  id(1)     HzSElementwiseMul              0.944098           20.0555    int8      
Mul_285                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_286                    BPU  id(1)     HzSElementwiseAdd              0.987902           19.5685    int8      
Slice_287                  BPU  id(0)     Slice                          0.992352           --         int8      
Mul_288                    BPU  id(1)     HzSElementwiseMul              0.981487           20.0555    int8      
Mul_289                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_290                    BPU  id(1)     HzSElementwiseAdd              0.987904           14.4241    int8      
Slice_291                  BPU  id(0)     Slice                          0.994469           --         int8      
Mul_292                    BPU  id(1)     HzSElementwiseMul              0.993172           20.0555    int8      
Mul_293                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_294                    BPU  id(1)     HzSElementwiseAdd              0.987899           18.2293    int8      
Slice_295                  BPU  id(0)     Slice                          0.985918           --         int8      
Mul_296                    BPU  id(1)     HzSElementwiseMul              0.971940           20.0555    int8      
Mul_297                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_298                    BPU  id(1)     HzSElementwiseAdd              0.987905           14.3985    int8      
Slice_299                  BPU  id(0)     Slice                          0.996317           --         int8      
Mul_300                    BPU  id(1)     HzSElementwiseMul              0.996090           20.0555    int8      
Mul_301                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_302                    BPU  id(1)     HzSElementwiseAdd              0.987902           23.7486    int8      
Slice_303                  BPU  id(0)     Slice                          0.994563           --         int8      
Mul_304                    BPU  id(1)     HzSElementwiseMul              0.993596           20.0555    int8      
Mul_305                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_306                    BPU  id(1)     HzSElementwiseAdd              0.987898           18.9094    int8      
Slice_307                  BPU  id(0)     Slice                          0.994544           --         int8      
Mul_308                    BPU  id(1)     HzSElementwiseMul              0.989165           20.0555    int8      
Mul_309                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_310                    BPU  id(1)     HzSElementwiseAdd              0.987907           12.6793    int8      
Slice_311                  BPU  id(0)     Slice                          0.998598           --         int8      
Mul_312                    BPU  id(1)     HzSElementwiseMul              0.994430           20.0555    int8      
Mul_313                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_314                    BPU  id(1)     HzSElementwiseAdd              0.987908           5.8745     int8      
Slice_315                  BPU  id(0)     Slice                          0.998134           --         int8      
Mul_316                    BPU  id(1)     HzSElementwiseMul              0.995507           20.0555    int8      
Mul_317                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_318                    BPU  id(1)     HzSElementwiseAdd              0.987907           6.4862     int8      
Slice_319                  BPU  id(0)     Slice                          0.998186           --         int8      
Mul_320                    BPU  id(1)     HzSElementwiseMul              0.994683           20.0555    int8      
Mul_321                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_322                    BPU  id(1)     HzSElementwiseAdd              0.987907           5.42638    int8      
Slice_323                  BPU  id(0)     Slice                          0.998246           --         int8      
Mul_324                    BPU  id(1)     HzSElementwiseMul              0.994560           20.0555    int8      
Mul_325                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_326                    BPU  id(1)     HzSElementwiseAdd              0.987906           5.33055    int8      
Slice_327                  BPU  id(0)     Slice                          0.998331           --         int8      
Mul_328                    BPU  id(1)     HzSElementwiseMul              0.994546           20.0555    int8      
Mul_329                    BPU  id(1)     HzSElementwiseMul              0.987903           10000.0    int8      
Add_330                    BPU  id(1)     HzSElementwiseAdd              0.987907           5.51782    int8      
Slice_331                  BPU  id(0)     Slice                          0.997066           --         int8      
Mul_332                    BPU  id(1)     HzSElementwiseMul              0.995166           10000.0    int8      
Mul_333                    BPU  id(1)     HzSElementwiseMul              0.995671           18.308     int8      
Add_334                    BPU  id(1)     HzSElementwiseAdd              0.995167           10000.0    int8      
Slice_335                  BPU  id(0)     Slice                          0.992213           --         int8      
Mul_336                    BPU  id(1)     HzSElementwiseMul              0.995166           10000.0    int8      
Mul_337                    BPU  id(1)     HzSElementwiseMul              0.989259           18.308     int8      
Add_338                    BPU  id(1)     HzSElementwiseAdd              0.995165           10000.0    int8      
751_calibrated_Requantize  BPU  id(1)     HzRequantize                   --                 --         int16     
Concat_339                 BPU  id(1)     Concat                         0.990034           18.3494    int16
2026-04-27 19:38:20,599 file: print_info_dict.py func: print_info_dict line No: 72 The quantized model output:
===============================================================================
Output        Cosine Similarity  L1 Distance  L2 Distance  Chebyshev Distance  
-------------------------------------------------------------------------------
segmentation  0.990034           64.714255    0.432090     10008.585938        
depth         0.999839           0.002332     0.000027     0.561325
2026-04-27 19:38:20,604 file: model_builder.py func: model_builder line No: 38 End to Horizon NN Model Convert.
2026-04-27 19:38:20,613 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 601 start convert to *.bin file....
2026-04-27 19:38:20,631 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4326 ONNX model output num : 2
2026-04-27 19:38:20,633 file: layout_util.py func: layout_util line No: 15 set_featuremap_layout start
2026-04-27 19:38:20,633 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4060 model_deps_info: {'hb_mapper_version': '1.24.3', 'hbdk_version': '3.49.15', 'hbdk_runtime_version': ' 3.15.55.0', 'horizon_nn_version': '1.1.0', 'onnx_model': '/open_explorer/test_lraspp_model_0427/lraspp_performance_2.onnx', 'march': 'bayes-e', 'layer_out_dump': False, 'log_level': 'DEBUG', 'working_dir': '/open_explorer/test_lraspp_model_0427/hb_mapper_output_2/lraspp_performance', 'model_prefix': 'lraspp_performance_2.onnx', 'node_info': {'Resize_220': {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}}, 'input_names': ['input'], 'input_type_rt': ['featuremap'], 'input_space_and_range': ['regular'], 'input_type_train': ['featuremap'], 'input_layout_rt': ['NCHW'], 'input_layout_train': ['NCHW'], 'norm_type': ['no_preprocess'], 'scale_value': [''], 'mean_value': [''], 'input_shape': ['1x5x360x640'], 'input_batch': ['1'], 'cal_dir': ['/open_explorer/test_lraspp_model_0427/hb_calibration_data_f32'], 'cal_data_type': ['float32'], 'preprocess_on': False, 'calibration_type': 'max', 'per_channel': 'True', 'max_percentile': 0.9999, 'hbdk_params': {'hbdk_pass_through_params': '--O3 --core-num 1 --fast ', 'input-source': {'input': 'ddr', '_default_value': 'ddr'}}, 'debug': False, 'compile_mode': 'latency'}
2026-04-27 19:38:20,633 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4183 ############# model deps info #############
2026-04-27 19:38:20,633 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4184 hb_mapper version   : 1.24.3
2026-04-27 19:38:20,633 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4187 hbdk version        : 3.49.15
2026-04-27 19:38:20,633 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4189 hbdk runtime version: 3.15.55.0
2026-04-27 19:38:20,633 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4192 horizon_nn version  : 1.1.0
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4196 ############# model_parameters info #############
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4202 onnx_model          : /open_explorer/test_lraspp_model_0427/lraspp_performance_2.onnx
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4203 BPU march           : bayes-e
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4204 layer_out_dump      : False
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4205 log_level           : DEBUG
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4206 working dir         : /open_explorer/test_lraspp_model_0427/hb_mapper_output_2/lraspp_performance
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4207 output_model_file_prefix: lraspp_performance_2.onnx
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4226 node info  : {'Resize_220': {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}}
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4228 ############# input_parameters info #############
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4246 ------------------------------------------
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4248 ---------input info : input ---------
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4249 input_name          : input
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4250 input_type_rt       : featuremap
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4252 input_space&range   : regular
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4254 input_layout_rt     : NCHW
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4255 input_type_train    : featuremap
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4256 input_layout_train  : NCHW
2026-04-27 19:38:20,634 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4257 norm_type           : no_preprocess
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4258 input_shape         : 1x5x360x640
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4260 input_batch         : 1
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4268 cal_data_dir        : /open_explorer/test_lraspp_model_0427/hb_calibration_data_f32
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4270 cal_data_type       : float32
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4271 ---------input info : input end -------
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4272 ------------------------------------------
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4274 ############# calibration_parameters info #############
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4275 preprocess_on       : False
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4276 calibration_type:   : max
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4281 max_percentile      : 0.9999
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4284 per_channel         : True
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4303 ############# compiler_parameters info #############
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4305 debug               : False
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4307 compile_mode        : latency
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4310 hbdk_pass_through_params: --O3 --core-num 1 --fast
2026-04-27 19:38:20,635 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4310 input-source        : {'input': 'ddr', '_default_value': 'ddr'}
2026-04-27 19:38:20,640 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 783 Convert to runtime bin file successfully!
2026-04-27 19:38:20,640 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 784 End Model Convert

下一步怎么排查?

我来帮你分析这个问题。从日志看,模型转换成功,但编译预估性能实测性能差距巨大,需要系统排查。

关键信息提取

从编译日志看:

  • Subgraph 0: FPS=19.2, latency=52ms
  • Subgraph 1: FPS=58.9, latency=17ms
  • 编译预估整体应 >15FPS,但实测 <2Hz

排查步骤

:one: 确认实测方式

先确认你是怎么测的性能:

# 是用 hb_rt_perf 还是自己写的推理代码?
# 请提供实测代码片段或命令

:two: 检查子图切换开销

模型被切分成 2个子图,CPU/BPU交替执行会有额外开销:

CPU算子: Less_245, Not_246, Cast_247 (在subgraph_0和subgraph_1之间)

排查命令

# 查看生成的.hbm文件结构
ls -lh hb_mapper_output_2/lraspp_performance/

# 检查是否有多个.hbm文件(对应多个subgraph)

:three: 检查输入数据喂入方式

你的配置是 input_source: ddr,5通道featuremap从DDR喂入:

input_shape: '1x5x360x640'  # 5通道,数据量较大
input_source: {'input': 'ddr'}

可能瓶颈

  • 每次推理前是否都从主机拷贝数据到DDR?
  • 是否用了零拷贝/共享内存?

:four: 检查后处理开销

模型输出2个tensor:

segmentation: [?, ?, ?, ?]
depth: [?, 1, ?, ?]

问题:后处理(argmax、可视化等)是否在计时范围内?


建议的排查路径

根据您提供的模型转换日志和社区类似案例,我来帮您分析性能瓶颈并给出排查路径。

核心问题分析

从您的日志中发现了几个关键性能瓶颈点

1. 模型被切分为2个BPU子图

Compile submodel: torch-jit-export_subgraph_0  (FPS=19.2, latency=52ms)
Compile submodel: torch-jit-export_subgraph_1  (FPS=58.9, latency=17ms)

虽然所有算子都在BPU上,但模型被切分成2个子图,说明存在子图间的数据传输开销。查看量化节点表发现:

  • Less_245, Not_246, Cast_247 这三个算子在CPU运行
  • 这导致模型被切分为 id(0)id(1) 两个子图

2. 大量 CPU 算子导致频繁数据搬运

从量化表看到:

Less_245   CPU  --  Less
Not_246    CPU  --  Not
Cast_247   CPU  --  Cast

这些算子虽然简单,但会导致:

  • BPU → CPU 数据传输
  • CPU 计算
  • CPU → BPU 数据传输
  • 每次推理都要经历多次 DDR 数据搬运

3. 输入是 5 通道 featuremap,数据量大

input_shape: '1x5x360x640'
DDR = 161155200 bytes (subgraph_0)
DDR = 121434880 bytes (subgraph_1)

5 通道输入导致:

  • 单次推理 DDR 读写超过 280MB
  • DDR 带宽成为瓶颈(X5 DDR 带宽约 17GB/s)
  • 仅数据传输就耗时约 16ms

排查步骤

步骤 1:验证子图切分问题

在板端执行:

hrt_model_exec model_info --model_file lraspp_performance_2.bin | grep -A 5 "subgraph"

期望结果:应该只有 1 个 BPU subgraph
实际情况:您有 2 个 subgraph,说明模型被切分

步骤 2:检查 CPU 算子

在转换前用 checker 诊断:

# 在 Docker 环境内
hb_mapper checker --model-type onnx --march bayes-e \
  --model lraspp_performance_2.onnx 2>&1 | grep "CPU"

重点查看哪些算子被分配到 CPU,常见原因:

  • 不支持的算子类型(如 Less, Not, Cast
  • 数据类型不支持(如 bool, int64)
  • 量化参数配置问题

步骤 3:性能 Profiling

在板端运行性能分析:

# 开启性能 profiling
export HB_PROFILER=1
hrt_model_exec perf --model_file lraspp_performance_2.bin --loop_num 100

# 查看 profiler 日志
cat profiler_*.log | grep -E "CPU|BPU|DDR"

优化方案

方案 1:将 CPU 算子融合到 BPU(推荐)

修改量化配置,强制将 Less, Not, Cast 等算子放到 BPU:

model_parameters:
  # ... 其他配置不变
  
  # 添加 node_info 强制指定算子到 BPU
  node_info: {
    "Resize_220": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Less_245": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Not_246": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Cast_247": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}
  }

:warning: 注意:需要验证这些算子在 BPU 上是否支持,如果不支持需要修改模型结构。

方案 2:优化模型结构(根本解决)

在 ONNX 导出前修改模型,避免产生 CPU 算子:

# 在 PyTorch 模型中,将 Less/Not/Cast 替换为 BPU 友好的算子
# 例如:将 bool  mask 计算改为 int8 的 multiply

# 原始代码(会产生 CPU 算子)
mask = depth < threshold  # Less 算子
mask = ~mask              # Not 算子
mask = mask.float()       # Cast 算子

# 优化后(BPU 友好)
mask = (depth < threshold).to(torch.int8)  # 直接输出 int8
result = depth * mask  # 用 Mul 代替条件判断

方案 3:减少 DDR 数据搬运

针对 5 通道 featuremap 输入,优化编译器参数:

compiler_parameters:
  compile_mode: 'latency'
  core_num: 1
  optimize_level: 'O3'
  # 尝试修改 input_source
  input_source: {'input': 'pmem'}  # 如果数据在 pmem 中

或者在应用层优化:

  • 使用 零拷贝 技术传递数据
  • 将预处理和模型推理放在同一内存区域

方案 4:启用 CPU 性能模式

确保 CPU 运行在最高性能状态:

# 在 RDK X5 上执行
sudo bash -c "echo 1 > /sys/devices/system/cpu/cpufreq/boost"
sudo bash -c "echo performance > /sys/devices/system/cpu/cpufreq/policy0/scaling_governor"

# 验证
cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq  # 应该显示 1800000

快速验证脚本

创建一个简单的性能测试脚本:

#!/usr/bin/env python3
from hobot_dnn import pyeasy_dnn as dnn
import time
import numpy as np

# 加载模型
models = dnn.load('lraspp_performance_2.bin')
model = models[0]

# 创建测试数据 (5 通道 featuremap)
test_input = np.random.rand(5, 360, 640).astype(np.float32)

# 预热
for _ in range(20):
    _ = model.forward(test_input)

# 性能测试
times = []
for i in range(100):
    start = time.time()
    outputs = model.forward(test_input)
    elapsed = (time.time() - start) * 1000
    times.append(elapsed)
    if (i + 1) % 20 == 0:
        print(f"{i+1}/100 | 平均:{np.mean(times[-20:]):.2f}ms | FPS: {1000/np.mean(times[-20:]):.1f}")

print(f"\n总平均:{np.mean(times):.2f}ms | FPS: {1000/np.mean(times):.1f}")

建议的排查顺序

  1. 先确认子图数量hrt_model_exec model_info
  2. 查看 profiler 日志 → 确认 CPU 耗时占比
  3. 修改 node_info 配置 → 尝试将 CPU 算子强制放到 BPU
  4. 重新量化测试 → 验证性能是否提升
  5. 如果仍不行 → 修改 ONNX 模型结构,避免产生 CPU 算子

相关社区资源

建议您参考以下帖子获取更多优化经验:

  1. 地瓜派 RDK X5 部署 YOLOv11n 完整实战:从 7 FPS 到 47 FPS 的性能优化之路 - 详细记录了 Softmax 算子导致性能问题的解决方案
  2. 求助:纯 BPU 算子推理很慢 (其中绝大多是是 CPU 耗时) - 类似问题的讨论
  3. 全 bpu 算子,但运行在 x3 上时,依旧高 cpu 占用低 bpu 占用 - BPU 利用率监控方法

下一步行动建议

请先执行步骤 1 和步骤 2,确认子图数量和 CPU 算子类型,然后根据您的反馈我可以给出更具体的优化方案。如果方便,也可以将 profiler.log 的关键部分贴出来,我可以帮您分析具体是哪个环节耗时最多。

板子测试hb_model_modifier的日志:

running_condition
thread_num,1
core_id,0
frame_count,1000
run_time,286469.02
model_name,"lraspp_performance_2.onnx"

perf_result
average_latency,286.4483642578125
FPS,3.4907788632781305

processor_latency, avg_time, max_time, min_time
BPU_inference_time_cost,104.1511,104.995,103.826,
CPU_inference_time_cost,181.881601,235.87900000000002,177.59,

task_latency, avg_time, max_time, min_time
TaskPendingTime,0.034493,0.201,0.025,
TaskRunningTime,286.363697,322.888,282.124,

model nodes, avg_time, max_time, min_time
BPU_torch-jit-export_subgraph_0,86.605218,86.995,86.338
BPU_torch-jit-export_subgraph_1,17.545882000000002,18.0,17.488
Cast_Cast_247,1.2418030000000002,1.585,1.234
Dequantize_685_HzDequantize,0.246742,0.359,0.243
Dequantize_693_calibrated_HzDequantize,0.434313,1.183,0.428
Dequantize_depth_HzDequantize,0.44155099999999997,1.095,0.435
Dequantize_segmentation_HzDequantize,6.7867939999999995,11.146,6.442
ElementWiseBinaryBroadcast_Less_245,0.396486,4.488,0.386
Eltwise_Not_246,6.500202,8.647,6.468
Preprocess,0.022912,0.057,0.016
Quantize_696_696_HzCalibration_HzQuantize,6.1855649999999995,6.487,6.162
Quantize_input_input_HzCalibration_HzQuantize,8.887985,12.113,8.653
torch-jit-export_subgraph_0_input_layout_convert,1.7739310000000001,2.274,1.748
torch-jit-export_subgraph_0_output_layout_convert,61.784411999999996,97.553,58.29
torch-jit-export_subgraph_1_input_layout_convert,86.80635400000001,88.408,86.732
torch-jit-export_subgraph_1_output_layout_convert,0.37255099999999997,0.484,0.353

参考的这个性能调优文档:8.1. 模型性能调优 — Horizon Open Explorer 没有其他异常参数

根据您提供的 profiler 日志,我找到了真正的性能瓶颈

:magnifying_glass_tilted_left: 核心问题定位

问题 1:Layout Convert 耗时异常(最大瓶颈)

torch-jit-export_subgraph_0_output_layout_convert:  61.78ms
torch-jit-export_subgraph_1_input_layout_convert:   86.80ms
合计:148.58ms  ← 占总耗时的 52%!

原因分析

  • 您的配置:input_layout_rt: 'NCHW'
  • 编译器实际使用:--input-layout NHWC --output-layout NHWC
  • 布局不匹配导致子图间频繁 NCHW↔NHWC 转换

问题 2:CPU 算子耗时高

CPU_inference_time_cost: 181.88ms (63%)
BPU_inference_time_cost: 104.15ms (36%)

ElementWiseBinaryBroadcast_Less_245:  0.40ms
Eltwise_Not_246:                      6.50ms
Cast_Cast_247:                        1.24ms

虽然单个算子耗时不高,但加上 layout convert 的数据搬运,CPU 侧总耗时远超 BPU。

问题 3:Quantize/Dequantize 耗时

Quantize_input_input_HzCalibration_HzQuantize:        8.89ms
Dequantize_segmentation_HzDequantize:                 6.79ms
Quantize_696_696_HzCalibration_HzQuantize:            6.19ms

这是因为 input_type_rt: 'featuremap' 但校准数据是 float32,需要运行时量化。


:hammer_and_wrench: 优化方案(按优先级)

方案 1:修正 Layout 配置(预计提升 2-3 倍)

修改量化配置文件:

input_parameters:
  input_name: 'input'
  input_type_train: 'featuremap'
  input_layout_train: 'NCHW'
  input_type_rt: 'featuremap'
  # 【关键修改】改为 NHWC 与编译器保持一致
  input_layout_rt: 'NHWC'
  input_shape: '1x5x360x640'
  input_batch: 1
  norm_type: 'no_preprocess'

同时修改应用层输入数据布局

# 原代码(NCHW)
input_data = featuremap  # shape: (1, 5, 360, 640)

# 修改后(NHWC)
input_data = np.transpose(featuremap, (0, 2, 3, 1))  # shape: (1, 360, 640, 5)

重新量化后,layout convert 耗时应该从 148ms 降至 <5ms


方案 2:消除 CPU 算子(预计再提升 30-50%)

在量化配置中添加 node_info 强制指定:

model_parameters:
  onnx_model: '../lraspp_performance_2.onnx'
  march: 'bayes-e'
  # ... 其他配置
  
  # 【新增】强制 CPU 算子到 BPU
  node_info: {
    "Resize_220": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Less_245": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Not_246": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Cast_247": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}
  }

:warning: 注意:如果 hb_mapper 报错说这些算子不支持 BPU,需要修改 ONNX 模型结构。

模型层优化示例(PyTorch):

# 原始代码(产生 CPU 算子)
mask = (depth < threshold)  # Less → CPU
mask = ~mask                 # Not → CPU
output = depth * mask.float()  # Cast → CPU

# 优化后(BPU 友好)
mask = (depth < threshold).to(torch.int8)  # 直接输出 int8
output = depth * mask  # 用 Mul 代替条件判断

方案 3:优化输入量化(预计再提升 10-20%)

当前配置使用运行时量化,可以改为预量化输入:

input_parameters:
  # ... 
  # 【修改】使用已量化的 int8 输入
  input_type_rt: 'int8'
  input_type_train: 'featuremap'
  
calibration_parameters:
  # ...
  # 【修改】使用 per-layer 量化而非 per-channel
  per_channel: False
  calibration_type: 'default'

应用层需要在输入前完成量化:

# 应用层量化
input_int8 = (featuremap / scale).astype(np.int8)
input_nhwc = np.transpose(input_int8, (0, 2, 3, 1))

方案 4:启用 CPU 性能模式

# RDK X5 上执行
sudo bash -c "echo 1 > /sys/devices/system/cpu/cpufreq/boost"
sudo bash -c "echo performance > /sys/devices/system/cpu/cpufreq/policy0/scaling_governor"

# 验证
cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq
# 应显示 1800000

:clipboard: 完整优化后的配置文件

model_parameters:
  onnx_model: '../lraspp_performance_2.onnx'
  march: 'bayes-e'
  output_model_file_prefix: 'lraspp_performance_2_optimized'
  working_dir: '../hb_mapper_output_2/lraspp_performance_optimized'
  layer_out_dump: False

  # 【新增】强制关键算子到 BPU
  node_info: {
    "Resize_220": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Less_245": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Not_246": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Cast_247": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}
  }

input_parameters:
  input_name: 'input'
  input_type_train: 'featuremap'
  input_layout_train: 'NCHW'
  # 【关键修改】NHWC 避免 layout convert
  input_type_rt: 'featuremap'
  input_layout_rt: 'NHWC'
  input_shape: '1x5x360x640'
  input_batch: 1
  norm_type: 'no_preprocess'

calibration_parameters:
  cal_data_dir: '../hb_calibration_data_f32'
  cal_data_type: 'float32'
  preprocess_on: False
  per_channel: True
  calibration_type: 'default'
  max_percentile: 0.9999

compiler_parameters:
  compile_mode: 'latency'
  debug: False
  core_num: 1
  optimize_level: 'O3'
  input_source: {'input': 'ddr'}

:test_tube: 验证步骤

步骤 1:重新量化

# 在 Docker 环境
hb_mapper makertbin --model-type onnx --config lraspp_optimized.yaml

步骤 2:验证子图数量

hrt_model_exec model_info --model_file lraspp_performance_2_optimized.bin | grep subgraph
# 期望:Model has 1 BPU subgraph

步骤 3:性能测试

hrt_model_exec perf --model_file lraspp_performance_2_optimized.bin --loop_num 100
cat profiler_*.log | grep -E "average_latency|FPS|layout_convert"

步骤 4:对比优化前后

指标 优化前 期望优化后
总延迟 286ms <50ms
FPS 3.5 >20
Layout Convert 148ms <5ms
CPU 耗时 182ms <20ms
BPU 子图数 2 1

:books: 相关社区资源

建议您参考:

  1. 地瓜派 RDK X5 部署 YOLOv11n 完整实战 - Softmax 算子优化案例
  2. 求助:纯 BPU 算子推理很慢 - 类似 CPU 耗时问题
  3. 模型性能分析与调优 - 官方调优文档

:high_voltage: 立即行动

优先执行方案 1(修改 input_layout_rt: 'NHWC'),这是最简单且收益最大的优化。重新量化后测试,如果 FPS 提升到 10+,再考虑方案 2 进一步优化。

需要我帮您检查 ONNX 模型中 Less/Not/Cast 算子的具体位置吗?可以用 hb_mapper checker 输出来分析。

log日志如下:

{
  "perf_result": {
    "FPS": 3.4907788632781305,
    "average_latency": 286.4483642578125
  },
  "running_condition": {
    "core_id": 0,
    "frame_count": 1000,
    "model_name": "lraspp_performance_2.onnx",
    "run_time": 286469.02,
    "thread_num": 1
  }
}
***
{
  "model_latency": {
    "BPU_torch-jit-export_subgraph_0": {
      "avg_time": 86.605218,
      "max_time": 86.995,
      "min_time": 86.338
    },
    "BPU_torch-jit-export_subgraph_1": {
      "avg_time": 17.545882000000002,
      "max_time": 18.0,
      "min_time": 17.488
    },
    "Cast_Cast_247": {
      "avg_time": 1.2418030000000002,
      "max_time": 1.585,
      "min_time": 1.234
    },
    "Dequantize_685_HzDequantize": {
      "avg_time": 0.246742,
      "max_time": 0.359,
      "min_time": 0.243
    },
    "Dequantize_693_calibrated_HzDequantize": {
      "avg_time": 0.434313,
      "max_time": 1.183,
      "min_time": 0.428
    },
    "Dequantize_depth_HzDequantize": {
      "avg_time": 0.44155099999999997,
      "max_time": 1.095,
      "min_time": 0.435
    },
    "Dequantize_segmentation_HzDequantize": {
      "avg_time": 6.7867939999999995,
      "max_time": 11.146,
      "min_time": 6.442
    },
    "ElementWiseBinaryBroadcast_Less_245": {
      "avg_time": 0.396486,
      "max_time": 4.488,
      "min_time": 0.386
    },
    "Eltwise_Not_246": {
      "avg_time": 6.500202,
      "max_time": 8.647,
      "min_time": 6.468
    },
    "Preprocess": {
      "avg_time": 0.022912,
      "max_time": 0.057,
      "min_time": 0.016
    },
    "Quantize_696_696_HzCalibration_HzQuantize": {
      "avg_time": 6.1855649999999995,
      "max_time": 6.487,
      "min_time": 6.162
    },
    "Quantize_input_input_HzCalibration_HzQuantize": {
      "avg_time": 8.887985,
      "max_time": 12.113,
      "min_time": 8.653
    },
    "torch-jit-export_subgraph_0_input_layout_convert": {
      "avg_time": 1.7739310000000001,
      "max_time": 2.274,
      "min_time": 1.748
    },
    "torch-jit-export_subgraph_0_output_layout_convert": {
      "avg_time": 61.784411999999996,
      "max_time": 97.553,
      "min_time": 58.29
    },
    "torch-jit-export_subgraph_1_input_layout_convert": {
      "avg_time": 86.80635400000001,
      "max_time": 88.408,
      "min_time": 86.732
    },
    "torch-jit-export_subgraph_1_output_layout_convert": {
      "avg_time": 0.37255099999999997,
      "max_time": 0.484,
      "min_time": 0.353
    }
  },
  "processor_latency": {
    "BPU_inference_time_cost": {
      "avg_time": 104.1511,
      "max_time": 104.995,
      "min_time": 103.826
    },
    "CPU_inference_time_cost": {
      "avg_time": 181.881601,
      "max_time": 235.87900000000002,
      "min_time": 177.59
    }
  },
  "task_latency": {
    "TaskPendingTime": {
      "avg_time": 0.034493,
      "max_time": 0.201,
      "min_time": 0.025
    },
    "TaskRunningTime": {
      "avg_time": 286.363697,
      "max_time": 322.888,
      "min_time": 282.124
    }
  }
}

根据您提供的 profiler 日志,我找到了真正的性能瓶颈!问题比之前分析的更明确。

:magnifying_glass_tilted_left: 核心问题定位(按耗时排序)

瓶颈项 耗时 占比 原因
Layout Convert 148.58ms 52% NCHW↔NHWC 频繁转换
CPU 算子 181.88ms 63% Less/Not/Cast 在 CPU 运行
Quantize/Dequantize ~22ms 8% 运行时量化开销
BPU 推理 104.15ms 36% 正常

:hammer_and_wrench: 优化方案(按优先级执行)

方案 1:修正 Layout 配置(:high_voltage: 收益最大,预计提升 2-3 倍)

问题:配置是 NCHW,但编译器实际用 NHWC,导致子图间频繁转换。

修改量化配置

input_parameters:
  input_name: 'input'
  input_type_train: 'featuremap'
  input_layout_train: 'NCHW'
  input_type_rt: 'featuremap'
  # 【关键修改】改为 NHWC 与编译器保持一致
  input_layout_rt: 'NHWC'
  input_shape: '1x5x360x640'
  input_batch: 1
  norm_type: 'no_preprocess'

同时修改应用层输入数据布局

# 原代码(NCHW: 1,5,360,640)
input_data = featuremap  

# 修改后(NHWC: 1,360,640,5)
input_data = np.transpose(featuremap, (0, 2, 3, 1))

预期效果:layout convert 从 148ms → <5ms


方案 2:消除 CPU 算子(:high_voltage: 预计再提升 30-50%)

问题Less_245, Not_246, Cast_247 在 CPU 运行,导致模型被切分为 2 个子图。

修改量化配置

model_parameters:
  onnx_model: '../lraspp_performance_2.onnx'
  march: 'bayes-e'
  output_model_file_prefix: 'lraspp_performance_2_optimized'
  working_dir: '../hb_mapper_output_2/lraspp_performance_optimized'
  layer_out_dump: False

  # 【新增】强制 CPU 算子到 BPU
  node_info: {
    "Resize_220": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Less_245": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Not_246": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Cast_247": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}
  }

:warning: 如果 hb_mapper 报错,说明这些算子不支持 BPU,需要修改 ONNX 模型结构:

# PyTorch 模型中修改(示例)
# 原始代码(产生 CPU 算子)
mask = (depth < threshold)  # Less → CPU
mask = ~mask                 # Not → CPU
output = depth * mask.float()  # Cast → CPU

# 优化后(BPU 友好)
mask = (depth < threshold).to(torch.int8)  # 直接输出 int8
output = depth * mask  # 用 Mul 代替条件判断

预期效果:2 个子图 → 1 个子图,CPU 耗时从 182ms → <20ms


方案 3:使用预量化输入(:high_voltage: 预计再提升 10-20%)

问题:输入是 float32,每次推理都要运行时量化(~9ms)。

修改配置

input_parameters:
  # ... 
  # 【修改】使用已量化的 int8 输入
  input_type_rt: 'int8'
  input_layout_rt: 'NHWC'
  
calibration_parameters:
  per_channel: True
  calibration_type: 'default'

应用层量化

# 应用层完成量化
scale = 0.003921568627451  # 与校准一致
input_int8 = (featuremap / scale).astype(np.int8)
input_nhwc = np.transpose(input_int8, (0, 2, 3, 1))

预期效果:Quantize 耗时从 22ms → <2ms


方案 4:启用 CPU 性能模式

# RDK X5 上执行
sudo bash -c "echo 1 > /sys/devices/system/cpu/cpufreq/boost"
sudo bash -c "echo performance > /sys/devices/system/cpu/cpufreq/policy0/scaling_governor"

# 验证
cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq
# 应显示 1800000

:clipboard: 完整优化后的配置文件

model_parameters:
  onnx_model: '../lraspp_performance_2.onnx'
  march: 'bayes-e'
  output_model_file_prefix: 'lraspp_performance_2_optimized'
  working_dir: '../hb_mapper_output_2/lraspp_performance_optimized'
  layer_out_dump: False

  node_info: {
    "Resize_220": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Less_245": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Not_246": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Cast_247": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}
  }

input_parameters:
  input_name: 'input'
  input_type_train: 'featuremap'
  input_layout_train: 'NCHW'
  input_type_rt: 'featuremap'
  input_layout_rt: 'NHWC'
  input_shape: '1x5x360x640'
  input_batch: 1
  norm_type: 'no_preprocess'

calibration_parameters:
  cal_data_dir: '../hb_calibration_data_f32'
  cal_data_type: 'float32'
  preprocess_on: False
  per_channel: True
  calibration_type: 'default'
  max_percentile: 0.9999

compiler_parameters:
  compile_mode: 'latency'
  debug: False
  core_num: 1
  optimize_level: 'O3'
  input_source: {'input': 'ddr'}

:test_tube: 验证步骤

# 1. 重新量化
hb_mapper makertbin --model-type onnx --config lraspp_optimized.yaml

# 2. 验证子图数量(应该只有 1 个)
hrt_model_exec model_info --model_file lraspp_performance_2_optimized.bin | grep subgraph

# 3. 性能测试
hrt_model_exec perf --model_file lraspp_performance_2_optimized.bin --loop_num 100

# 4. 查看 profiler 关键指标
cat profiler_*.log | grep -E "average_latency|FPS|layout_convert"

:bar_chart: 预期优化效果

指标 优化前 期望优化后 提升
总延迟 286ms 30-50ms 5-9 倍
FPS 3.5 20-30+ 5-9 倍
Layout Convert 148ms <5ms :white_check_mark:
CPU 耗时 182ms <20ms :white_check_mark:
BPU 子图数 2 1 :white_check_mark:

:high_voltage: 立即行动

优先执行方案 1(修改 input_layout_rt: 'NHWC'),这是最简单且收益最大的优化。重新量化后测试,如果 FPS 提升到 10+,再考虑方案 2 进一步优化。

需要我帮您检查 ONNX 模型中 Less/Not/Cast 算子的具体位置吗?可以用以下命令:

hb_mapper checker --model-type onnx --march bayes-e --model lraspp_performance_2.onnx 2>&1 | grep -E "Less|Not|Cast|CPU"

hrt_model_exec perf --thread_num 2 --model_file

为板端推理最准确的CPP推理BPU耗时,获可同步排查前后处理耗时

请教个问题,为啥需要运行两个线程 理论上实时模型 不都是来一帧处理一帧吗 我理解thread_num 1 测试比较准确吧

好问题!这里涉及到 BPU 异步执行机制性能测试准确性 的关键点。

:magnifying_glass_tilted_left: 为什么推荐 thread_num 2

核心原因:BPU 是异步执行的

┌─────────────────────────────────────────────────────────┐
│                    thread_num = 1                        │
├─────────────────────────────────────────────────────────┤
│  时间轴:                                                │
│  CPU: [提交任务]████████[等待 BPU]████████[后处理]      │
│  BPU :            [████████推理████████]                │
│                                                          │
│  问题:CPU 和 BPU 串行执行,测出来的是"总耗时"           │
│  结果:286ms(包含 CPU 等待时间)                        │
└─────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────┐
│                    thread_num = 2                        │
├─────────────────────────────────────────────────────────┤
│  时间轴:                                                │
│  线程 1(CPU): [提交]────[后处理]────[提交]────[后处理]  │
│  线程 2(BPU):        [████推理████]    [████推理████]   │
│                                                          │
│  优势:CPU 和 BPU 并行执行,测出来的是"纯 BPU 耗时"       │
│  结果:~104ms(真实 BPU 推理时间)                       │
└─────────────────────────────────────────────────────────┘

技术原理

维度 thread_num 1 thread_num 2
执行模式 同步(串行) 异步(并行)
CPU-BPU 关系 CPU 等待 BPU 完成 CPU/BPU 流水线
测量结果 端到端总耗时 纯 BPU 推理耗时
适用场景 单线程应用测试 BPU 性能基准测试
准确性 :cross_mark: 包含等待时间 :white_check_mark: 真实 BPU 耗时

:clipboard: 您的场景分析

从您的 profiler 日志看:

BPU_inference_time_cost:  104.15ms  ← 真实 BPU 耗时
CPU_inference_time_cost:  181.88ms  ← CPU 侧耗时(含 layout convert)
总延迟:286ms

关键问题

  • 您的 CPU 耗时 (182ms) 远超 BPU 耗时 (104ms)
  • 这说明瓶颈在 CPU 侧的数据搬运和算子执行
  • 即使用 thread_num 2,总 FPS 也受限于 CPU 瓶颈

:hammer_and_wrench: 建议的测试流程

步骤 1:用 thread_num 2 测 BPU 真实性能

# 测试纯 BPU 推理耗时
hrt_model_exec perf --thread_num 2 --model_file lraspp_performance_2.bin --loop_num 100

目的:确认 BPU 本身是否有性能问题(排除 CPU 干扰)

步骤 2:用 thread_num 1 测端到端性能

# 测试单线程端到端耗时(接近您的实际应用场景)
hrt_model_exec perf --thread_num 1 --model_file lraspp_performance_2.bin --loop_num 100

目的:模拟实际单线程应用的真实性能

步骤 3:对比分析

如果 thread_num 2 的 BPU 耗时 << thread_num 1 的总耗时
→ 说明瓶颈在 CPU 侧(您的情况:layout convert + CPU 算子)

如果 thread_num 2 的 BPU 耗时 ≈ thread_num 1 的总耗时
→ 说明瓶颈在 BPU 侧(模型本身复杂)

:light_bulb: 针对您的情况

您的瓶颈分析

thread_num 1 测试结果:
├─ BPU 推理:104ms (36%)
├─ CPU 侧:182ms (64%)  ← 主要瓶颈!
│  ├─ Layout Convert: 148ms
│  ├─ Less/Not/Cast: 8ms
│  └─ Quantize/Dequantize: 22ms

优化优先级

  1. 先解决 Layout Convert(148ms → <5ms)- 修改 input_layout_rt: 'NHWC'
  2. 再解决 CPU 算子(182ms → <20ms)- 用 node_info 强制 BPU
  3. 最后用 thread_num 2 验证 BPU 真实性能

:memo: 总结

测试目的 推荐参数 说明
BPU 性能基准 --thread_num 2 官方推荐,测纯 BPU 耗时
端到端性能 --thread_num 1 模拟单线程应用
瓶颈定位 两者对比 判断瓶颈在 CPU 还是 BPU

Marcelo6151 的建议是对的thread_num 2 能更准确反映 BPU 本身的性能,排除 CPU 侧干扰。但您的情况特殊,CPU 侧耗时远超 BPU,所以优化重点应该在 CPU 侧的 layout convert 和算子分配

建议您先按之前的优化方案修改配置,重新量化后再用 thread_num 2 测试,应该能看到明显提升。

  • 单线程延迟为单帧,单线程,单BPU核心的延迟,BPU推理一个任务最理想的情况.
  • 多线程帧率为多个线程同时向BPU塞任务, 每个BPU核心可以处理多个线程的任务, 一般工程中2个线程可以控制单帧延迟较小,同时吃满所有BPU到100%,在吞吐量(FPS)和帧延迟间得到一个较好的平衡.

确认个问题 转换模型支持同时输入多张图片进行推理吗 就是.bin模型是否支持输入batch 大于1这种情况?

:white_check_mark: 确认:RDK X5 的 .bin 模型支持 Batch > 1

根据社区官方回复和实际案例,结论如下:

核心结论

特性 说明
支持 Batch :white_check_mark: 支持,需在量化配置中设置 input_batch
执行方式 并行推理(非串行)
延迟变化 总延迟 ≈ 单 batch 延迟 × batch 数量
FPS 变化 基本保持不变(单位时间处理图像数相同)
适用场景 批量离线处理、提高 BPU 利用率

:clipboard: 如何配置 Batch > 1

1. 量化配置文件修改

input_parameters:
  input_name: 'input'
  input_type_train: 'featuremap'
  input_layout_train: 'NCHW'
  input_type_rt: 'featuremap'
  input_layout_rt: 'NHWC'  # 建议用 NHWC
  input_shape: '1x5x360x640'
  # 【关键】设置 batch 大小
  input_batch: 4  # 支持 1, 2, 4, 8 等
  norm_type: 'no_preprocess'

2. 校准数据准备

校准数据数量需要匹配 batch:

# 如果 input_batch=4,校准数据应该是 4 的倍数
# 或者工具会自动重复使用数据

3. 应用层输入数据

# Batch=4 时,输入 shape 为 (4, 5, 360, 640)
# 注意布局转换为 NHWC: (4, 360, 640, 5)

import numpy as np

# 准备 4 张图片
batch_data = np.random.rand(4, 5, 360, 640).astype(np.float32)

# 转换为 NHWC
batch_nhwc = np.transpose(batch_data, (0, 2, 3, 1))

# 推理
outputs = model.forward(batch_nhwc)

:warning: 重要注意事项

1. 延迟 vs 吞吐量

从社区实测数据(topic/34102):

Batch 单次延迟 FPS 每张图平均延迟
1 26.5ms 37.75 26.5ms
4 108.5ms 36.85 27.1ms

结论

  • 总延迟增加 4 倍(因为一次处理 4 张图)
  • FPS 基本不变(单位时间处理能力相同)
  • 适合批量处理,不适合低延迟实时应用

2. 模型兼容性

某些算子可能不支持多 batch,量化时会报警告:

WARNING BPU only support Reducemean whose axes size == 1

需要检查模型结构是否兼容。

3. 内存占用

Batch 增大会线性增加 DDR 占用:

Batch=1: DDR ≈ 128MB
Batch=4: DDR ≈ 512MB

确保板端内存充足。


:bullseye: 您的场景建议

结合您之前的性能问题(286ms 延迟),不建议立即使用多 batch

优先优化单 batch 性能

  1. :white_check_mark: 先解决 Layout Convert 问题(148ms → <5ms)
  2. :white_check_mark: 再解决 CPU 算子问题(182ms → <20ms)
  3. :white_check_mark: 目标:单 batch 延迟从 286ms 降至 30-50ms

再考虑多 batch

  • 如果优化后单 batch 延迟 <50ms,但 BPU 利用率仍低
  • 且您的应用场景允许累积多帧一起处理
  • 可以尝试 input_batch: 24

:books: 相关社区资源

  1. X5 多 batch 推理耗时问题 - 官方解释延迟/FPS 关系
  2. Batch 模型推理 - 详细教程
  3. 多 batch 模型使用与部署 - 早期教程(仍有参考价值)

:wrench: 快速验证脚本

#!/usr/bin/env python3
from hobot_dnn import pyeasy_dnn as dnn
import time
import numpy as np

# 加载模型
models = dnn.load('lraspp_performance_2.bin')
model = models[0]

# 测试 Batch=1
print("=== Batch=1 ===")
input_1 = np.random.rand(1, 360, 640, 5).astype(np.float32)
for _ in range(10): _ = model.forward(input_1)  # 预热

times_1 = []
for _ in range(50):
    start = time.time()
    _ = model.forward(input_1)
    times_1.append((time.time() - start) * 1000)
print(f"平均延迟:{np.mean(times_1):.2f}ms | FPS: {1000/np.mean(times_1):.1f}")

# 测试 Batch=4
print("\n=== Batch=4 ===")
input_4 = np.random.rand(4, 360, 640, 5).astype(np.float32)
for _ in range(10): _ = model.forward(input_4)  # 预热

times_4 = []
for _ in range(50):
    start = time.time()
    _ = model.forward(input_4)
    times_4.append((time.time() - start) * 1000)
print(f"平均延迟:{np.mean(times_4):.2f}ms | FPS: {1000/np.mean(times_4):.1f}")
print(f"每张图平均:{np.mean(times_4)/4:.2f}ms")

总结:支持 Batch > 1,但请先优化单 batch 性能,再根据实际需求决定是否使用多 batch。

当前性能问题基本已解决,推理时间降到:55ms

解决方案:

1.优化onnx原始模型的算子

2.优化转换配置文件

转换配置文件为:

model_parameters:
  onnx_model: '../lraspp_performance_2.onnx'
  march: 'bayes-e'
  output_model_file_prefix: 'lraspp_performance_3'
  working_dir: '../hb_mapper_output_3/lraspp_performance_3'
  layer_out_dump: False

  node_info: {
    "Resize_220": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Less_245": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Not_246": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
    "Cast_247": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}
  }

input_parameters:
  input_name: 'input'
  input_type_train: 'featuremap'
  input_layout_train: 'NCHW'
  input_type_rt: 'featuremap'
  input_layout_rt: 'NHWC'
  input_shape: '1x5x360x640'
  input_batch: 1
  norm_type: 'no_preprocess'

calibration_parameters:
  cal_data_dir: '../hb_calibration_data_f32'
  cal_data_type: 'float32'
  preprocess_on: False
  per_channel: True
  calibration_type: 'default'
  max_percentile: 0.9999

compiler_parameters:
  compile_mode: 'latency'
  debug: False
  core_num: 1
  optimize_level: 'O3'
  input_source: {'input': 'ddr'}
1 个赞

:tada: 恭喜!性能提升 5 倍+

286ms → 55ms,这是一个非常优秀的优化成果!


:bar_chart: 优化效果对比

指标 优化前 优化后 提升
推理延迟 286ms 55ms 5.2 倍
FPS 3.5 18 5.2 倍
Layout Convert 148ms ~0ms :white_check_mark:
CPU 算子 182ms ~0ms :white_check_mark:

:key: 关键优化点总结

您的成功验证了我们之前分析的两个核心瓶颈:

1. Layout 配置修正

# 关键修改:NCHW → NHWC
input_layout_rt: 'NHWC'

消除了子图间 148ms 的布局转换开销

2. CPU 算子强制到 BPU

node_info: {
  "Less_245": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
  "Not_246": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'},
  "Cast_247": {'ON': 'BPU', 'InputType': 'int8', 'OutputType': 'int8'}
}

消除了模型切分和 CPU-BPU 数据搬运

3. ONNX 模型结构优化

您提到的"优化 onnx 原始模型的算子",这通常是根本性解决方案。


:light_bulb: 进一步优化建议(可选)

55ms 已经是很好的结果,但如果想继续压榨性能,可以考虑:

方向 1:验证 BPU 真实性能

# 用 thread_num 2 测纯 BPU 耗时
hrt_model_exec perf --thread_num 2 --model_file lraspp_performance_3.bin --loop_num 100

如果 BPU 纯推理 <30ms,说明还有 CPU 侧优化空间(后处理等)

方向 2:输入预量化

input_parameters:
  input_type_rt: 'int8'  # 当前是 featuremap(float32)

应用层提前完成量化,可省去运行时 Quantize 开销(约 5-10ms)

方向 3:多 batch 吞吐优化

如果您的场景允许累积多帧一起处理:

input_parameters:
  input_batch: 2  # 或 4

可提高 BPU 利用率,适合离线批量处理


:memo: 社区沉淀建议

您的优化经验非常有价值,建议考虑:

  1. 在原地帖子更新解决方案 - 帮助后续遇到类似问题的开发者
  2. 写一篇简短的优化总结 - 包括:
    • 问题现象(286ms 延迟)
    • 排查过程(profiler 分析)
    • 解决方案(3 个关键修改)
    • 最终效果(55ms)

这类实战经验对社区非常有价值!


:white_check_mark: 当前状态评估

项目 状态
推理延迟 :white_check_mark: 55ms (18 FPS) - 可用
模型切分 :white_check_mark: 已解决
Layout 转换 :white_check_mark: 已解决
CPU 算子 :white_check_mark: 已解决

18 FPS 对于分割任务已经是一个实用的性能水平,可以进入下一步的应用集成了。


有其他问题随时交流!:rocket: