yolo11模型转换时，在calibration阶段发生很大的精度损失，有些类型没有检测出来，但是onnx原始模型是可以检测出来

开发与问题 Model Zoo
1

求助：这个是否有优化的方案，配置转换文件或者标定校验数据集生成等方面

问题出现在：标定转换阶段

转换配置文件如下：

# yolo11n_config_optimized.yaml - 最终优化版配置

# ============================================
# YOLOv11n RDK X5 高性能PTQ量化配置
# 核心：解决Softmax性能瓶颈，实现100+ FPS
# ============================================

model_parameters:
  # ONNX模型路径（Docker内路径）
  onnx_model: '../onnx_model/best_1026.onnx'
  
  # 目标平台：RDK X5使用bayes-e架构
  march: 'bayes-e'
  
  # 不输出中间层结果（加快编译速度）
  layer_out_dump: False
  
  # 输出目录和文件前缀
  working_dir: '../hb_mapper_output_3'
  output_model_file_prefix: 'yolo26_detect_bayese_640x384_nv12'
  
  # ============================================
  # 【核心配置】解决Softmax性能瓶颈
  # ============================================
  # 将C2PSA注意力模块中的Softmax算子指定到BPU运行
  # 使用int16精度（Softmax不支持int8，但支持int16）
  # 
  # 不同模型的配置：
  # - YOLOv11 n/s/m: 只有1个Softmax
  # - YOLOv11 l/x: 有2个Softmax（需要配置两个）
  node_info: {
    "/model.10/m/m.0/attn/Softmax": {
      'ON': 'BPU',
      'InputType': 'int16',
      'OutputType': 'int16'
    },
    "/model.22/m.0/m.0.1/attn/Softmax": {
      'ON': 'BPU',
      'InputType': 'int16',
      'OutputType': 'int16'
    }
  }

# ============================================
# 输入参数配置
# ============================================
input_parameters:
  # 输入节点名称（留空自动识别）
  input_name: "images"
  
  # 【关键】运行时输入格式：nv12
  # RDK X5的BPU原生支持nv12格式，无需CPU转换
  # 性能提升：省去BGR->RGB转换时间
  input_type_rt: 'nv12'
  
  # 训练时输入格式：rgb
  # YOLO模型训练使用RGB格式
  input_type_train: 'rgb'
  
  # 输入数据布局：NCHW
  # PyTorch/ONNX默认使用NCHW (Batch, Channels, Height, Width)
  input_layout_train: 'NCHW'
  
  # 输入shape（留空自动从ONNX读取）
  input_shape: '2x3x384x640'

  # input_batch: 1
  
  # 归一化类型：data_scale
  # 使用简单的缩放归一化 (x * scale)
  norm_type: 'data_scale'
  
  # 归一化系数：1/255
  # YOLO标准预处理：将[0,255]映射到[0,1]
  scale_value: 0.003921568627451

# ============================================
# 校准参数配置
# ============================================
calibration_parameters:
  # 校准数据目录
  cal_data_dir: '../calibration_data_bgr_f32_255_generated'
  
  # 校准数据类型
  cal_data_type: 'float32'
  
  # 【关键】校准方法：default
  # default: 自动搜索最佳量化策略（推荐）
  # 其他选项：
  #   - kl: KL散度最小化（精度高，编译慢）
  #   - max: 最大值法（速度快，精度略低）
  #   - mix: 混合策略（精度最高，编译最慢）
  calibration_type: 'default'

# ============================================
# 编译器参数配置
# ============================================
compiler_parameters:
  # 编译模式：latency
  # latency: 低延迟优先（单帧推理快）
  # throughput: 吞吐量优先（批量推理快）
  compile_mode: 'latency'
  
  # 调试模式：关闭（加快编译）
  debug: False
  
  # 【关键】优化等级：O3
  # O0: 不优化（调试用）
  # O1: 基本优化
  # O2: 标准优化
  # O3: 最高优化（推荐）
  optimize_level: 'O3'

转换过程的makertbin日志如下：

======================================================================================================================================
Node                                                ON   Subgraph  Type                       Cosine Similarity  Threshold  DataType  
--------------------------------------------------------------------------------------------------------------------------------------
HZ_PREPROCESS_FOR_images                            BPU  id(0)     HzSQuantizedPreprocess     1.000019           127.0      int8      
/model.0/conv/Conv                                  BPU  id(0)     HzSQuantizedConv           0.999732           1.06558    int8      
/model.0/act/Mul                                    BPU  id(0)     HzLut                      0.999489           52.7776    int8      
/model.1/conv/Conv                                  BPU  id(0)     HzSQuantizedConv           0.997429           49.5613    int8      
/model.1/act/Mul                                    BPU  id(0)     HzLut                      0.997215           81.4011    int8      
/model.2/cv1/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.991194           71.512     int8      
/model.2/cv1/act/Mul                                BPU  id(0)     HzLut                      0.991340           68.5251    int8      
/model.2/Split                                      BPU  id(0)     Split                      0.991442           35.0651    int8      
/model.2/m.0/cv1/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.997627           35.0651    int8      
/model.2/m.0/cv1/act/Mul                            BPU  id(0)     HzLut                      0.997632           6.50954    int8      
/model.2/m.0/cv2/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.991058           5.2504     int8      
/model.2/m.0/cv2/act/Mul                            BPU  id(0)     HzLut                      0.993152           35.6744    int8      
/model.2/m.0/Add                                    BPU  id(0)     HzSElementwiseAdd          0.993931           35.0651    int8      
/model.2/Split_output_0_calibrated_Requantize       BPU  id(0)     HzRequantize               --                 --         int8      
/model.2/Split_output_1_calibrated_Requantize       BPU  id(0)     HzRequantize               --                 --         int8      
/model.2/Concat                                     BPU  id(0)     Concat                     0.992831           35.0651    int8      
/model.2/cv2/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.988324           38.0301    int8      
/model.2/cv2/act/Mul                                BPU  id(0)     HzLut                      0.987918           49.578     int8      
/model.3/conv/Conv                                  BPU  id(0)     HzSQuantizedConv           0.993886           16.4691    int8      
/model.3/act/Mul                                    BPU  id(0)     HzLut                      0.996837           11.5675    int8      
/model.4/cv1/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.994723           11.5555    int8      
/model.4/cv1/act/Mul                                BPU  id(0)     HzLut                      0.995272           8.77371    int8      
/model.4/Split                                      BPU  id(0)     Split                      0.994949           8.19899    int8      
/model.4/m.0/cv1/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.993011           8.19899    int8      
/model.4/m.0/cv1/act/Mul                            BPU  id(0)     HzLut                      0.993700           5.54661    int8      
/model.4/m.0/cv2/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.996701           5.52421    int8      
/model.4/m.0/cv2/act/Mul                            BPU  id(0)     HzLut                      0.996739           12.4207    int8      
/model.4/m.0/Add                                    BPU  id(0)     HzSElementwiseAdd          0.997926           8.19899    int8      
/model.4/Split_output_0_calibrated_Requantize       BPU  id(0)     HzRequantize               --                 --         int8      
/model.4/Split_output_1_calibrated_Requantize       BPU  id(0)     HzRequantize               --                 --         int8      
/model.4/Concat                                     BPU  id(0)     Concat                     0.997242           8.19899    int8      
/model.4/cv2/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.990299           13.8193    int8      
/model.4/cv2/act/Mul                                BPU  id(0)     HzLut                      0.982934           7.36125    int8      
/model.5/conv/Conv                                  BPU  id(0)     HzSQuantizedConv           0.996112           5.13148    int8      
/model.5/act/Mul                                    BPU  id(0)     HzLut                      0.997732           11.6098    int8      
/model.6/cv1/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.993812           8.8449     int8      
/model.6/cv1/act/Mul                                BPU  id(0)     HzLut                      0.995297           10.984     int8      
/model.6/Split                                      BPU  id(0)     Split                      0.995547           9.99939    int8      
/model.6/m.0/cv1/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.994160           9.99939    int8      
/model.6/m.0/cv2/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.992959           9.99939    int8      
/model.6/m.0/cv1/act/Mul                            BPU  id(0)     HzLut                      0.994629           7.63686    int8      
/model.6/m.0/cv2/act/Mul                            BPU  id(0)     HzLut                      0.987342           10.4442    int8      
/model.6/m.0/m/m.0/cv1/conv/Conv                    BPU  id(0)     HzSQuantizedConv           0.993997           7.62414    int8      
/model.6/m.0/m/m.0/cv1/act/Mul                      BPU  id(0)     HzLut                      0.989666           8.5377     int8      
/model.6/m.0/m/m.0/cv2/conv/Conv                    BPU  id(0)     HzSQuantizedConv           0.994245           6.8799     int8      
/model.6/m.0/m/m.0/cv2/act/Mul                      BPU  id(0)     HzLut                      0.994528           9.12017    int8      
/model.6/m.0/m/m.0/Add                              BPU  id(0)     HzSElementwiseAdd          0.995331           7.62414    int8      
/model.6/m.0/m/m.1/cv1/conv/Conv                    BPU  id(0)     HzSQuantizedConv           0.994267           10.1904    int8      
/model.6/m.0/m/m.1/cv1/act/Mul                      BPU  id(0)     HzLut                      0.992772           9.79045    int8      
/model.6/m.0/m/m.1/cv2/conv/Conv                    BPU  id(0)     HzSQuantizedConv           0.997495           7.37776    int8      
/model.6/m.0/m/m.1/cv2/act/Mul                      BPU  id(0)     HzLut                      0.997742           17.0881    int8      
/model.6/m.0/m/m.1/Add                              BPU  id(0)     HzSElementwiseAdd          0.997956           10.1904    int8      
/model.6/m.0/Concat                                 BPU  id(0)     Concat                     0.997583           18.4174    int8      
/model.6/m.0/cv3/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.993775           18.4174    int8      
/model.6/m.0/cv3/act/Mul                            BPU  id(0)     HzLut                      0.994581           13.0001    int8      
/model.6/Split_output_0_calibrated_Requantize       BPU  id(0)     HzRequantize               --                 --         int8      
/model.6/Split_output_1_calibrated_Requantize       BPU  id(0)     HzRequantize               --                 --         int8      
/model.6/Concat                                     BPU  id(0)     Concat                     0.994917           9.99939    int8      
/model.6/cv2/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.991047           12.1735    int8      
/model.6/cv2/act/Mul                                BPU  id(0)     HzLut                      0.988720           11.6538    int8      
/model.7/conv/Conv                                  BPU  id(0)     HzSQuantizedConv           0.993862           8.9875     int8      
/model.7/act/Mul                                    BPU  id(0)     HzLut                      0.992382           12.1365    int8      
/model.8/cv1/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.992455           10.9503    int8      
/model.8/cv1/act/Mul                                BPU  id(0)     HzLut                      0.989855           13.5282    int8      
/model.8/Split                                      BPU  id(0)     Split                      0.990514           12.9413    int8      
/model.8/m.0/cv1/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.994192           12.9413    int8      
/model.8/m.0/cv2/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.996736           12.9413    int8      
/model.8/m.0/cv1/act/Mul                            BPU  id(0)     HzLut                      0.993121           12.7957    int8      
/model.8/m.0/cv2/act/Mul                            BPU  id(0)     HzLut                      0.982519           13.0792    int8      
/model.8/m.0/m/m.0/cv1/conv/Conv                    BPU  id(0)     HzSQuantizedConv           0.996342           12.7957    int8      
/model.8/m.0/m/m.0/cv1/act/Mul                      BPU  id(0)     HzLut                      0.993493           12.0771    int8      
/model.8/m.0/m/m.0/cv2/conv/Conv                    BPU  id(0)     HzSQuantizedConv           0.995928           11.671     int8      
/model.8/m.0/m/m.0/cv2/act/Mul                      BPU  id(0)     HzLut                      0.994394           15.6022    int8      
/model.8/m.0/m/m.0/Add                              BPU  id(0)     HzSElementwiseAdd          0.993994           12.7957    int8      
/model.8/m.0/m/m.1/cv1/conv/Conv                    BPU  id(0)     HzSQuantizedConv           0.996991           13.8258    int8      
/model.8/m.0/m/m.1/cv1/act/Mul                      BPU  id(0)     HzLut                      0.994932           9.5978     int8      
/model.8/m.0/m/m.1/cv2/conv/Conv                    BPU  id(0)     HzSQuantizedConv           0.997265           9.37617    int8      
/model.8/m.0/m/m.1/cv2/act/Mul                      BPU  id(0)     HzLut                      0.997877           17.9276    int8      
/model.8/m.0/m/m.1/Add                              BPU  id(0)     HzSElementwiseAdd          0.997401           13.8258    int8      
/model.8/m.0/Concat                                 BPU  id(0)     Concat                     0.997019           18.8416    int8      
/model.8/m.0/cv3/conv/Conv                          BPU  id(0)     HzSQuantizedConv           0.996908           18.8416    int8      
/model.8/m.0/cv3/act/Mul                            BPU  id(0)     HzLut                      0.996919           10.6319    int8      
/model.8/Split_output_0_calibrated_Requantize       BPU  id(0)     HzRequantize               --                 --         int8      
/model.8/Split_output_1_calibrated_Requantize       BPU  id(0)     HzRequantize               --                 --         int8      
/model.8/Concat                                     BPU  id(0)     Concat                     0.992958           12.9413    int8      
/model.8/cv2/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.994937           12.5725    int8      
/model.8/cv2/act/Mul                                BPU  id(0)     HzLut                      0.992459           11.611     int8      
/model.9/cv1/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.994406           8.31991    int8      
/model.9/m/MaxPool                                  BPU  id(0)     HzQuantizedMaxPool         0.996616           15.6738    int8      
/model.9/m_1/MaxPool                                BPU  id(0)     HzQuantizedMaxPool         0.997294           15.6738    int8      
/model.9/m_2/MaxPool                                BPU  id(0)     HzQuantizedMaxPool         0.997619           15.6738    int8      
/model.9/Concat                                     BPU  id(0)     Concat                     0.996923           15.6738    int8      
/model.9/cv2/conv/Conv                              BPU  id(0)     HzSQuantizedConv           0.994304           15.6738    int8      
/model.9/cv2/act/Mul                                BPU  id(0)     HzLut                      0.988605           12.7024    int8      
/model.9/Add                                        BPU  id(0)     HzSElementwiseAdd          0.991013           12.143     int8      
/model.10/cv1/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.988946           12.2601    int8      
/model.10/cv1/act/Mul                               BPU  id(0)     HzLut                      0.987605           18.0694    int8      
/model.10/Split                                     BPU  id(0)     Split                      0.983788           15.9709    int8      
/model.10/m/m.0/attn/qkv/conv/Conv                  BPU  id(0)     HzSQuantizedConv           0.994459           15.9709    int8      
/model.10/m/m.0/attn/Reshape                        BPU  id(0)     Reshape                    0.994459           8.6198     int8      
/model.10/m/m.0/attn/Split                          BPU  id(0)     Split                      0.994087           8.6198     int8      
/model.10/m/m.0/attn/Mul                            BPU  id(0)     HzSElementwiseMul          0.994016           8.6198     int8      
/model.10/m/m.0/attn/Reshape_2                      BPU  id(0)     Reshape                    0.992376           8.6198     int8      
/model.10/m/m.0/attn/Transpose                      BPU  id(0)     Transpose                  0.994016           1.11505    int8      
/model.10/m/m.0/attn/MatMul                         BPU  id(0)     HzSQuantizedMatmul         0.993305           1.11505    int8      
...0/attn/Softmax_reducemax_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzQuantizedReduceMax       0.997026           19.6331    int16     
...0/m/m.0/attn/Softmax_sub_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzSElementwiseSub          0.992263           19.6331    int16     
...0/m/m.0/attn/Softmax_exp_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzLut2Layer                0.978543           11.0903    int16     
...0/attn/Softmax_reducesum_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzSQuantizedReduceSum      0.987868           1.0        int16     
.../attn/Softmax_reciprocal_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzLut2Layer                0.969905           211.834    int16     
...0/m/m.0/attn/Softmax_mul_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzSElementwiseMul          0.955807           1.0        int16     
/model.10/m/m.0/attn/Transpose_1                    BPU  id(0)     Transpose                  0.955807           0.286957   int8      
/model.10/m/m.0/attn/MatMul_1                       BPU  id(0)     HzSQuantizedMatmul         0.996460           8.6198     int8      
/model.10/m/m.0/attn/Reshape_1                      BPU  id(0)     Reshape                    0.996460           6.29675    int8      
/model.10/m/m.0/attn/pe/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.993123           8.6198     int8      
/model.10/m/m.0/attn/proj/conv/Conv                 BPU  id(0)     HzSQuantizedConv           0.997752           14.9343    int8      
/model.10/m/m.0/ffn/ffn.0/conv/Conv                 BPU  id(0)     HzSQuantizedConv           0.997853           9.50766    int8      
/model.10/m/m.0/ffn/ffn.0/act/Mul                   BPU  id(0)     HzLut                      0.984695           11.0806    int8      
/model.10/m/m.0/ffn/ffn.1/conv/Conv                 BPU  id(0)     HzSQuantizedConv           0.998866           5.87139    int8      
/model.10/Split_output_0_calibrated_Requantize      BPU  id(0)     HzRequantize               --                 --         int8      
/model.10/Concat                                    BPU  id(0)     Concat                     0.996307           15.9709    int8      
/model.10/cv2/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.996109           17.322     int8      
/model.10/cv2/act/Mul                               BPU  id(0)     HzLut                      0.988603           12.8848    int8      
/model.11/Resize                                    BPU  id(0)     HzQuantizedResizeUpsample  0.988596           11.1868    int8      
/model.11/Resize_output_0_calibrated_Requantize     BPU  id(0)     HzRequantize               --                 --         int8      
...el.6/cv2/act/Mul_output_0_calibrated_Requantize  BPU  id(0)     HzRequantize               --                 --         int8      
/model.12/Concat                                    BPU  id(0)     Concat                     0.988436           11.1868    int8      
/model.13/cv1/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.992549           11.232     int8      
/model.13/cv1/act/Mul                               BPU  id(0)     HzLut                      0.993898           11.2164    int8      
/model.13/Split                                     BPU  id(0)     Split                      0.989571           8.10003    int8      
/model.13/m.0/cv1/conv/Conv                         BPU  id(0)     HzSQuantizedConv           0.990191           8.10003    int8      
/model.13/m.0/cv2/conv/Conv                         BPU  id(0)     HzSQuantizedConv           0.992602           8.10003    int8      
/model.13/m.0/cv1/act/Mul                           BPU  id(0)     HzLut                      0.990692           5.98212    int8      
/model.13/m.0/cv2/act/Mul                           BPU  id(0)     HzLut                      0.992617           12.0107    int8      
/model.13/m.0/m/m.0/cv1/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.991340           5.8906     int8      
/model.13/m.0/m/m.0/cv1/act/Mul                     BPU  id(0)     HzLut                      0.979534           9.4477     int8      
/model.13/m.0/m/m.0/cv2/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.987153           5.40344    int8      
/model.13/m.0/m/m.0/cv2/act/Mul                     BPU  id(0)     HzLut                      0.987669           9.0412     int8      
/model.13/m.0/m/m.0/Add                             BPU  id(0)     HzSElementwiseAdd          0.990258           5.8906     int8      
/model.13/m.0/m/m.1/cv1/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.990033           10.0449    int8      
/model.13/m.0/m/m.1/cv1/act/Mul                     BPU  id(0)     HzLut                      0.984161           8.4322     int8      
/model.13/m.0/m/m.1/cv2/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.996380           5.98247    int8      
/model.13/m.0/m/m.1/cv2/act/Mul                     BPU  id(0)     HzLut                      0.996621           15.7835    int8      
/model.13/m.0/m/m.1/Add                             BPU  id(0)     HzSElementwiseAdd          0.996704           10.0449    int8      
/model.13/m.0/Concat                                BPU  id(0)     Concat                     0.996453           18.8438    int8      
/model.13/m.0/cv3/conv/Conv                         BPU  id(0)     HzSQuantizedConv           0.989163           18.8438    int8      
/model.13/m.0/cv3/act/Mul                           BPU  id(0)     HzLut                      0.988511           9.67236    int8      
/model.13/Split_output_0_calibrated_Requantize      BPU  id(0)     HzRequantize               --                 --         int8      
/model.13/Split_output_1_calibrated_Requantize      BPU  id(0)     HzRequantize               --                 --         int8      
/model.13/Concat                                    BPU  id(0)     Concat                     0.992321           8.10003    int8      
/model.13/cv2/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.994761           7.99959    int8      
/model.13/cv2/act/Mul                               BPU  id(0)     HzLut                      0.993724           11.7391    int8      
/model.14/Resize                                    BPU  id(0)     HzQuantizedResizeUpsample  0.993735           9.15482    int8      
/model.14/Resize_output_0_calibrated_Requantize     BPU  id(0)     HzRequantize               --                 --         int8      
...el.4/cv2/act/Mul_output_0_calibrated_Requantize  BPU  id(0)     HzRequantize               --                 --         int8      
/model.15/Concat                                    BPU  id(0)     Concat                     0.991329           9.15482    int8      
/model.16/cv1/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.993768           7.85915    int8      
/model.16/cv1/act/Mul                               BPU  id(0)     HzLut                      0.996146           7.72201    int8      
/model.16/Split                                     BPU  id(0)     Split                      0.996802           7.64008    int8      
/model.16/m.0/cv1/conv/Conv                         BPU  id(0)     HzSQuantizedConv           0.987927           7.64008    int8      
/model.16/m.0/cv2/conv/Conv                         BPU  id(0)     HzSQuantizedConv           0.992545           7.64008    int8      
/model.16/m.0/cv1/act/Mul                           BPU  id(0)     HzLut                      0.989752           2.63829    int8      
/model.16/m.0/cv2/act/Mul                           BPU  id(0)     HzLut                      0.995837           8.67095    int8      
/model.16/m.0/m/m.0/cv1/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.977111           2.39589    int8      
/model.16/m.0/m/m.0/cv1/act/Mul                     BPU  id(0)     HzLut                      0.972480           6.46695    int8      
/model.16/m.0/m/m.0/cv2/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.963601           6.45632    int8      
/model.16/m.0/m/m.0/cv2/act/Mul                     BPU  id(0)     HzLut                      0.962082           7.08995    int8      
/model.16/m.0/m/m.0/Add                             BPU  id(0)     HzSElementwiseAdd          0.973204           2.39589    int8      
/model.16/m.0/m/m.1/cv1/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.983897           6.713      int8      
/model.16/m.0/m/m.1/cv1/act/Mul                     BPU  id(0)     HzLut                      0.992543           6.95126    int8      
/model.16/m.0/m/m.1/cv2/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.991065           5.24391    int8      
/model.16/m.0/m/m.1/cv2/act/Mul                     BPU  id(0)     HzLut                      0.992416           18.7882    int8      
/model.16/m.0/m/m.1/Add                             BPU  id(0)     HzSElementwiseAdd          0.992266           6.713      int8      
/model.16/m.0/Concat                                BPU  id(0)     Concat                     0.992603           20.9158    int8      
/model.16/m.0/cv3/conv/Conv                         BPU  id(0)     HzSQuantizedConv           0.977139           20.9158    int8      
/model.16/m.0/cv3/act/Mul                           BPU  id(0)     HzLut                      0.982545           10.3113    int8      
/model.16/Split_output_0_calibrated_Requantize      BPU  id(0)     HzRequantize               --                 --         int8      
/model.16/Split_output_1_calibrated_Requantize      BPU  id(0)     HzRequantize               --                 --         int8      
/model.16/Concat                                    BPU  id(0)     Concat                     0.989395           7.64008    int8      
/model.16/cv2/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.983839           9.18459    int8      
/model.16/cv2/act/Mul                               BPU  id(0)     HzLut                      0.988381           10.9257    int8      
/model.17/conv/Conv                                 BPU  id(0)     HzSQuantizedConv           0.972366           8.28969    int8      
/model.23/one2one_cv2.0/one2one_cv2.0.0/conv/Conv   BPU  id(0)     HzSQuantizedConv           0.990267           8.28969    int8      
...3.0/one2one_cv3.0.0/one2one_cv3.0.0.0/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.997193           8.28969    int8      
/model.17/act/Mul                                   BPU  id(0)     HzLut                      0.971700           10.5794    int8      
/model.23/one2one_cv2.0/one2one_cv2.0.0/act/Mul     BPU  id(0)     HzLut                      0.992458           14.407     int8      
...cv3.0/one2one_cv3.0.0/one2one_cv3.0.0.0/act/Mul  BPU  id(0)     HzLut                      0.997646           26.879     int8      
/model.18/Concat                                    BPU  id(0)     Concat                     0.982128           9.15482    int8      
/model.23/one2one_cv2.0/one2one_cv2.0.1/conv/Conv   BPU  id(0)     HzSQuantizedConv           0.987804           14.3902    int8      
...3.0/one2one_cv3.0.0/one2one_cv3.0.0.1/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.985759           26.8655    int8      
/model.19/cv1/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.987044           9.15482    int8      
/model.23/one2one_cv2.0/one2one_cv2.0.1/act/Mul     BPU  id(0)     HzLut                      0.978245           10.8777    int8      
...cv3.0/one2one_cv3.0.0/one2one_cv3.0.0.1/act/Mul  BPU  id(0)     HzLut                      0.987041           16.757     int8      
/model.19/cv1/act/Mul                               BPU  id(0)     HzLut                      0.987558           10.5891    int8      
/model.23/one2one_cv2.0/one2one_cv2.0.2/Conv        BPU  id(0)     HzSQuantizedConv           0.991145           3.23666    int8      
...3.0/one2one_cv3.0.1/one2one_cv3.0.1.0/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.983911           16.4124    int8      
/model.19/Split                                     BPU  id(0)     Split                      0.992451           10.1841    int8      
/model.19/m.0/cv1/conv/Conv                         BPU  id(0)     HzSQuantizedConv           0.978631           10.1841    int8      
/model.19/m.0/cv2/conv/Conv                         BPU  id(0)     HzSQuantizedConv           0.966232           10.1841    int8      
...cv3.0/one2one_cv3.0.1/one2one_cv3.0.1.0/act/Mul  BPU  id(0)     HzLut                      0.986016           28.9536    int8      
...3.0/one2one_cv3.0.1/one2one_cv3.0.1.1/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.978907           25.1463    int8      
/model.19/m.0/cv1/act/Mul                           BPU  id(0)     HzLut                      0.980538           4.91409    int8      
/model.19/m.0/cv2/act/Mul                           BPU  id(0)     HzLut                      0.946003           11.7768    int8      
/model.19/m.0/m/m.0/cv1/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.963481           4.14478    int8      
...cv3.0/one2one_cv3.0.1/one2one_cv3.0.1.1/act/Mul  BPU  id(0)     HzLut                      0.972869           41.4765    int8      
/model.23/one2one_cv3.0/one2one_cv3.0.2/Conv        BPU  id(0)     HzSQuantizedConv           0.995471           41.4765    int8      
/model.19/m.0/m/m.0/cv1/act/Mul                     BPU  id(0)     HzLut                      0.919875           9.64124    int8      
/model.19/m.0/m/m.0/cv2/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.956819           7.01682    int8      
/model.19/m.0/m/m.0/cv2/act/Mul                     BPU  id(0)     HzLut                      0.959501           9.0007     int8      
/model.19/m.0/m/m.0/Add                             BPU  id(0)     HzSElementwiseAdd          0.972265           4.14478    int8      
/model.19/m.0/m/m.1/cv1/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.955313           7.50536    int8      
/model.19/m.0/m/m.1/cv1/act/Mul                     BPU  id(0)     HzLut                      0.927178           9.01211    int8      
/model.19/m.0/m/m.1/cv2/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.967776           8.05438    int8      
/model.19/m.0/m/m.1/cv2/act/Mul                     BPU  id(0)     HzLut                      0.973260           25.0082    int8      
/model.19/m.0/m/m.1/Add                             BPU  id(0)     HzSElementwiseAdd          0.976059           7.50536    int8      
/model.19/m.0/Concat                                BPU  id(0)     Concat                     0.974777           25.6693    int8      
/model.19/m.0/cv3/conv/Conv                         BPU  id(0)     HzSQuantizedConv           0.949662           25.6693    int8      
/model.19/m.0/cv3/act/Mul                           BPU  id(0)     HzLut                      0.952762           15.021     int8      
/model.19/Split_output_0_calibrated_Requantize      BPU  id(0)     HzRequantize               --                 --         int8      
/model.19/Split_output_1_calibrated_Requantize      BPU  id(0)     HzRequantize               --                 --         int8      
/model.19/Concat                                    BPU  id(0)     Concat                     0.970868           10.1841    int8      
/model.19/cv2/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.982390           12.046     int8      
/model.19/cv2/act/Mul                               BPU  id(0)     HzLut                      0.976443           14.3812    int8      
/model.20/conv/Conv                                 BPU  id(0)     HzSQuantizedConv           0.985337           11.8647    int8      
/model.23/one2one_cv2.1/one2one_cv2.1.0/conv/Conv   BPU  id(0)     HzSQuantizedConv           0.977505           11.8647    int8      
...3.1/one2one_cv3.1.0/one2one_cv3.1.0.0/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.991116           11.8647    int8      
/model.20/act/Mul                                   BPU  id(0)     HzLut                      0.965896           9.32107    int8      
/model.23/one2one_cv2.1/one2one_cv2.1.0/act/Mul     BPU  id(0)     HzLut                      0.982109           14.0359    int8      
...cv3.1/one2one_cv3.1.0/one2one_cv3.1.0.0/act/Mul  BPU  id(0)     HzLut                      0.991809           22.1297    int8      
/model.21/Concat                                    BPU  id(0)     Concat                     0.981236           11.1868    int8      
/model.23/one2one_cv2.1/one2one_cv2.1.1/conv/Conv   BPU  id(0)     HzSQuantizedConv           0.968720           14.0359    int8      
...3.1/one2one_cv3.1.0/one2one_cv3.1.0.1/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.952954           21.8763    int8      
/model.22/cv1/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.993625           11.1868    int8      
/model.23/one2one_cv2.1/one2one_cv2.1.1/act/Mul     BPU  id(0)     HzLut                      0.968222           9.74742    int8      
...cv3.1/one2one_cv3.1.0/one2one_cv3.1.0.1/act/Mul  BPU  id(0)     HzLut                      0.957260           17.717     int8      
/model.22/cv1/act/Mul                               BPU  id(0)     HzLut                      0.992123           10.6386    int8      
/model.23/one2one_cv2.1/one2one_cv2.1.2/Conv        BPU  id(0)     HzSQuantizedConv           0.988420           3.93149    int8      
...3.1/one2one_cv3.1.1/one2one_cv3.1.1.0/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.969618           17.693     int8      
/model.22/Split                                     BPU  id(0)     Split                      0.992194           10.6337    int8      
/model.22/m.0/m.0.0/cv1/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.995042           10.6337    int8      
...cv3.1/one2one_cv3.1.1/one2one_cv3.1.1.0/act/Mul  BPU  id(0)     HzLut                      0.969315           44.1044    int8      
...3.1/one2one_cv3.1.1/one2one_cv3.1.1.1/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.933211           36.2543    int8      
/model.22/m.0/m.0.0/cv1/act/Mul                     BPU  id(0)     HzLut                      0.989464           7.40366    int8      
/model.22/m.0/m.0.0/cv2/conv/Conv                   BPU  id(0)     HzSQuantizedConv           0.993004           6.04107    int8      
...cv3.1/one2one_cv3.1.1/one2one_cv3.1.1.1/act/Mul  BPU  id(0)     HzLut                      0.923929           75.141     int8      
/model.23/one2one_cv3.1/one2one_cv3.1.2/Conv        BPU  id(0)     HzSQuantizedConv           0.959172           72.0246    int8      
/model.22/m.0/m.0.0/cv2/act/Mul                     BPU  id(0)     HzLut                      0.992261           9.19344    int8      
/model.22/m.0/m.0.0/Add                             BPU  id(0)     HzSElementwiseAdd          0.992202           10.6337    int8      
/model.22/m.0/m.0.1/attn/qkv/conv/Conv              BPU  id(0)     HzSQuantizedConv           0.995455           9.45108    int8      
/model.22/m.0/m.0.1/attn/Reshape                    BPU  id(0)     Reshape                    0.995455           8.50838    int8      
/model.22/m.0/m.0.1/attn/Split                      BPU  id(0)     Split                      0.995945           8.50838    int8      
/model.22/m.0/m.0.1/attn/Mul                        BPU  id(0)     HzSElementwiseMul          0.995912           8.50838    int8      
/model.22/m.0/m.0.1/attn/Reshape_2                  BPU  id(0)     Reshape                    0.996680           8.50838    int8      
/model.22/m.0/m.0.1/attn/Transpose                  BPU  id(0)     Transpose                  0.995913           1.42247    int8      
/model.22/m.0/m.0.1/attn/MatMul                     BPU  id(0)     HzSQuantizedMatmul         0.987868           1.42247    int8      
...1/attn/Softmax_reducemax_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzQuantizedReduceMax       0.995980           96.9579    int16     
...0/m.0.1/attn/Softmax_sub_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzSElementwiseSub          0.878041           96.9579    int16     
...0/m.0.1/attn/Softmax_exp_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzLut2Layer                0.896811           11.0903    int16     
...1/attn/Softmax_reducesum_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzSQuantizedReduceSum      0.881342           1.0        int16     
.../attn/Softmax_reciprocal_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzLut2Layer                0.967451           104.844    int16     
...0/m.0.1/attn/Softmax_mul_FROM_QUANTIZED_SOFTMAX  BPU  id(0)     HzSElementwiseMul          0.920856           1.0        int16     
/model.22/m.0/m.0.1/attn/Transpose_1                BPU  id(0)     Transpose                  0.920856           1.0        int8      
/model.22/m.0/m.0.1/attn/MatMul_1                   BPU  id(0)     HzSQuantizedMatmul         0.994218           8.50838    int8      
/model.22/m.0/m.0.1/attn/Reshape_1                  BPU  id(0)     Reshape                    0.994218           8.28321    int8      
/model.22/m.0/m.0.1/attn/pe/conv/Conv               BPU  id(0)     HzSQuantizedConv           0.998944           8.50838    int8      
/model.22/m.0/m.0.1/attn/proj/conv/Conv             BPU  id(0)     HzSQuantizedConv           0.999894           13.9428    int8      
/model.22/m.0/m.0.1/ffn/ffn.0/conv/Conv             BPU  id(0)     HzSQuantizedConv           0.994360           4.47673    int8      
/model.22/m.0/m.0.1/ffn/ffn.0/act/Mul               BPU  id(0)     HzLut                      0.983103           11.1294    int8      
/model.22/m.0/m.0.1/ffn/ffn.1/conv/Conv             BPU  id(0)     HzSQuantizedConv           0.999855           7.66488    int8      
/model.22/Split_output_0_calibrated_Requantize      BPU  id(0)     HzRequantize               --                 --         int8      
/model.22/Split_output_1_calibrated_Requantize      BPU  id(0)     HzRequantize               --                 --         int8      
/model.22/Concat                                    BPU  id(0)     Concat                     0.999550           10.6337    int8      
/model.22/cv2/conv/Conv                             BPU  id(0)     HzSQuantizedConv           0.995982           31.4349    int8      
/model.22/cv2/act/Mul                               BPU  id(0)     HzLut                      0.995175           13.6214    int8      
/model.23/one2one_cv2.2/one2one_cv2.2.0/conv/Conv   BPU  id(0)     HzSQuantizedConv           0.998250           10.7788    int8      
...3.2/one2one_cv3.2.0/one2one_cv3.2.0.0/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.997529           10.7788    int8      
/model.23/one2one_cv2.2/one2one_cv2.2.0/act/Mul     BPU  id(0)     HzLut                      0.998403           8.09273    int8      
...cv3.2/one2one_cv3.2.0/one2one_cv3.2.0.0/act/Mul  BPU  id(0)     HzLut                      0.997583           12.7006    int8      
/model.23/one2one_cv2.2/one2one_cv2.2.1/conv/Conv   BPU  id(0)     HzSQuantizedConv           0.994550           8.04728    int8      
...3.2/one2one_cv3.2.0/one2one_cv3.2.0.1/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.995058           12.6614    int8      
/model.23/one2one_cv2.2/one2one_cv2.2.1/act/Mul     BPU  id(0)     HzLut                      0.993851           6.56473    int8      
...cv3.2/one2one_cv3.2.0/one2one_cv3.2.0.1/act/Mul  BPU  id(0)     HzLut                      0.994712           8.38742    int8      
/model.23/one2one_cv2.2/one2one_cv2.2.2/Conv        BPU  id(0)     HzSQuantizedConv           0.998870           5.93953    int8      
...3.2/one2one_cv3.2.1/one2one_cv3.2.1.0/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.998412           8.37452    int8      
...cv3.2/one2one_cv3.2.1/one2one_cv3.2.1.0/act/Mul  BPU  id(0)     HzLut                      0.998161           10.658     int8      
...3.2/one2one_cv3.2.1/one2one_cv3.2.1.1/conv/Conv  BPU  id(0)     HzSQuantizedConv           0.997855           9.84848    int8      
...cv3.2/one2one_cv3.2.1/one2one_cv3.2.1.1/act/Mul  BPU  id(0)     HzLut                      0.998135           13.6558    int8      
/model.23/one2one_cv3.2/one2one_cv3.2.2/Conv        BPU  id(0)     HzSQuantizedConv           0.999841           13.4876    int8
2026-10-26 10:14:10,705 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  
-----------------------------------------------------------------------------
output0     0.991145           0.209187     0.001906     3.802984            
output1     0.995471           2.425760     0.017688     38.828159           
output2     0.988420           0.323472     0.006308     5.715887            
output3     0.959172           4.288767     0.101424     139.803467          
output4     0.998870           0.133608     0.004822     2.200786            
output5     0.999841           0.408082     0.012171     2.398066
2026-10-26 10:14:10,710 file: model_builder.py func: model_builder line No: 38 End to Horizon NN Model Convert.
2026-10-26 10:14:10,718 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 601 start convert to *.bin file....
2026-10-26 10:14:10,735 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4326 ONNX model output num : 6
2026-10-26 10:14:10,737 file: layout_util.py func: layout_util line No: 15 set_featuremap_layout start
2026-10-26 10:14:10,737 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/yolo_new_model_1026/onnx_model/best_1026.onnx', 'march': 'bayes-e', 'layer_out_dump': False, 'log_level': 'DEBUG', 'working_dir': '/open_explorer/yolo_new_model_1026/hb_mapper_output', 'model_prefix': 'yolo26_detect_bayese_640x384_nv12', 'node_info': {'/model.10/m/m.0/attn/Softmax': {'ON': 'BPU', 'InputType': 'int16', 'OutputType': 'int16'}, '/model.22/m.0/m.0.1/attn/Softmax': {'ON': 'BPU', 'InputType': 'int16', 'OutputType': 'int16'}}, 'input_names': ['images'], 'input_type_rt': ['nv12'], 'input_space_and_range': ['regular'], 'input_type_train': ['rgb'], 'input_layout_rt': [''], 'input_layout_train': ['NCHW'], 'norm_type': ['data_scale'], 'scale_value': ['0.003921568627451,'], 'mean_value': [''], 'input_shape': ['2x3x384x640'], 'input_batch': [], 'cal_dir': ['/open_explorer/yolo_new_model_1026/calibration_data_f32_255'], 'cal_data_type': ['float32'], 'preprocess_on': False, 'calibration_type': 'default', 'per_channel': 'False', 'hbdk_params': {'hbdk_pass_through_params': '--O3 --core-num 1 --fast ', 'input-source': {'images': 'pyramid', '_default_value': 'ddr'}}, 'debug': False, 'compile_mode': 'latency'}
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4183 ############# model deps info #############
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4184 hb_mapper version   : 1.24.3
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4187 hbdk version        : 3.49.15
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4189 hbdk runtime version: 3.15.55.0
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4192 horizon_nn version  : 1.1.0
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4196 ############# model_parameters info #############
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4202 onnx_model          : /open_explorer/yolo_new_model_1026/onnx_model/best_1026.onnx
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4203 BPU march           : bayes-e
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4204 layer_out_dump      : False
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4205 log_level           : DEBUG
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4206 working dir         : /open_explorer/yolo_new_model_1026/hb_mapper_output
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4207 output_model_file_prefix: yolo26_detect_bayese_640x384_nv12
2026-10-26 10:14:10,737 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4226 node info  : {'/model.10/m/m.0/attn/Softmax': {'ON': 'BPU', 'InputType': 'int16', 'OutputType': 'int16'}, '/model.22/m.0/m.0.1/attn/Softmax': {'ON': 'BPU', 'InputType': 'int16', 'OutputType': 'int16'}}
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4228 ############# input_parameters info #############
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4246 ------------------------------------------
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4248 ---------input info : images ---------
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4249 input_name          : images
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4250 input_type_rt       : nv12
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4252 input_space&range   : regular
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4254 input_layout_rt     : None
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4255 input_type_train    : rgb
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4256 input_layout_train  : NCHW
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4257 norm_type           : data_scale
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4258 input_shape         : 2x3x384x640
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4266 scale_value         : 0.003921568627451,
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4268 cal_data_dir        : /open_explorer/yolo_new_model_1026/calibration_data_f32_255
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4270 cal_data_type       : float32
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4271 ---------input info : images end -------
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4272 ------------------------------------------
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4274 ############# calibration_parameters info #############
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4275 preprocess_on       : False
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4276 calibration_type:   : default
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4284 per_channel         : False
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4303 ############# compiler_parameters info #############
2026-10-26 10:14:10,738 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4305 debug               : False
2026-10-26 10:14:10,739 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4307 compile_mode        : latency
2026-10-26 10:14:10,739 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4310 hbdk_pass_through_params: --O3 --core-num 1 --fast
2026-10-26 10:14:10,739 file: onnx2horizonrt.py func: onnx2horizonrt line No: 4310 input-source        : {'images': 'pyramid', '_default_value': 'ddr'}
2026-10-26 10:14:10,740 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 783 Convert to runtime bin file successfully!
2026-10-26 10:14:10,741 file: hb_mapper_makertbin.py func: hb_mapper_makertbin line No: 784 End Model Convert
2

配置能转通，输出余弦 0.96~0.999，说明量化没有崩，属于"标定数据/预处理不一致 + 个别敏感算子"造成的类别级漏检，先按一致性→标定参数→敏感算子三步查，官方流程见 PTQ 模型精度调优。

1）先查一致性（最可能的原因）

你贴的 yaml 和这段日志不是同一次运行：yaml 里 working_dir: ../hb_mapper_output_3、cal_data_dir: calibration_data_bgr_f32_255_generated，而日志里是 hb_mapper_output 和 calibration_data_f32_255。先确认最终用的是哪份标定数据。

目录名写着 bgr，但配置里 input_type_train: 'rgb'。标定二进制会被原样喂给模型，通道顺序反了会让 conv 的量化阈值按错误分布算，表现就是余弦看着还行、某些类别边界被压掉。官方把 BGR/RGB 顺序、mean/scale 是否重复施加列为 yaml 头号误区。需要逐项核对：

  • 标定数据顺序与 input_type_train 一致，数值为 0~255、float32、CHW；
  • resize/letterbox 方式与训练一致（YOLO 训练多是 letterbox，标定时直接 resize 会改分布，直接用 transformer.py 默认的 opencv linear 不一定对齐）；
  • 数量百张左右，必须包含漏检那几类的样本，避免纯背景图、过曝图。

2）确认是不是量化的问题

用同一批图分别跑 *_original_float_model.onnx 和原 ONNX：若两者结果对不上，问题在数据处理而不在量化，可先不动量化参数。工具给了 hb_mapper infer 做这种比对。

3）标定与算子参数

  • 日志里 per_channel: False，建议打开 per_channel: True；
  • calibration_type 手动试 mix，不行再退 kl，并看日志是否出现 Perchannel quantization is enabled / Asymmetric quantization is enabled，或自动搜索选了哪种（Select kl method.）；
  • 敏感节点：第二个 C2PSA 的 Softmax 分解链余弦已掉到 0.878/0.897/0.881，紧随的 Transpose_1 0.9208，output3（one2one_cv3.1）0.959 最低。int16 的作用范围包含前后上下文，若调 int16 没改善，可把尾部 1~2 个算子回退 CPU 试，不要大面积回退，参数写法见 hb_mapper makertbin（int16 配置说明）。

4）两个顺带要确认的点

  • input_shape: '2x3x384x640' 是 batch=2，板上单帧推理通常是 1，确认是否有意为之；
  • 模型有 6 个输出（含 one2one_cv2/one2one_cv3 系列），即 one-to-one 分支被保留了。板端后处理若按单分支解析，也会表现成"某几类检不出"，先确认消费的是哪几个 output。

以上都不改善时，可用官方精度 debug 工具做节点级误差定位，再考虑 QAT。

方便的话补充两点：漏检是「这几类完全没有输出」还是「分数普遍偏低」？标定数据是 BGR 还是 RGB、共多少张？

3

先别继续改 Softmax，日志里已经是 int16。先对齐标定数据：实际读取的 calibration_data_f32_255 和贴出的 YAML 目录不同；按当前配置，数据应为 RGB、NCHW、float32、0～255，缩放补边与原 ONNX 验证一致，不要提前再除255。选百张左右、覆盖漏检类别的图片重新生成，并确认新日志用了这个目录。

对齐后用同一批图、同一后处理比较浮点和量化结果；若只有量化后漏检，再把 calibration_type 改为 mix 对照各类召回率。日志里的余弦只是首张标定图的结果，不能代表整个数据集精度。

参考：https://developer.d-robotics.cc/oe_x5_doc/cn/oe_mapper/source/tune_content/accuracy_tune.html

4

建议 直接参考Model Zoo的方案来量化

RDK Model Zoo 精度 性能可供参考