用户您好，请详细描述您所遇到的问题。
1.硬件获取渠道：
2.当前系统镜像版本：
3.当前天工开物版本：
4.问题定位：
5.开发的demo/案例：
6.需要提供的解决方案：
使用OE1.1.37，ddk/samples/vdsp_rpc_sample路径下提供的Softmax自定义算子示例上板运行正常，示例中使用的模型是mobilenet-v1。
当前我手中没有DSP工具链，想替换模型，然后测试下DSP侧Softmax算子能不能用。
将模型替换为Resnet模型，运行时返回错误码 -6000002，查了下说是模型无效，但是这个模型我在CPU运行都是OK的。
当前已经没有思路就怎么搞了~，哪位朋友有时间麻烦帮忙看下~
您先用1.1.37OE包提供的hrt_model_exec工具，在板端对你的resnet模型跑一下perf看能不能成功出结果呢？
hrt_model_exec位置：ddk/package/board/hrt_tools/bin/hrt_model_exec
参考指令：./hrt_model_exec perf --model_file resnet.bin
如果能正常打印FPS和latency等信息，说明模型没问题。
你好~，多谢回复
模型我已经在板子上跑起来了，结果已经验证是正确的
只是Softmax运行在CPU上，导致FPS很低，所以想用DSP
一会回办公室，我再用这个工具试下
这个是在x86上跑出来的结果，J5没在手上，帧率会比这个高很多
./hrt_model_exec perf --core_id 0 --frame_count 1 --model_file /home/shiyucun/workspace/zero_one/src/perception/lidars_fpn_resnet18/model_inference_cc/data/fpn_resnet.bin
I0000 00:00:00.000000 179 vlog_is_on.cc:197] RAW: Set VLOG level for “*” to 3
core[0] open!
core[1] open!
[HBRT] set log level as 0. version = 3.15.8.0
[DNN] Runtime version = 1.14.4_(3.15.8 HBRT)
[A][DNN][initializer.cpp:37](1686279622307) Direct mode
[A][DNN][packed_model.cpp:208](1686279622389) [HorizonRT] The model builder version = 1.13.5
Load model to DDR cost 127.18ms.
I0609 11:00:22.434839 179 function_util.cpp:319] get model handle success
I0609 11:00:22.434887 179 function_util.cpp:607] get model input count success
I0609 11:00:22.435356 179 function_util.cpp:638] prepare input tensor success!
I0609 11:00:22.435359 179 function_util.cpp:644] get model output count success
WARNING: hb_bpu_core_estimate_loading is ignored in pseudo_firmware
WARNING: cnn_core_fc_avl_cap is ignored in pseudo_firmware
WARNING: bpu_mem_cache_flush is ignored in pseudo_firmware
Frame count: 1, Thread Average: 70910.718750 ms, thread max latency: 70910.718750 ms, thread min latency: 70910.718750 ms, FPS: 0.014102
Running condition:
Thread number is: 1
Frame count is: 1
Program run time: 70911.344000 ms
Perf result:
Frame totally latency is: 70910.718750 ms
Average latency is: 70910.718750 ms
Frame rate is: 0.014102 FPS
把Softmax部署到BPU：
run_on_bpu: /Softmax;/Softmax_1;/Softmax_2;/Softmax_3;
run_on_bpu: Softmax;Softmax_1;Softmax_2;Softmax_3;
run_on_bpu: Softmax;
run_on_bpu: {/Softmax, /Softmax_1, /Softmax_2, /Softmax_3}
run_on_bpu: {Softmax, Softmax_1, Softmax_2, Softmax_3}
run_on_bpu: {Softmax}
转换使用的配置文件，使用了上面的参数，转换后Softmax依然跑在CPU：
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs# hb_mapper makertbin --config /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/config.yaml --model-type onnx
2023-06-09 14:27:45,249 INFO log will be stored in /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs/hb_mapper_makertbin.log
2023-06-09 14:27:45,249 INFO Start hb_mapper…
2023-06-09 14:27:45,249 INFO hbdk version 3.41.5
2023-06-09 14:27:45,249 INFO horizon_nn version 0.15.5
2023-06-09 14:27:45,249 INFO hb_mapper version 1.13.5
2023-06-09 14:27:45,249 INFO Start Model Convert…
2023-06-09 14:27:45,252 INFO Using onnx model file: /mnt/sda/shiyucun/pcp_j5/larry_pcp/src/super_fast_object_detection/src/sfa/model2onnx/fpn_resnet0420.onnx
2023-06-09 14:27:45,282 INFO Model has 1 inputs according to model file
2023-06-09 14:27:45,282 INFO 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.
2023-06-09 14:27:45,282 INFO custom_op does not exist, skipped
2023-06-09 14:27:45,282 WARNING Input node data’s input_source not set, it will be set to ddr by default
2023-06-09 14:27:45,284 INFO *******************************************
2023-06-09 14:27:45,284 INFO First calibration picture name: 1589168947_0.pcd.bgr
2023-06-09 14:27:45,284 INFO First calibration picture md5:
6f5602903a2881239cdec090376ae782 /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/calibration_data_bgr_f32/1589168947_0.pcd.bgr
2023-06-09 14:27:45,290 INFO *******************************************
2023-06-09 14:27:45,513 INFO [Fri Jun 9 14:27:45 2023] Start to Horizon NN Model Convert.
2023-06-09 14:27:45,513 INFO Parsing the input parameter:{‘data’: {‘input_shape’: [1, 3, 608, 608], ‘input_batch’: 1, ‘expected_input_type’: ‘BGR_128’, ‘original_input_type’: ‘BGR’, ‘original_input_layout’: ‘NCHW’}}
2023-06-09 14:27:45,513 INFO Parsing the calibration parameter
2023-06-09 14:27:45,513 INFO There are 1 nodes designated to run on the bpu: [‘{ Softmax : None}’].
2023-06-09 14:27:45,513 INFO Parsing the hbdk parameter:{‘hbdk_pass_through_params’: '–O3 --core-num 1 --fast ', ‘input-source’: {‘data’: ‘ddr’, ‘_default_value’: ‘ddr’}}
2023-06-09 14:27:45,513 INFO HorizonNN version: 0.15.5
2023-06-09 14:27:45,513 INFO HBDK version: 3.41.5
2023-06-09 14:27:45,514 INFO [Fri Jun 9 14:27:45 2023] Start to parse the onnx model.
2023-06-09 14:27:45,538 INFO Input ONNX model infomation:
ONNX IR version: 6
Opset version: [11]
Producer: pytorch2.0.1
Domain: none
Input name: data, [1, 3, 608, 608]
Output name: output, [1, 3, 152, 152]
Output name: 342, [1, 2, 152, 152]
Output name: 368, [1, 2, 152, 152]
Output name: 394, [1, 1, 152, 152]
Output name: 420, [1, 3, 152, 152]
2023-06-09 14:27:45,717 INFO [Fri Jun 9 14:27:45 2023] End to parse the onnx model.
2023-06-09 14:27:45,718 INFO Model input names parsed from model: [‘data’]
2023-06-09 14:27:45,718 INFO Create a preprocessing operator for input_name data with means=None, std=None, original_input_layout=NCHW, color convert from ‘BGR’ to ‘BGR’.
2023-06-09 14:27:45,994 INFO Saving the original float model: fpn_resnet_original_float_model.onnx.
2023-06-09 14:27:45,994 INFO [Fri Jun 9 14:27:45 2023] Start to optimize the model.
2023-06-09 14:27:46,294 INFO [Fri Jun 9 14:27:46 2023] End to optimize the model.
2023-06-09 14:27:46,489 INFO Saving the optimized model: fpn_resnet_optimized_float_model.onnx.
2023-06-09 14:27:46,489 INFO [Fri Jun 9 14:27:46 2023] Start to calibrate the model.
2023-06-09 14:27:46,490 INFO There are 102 samples in the calibration data set.
2023-06-09 14:27:46,696 INFO Run calibration model with kl method.
kl calibration in progress: 0%| | 0/13 [00:00<?, ?it/s]2023-06-09 14:27:49.684791867 [E:onnxruntime:, sequential_executor.cc:183 Execute] Non-zero status code returned while running Reshape node. Name:‘/Unsqueeze_14’ Status Message: /home/jenkins/agent/workspace/model_convert/onnxruntime/onnxruntime/core/providers/cpu/tensor/reshape_helper.h:43 onnxruntime::ReshapeHelper::ReshapeHelper(const onnxruntime::TensorShape&, std::vector&) gsl::narrow_cast<int64_t>(input_shape.Size()) == size was false. The input tensor cannot be reshaped to the requested shape. Input shape:{8,3,152,152}, requested shape:{1,3,152,152,1}
kl calibration in progress: 0%| | 0/13 [00:02<?, ?it/s]
2023-06-09 14:27:49,685 INFO Above info is caused by batch mode infer and can be ignored
2023-06-09 14:27:49,685 INFO Reset batch_size=1 and execute calibration again…
kl calibration in progress: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 102/102 [01:51<00:00, 1.09s/it]
2023-06-09 14:29:40,980 INFO [Fri Jun 9 14:29:40 2023] End to calibrate the model.
2023-06-09 14:29:40,981 INFO [Fri Jun 9 14:29:40 2023] Start to quantize the model.
2023-06-09 14:29:50,457 INFO [Fri Jun 9 14:29:50 2023] End to quantize the model.
2023-06-09 14:29:51,016 INFO Saving the quantized model: fpn_resnet_quantized_model.onnx.
2023-06-09 14:29:51,840 INFO [Fri Jun 9 14:29:51 2023] Start to compile the model with march bayes.
2023-06-09 14:29:52,208 INFO Compile submodel: torch_jit_subgraph_0
2023-06-09 14:29:52,768 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr’]
2023-06-09 14:29:52,781 WARNING Can not find the scale for node HZ_PREPROCESS_FOR_data_NCHW2NHWC_LayoutConvert_Input0
2023-06-09 14:41:54,504 INFO Compile submodel: torch_jit_subgraph_1
2023-06-09 14:41:54,542 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 14:41:54,717 INFO Compile submodel: torch_jit_subgraph_2
2023-06-09 14:41:54,754 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 14:41:54,872 INFO Compile submodel: torch_jit_subgraph_3
2023-06-09 14:41:54,910 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 14:41:55,028 INFO Compile submodel: torch_jit_subgraph_4
2023-06-09 14:41:55,067 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 14:41:55,175 INFO Compile submodel: torch_jit_subgraph_5
2023-06-09 14:41:55,211 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 14:41:55,730 INFO [Fri Jun 9 14:41:55 2023] End to compile the model with march bayes.
2023-06-09 14:41:55,731 INFO The converted model node information:
==============================================================================================================================================
Node ON Subgraph Type Cosine Similarity Threshold
-----------------------------------------------------------------------------------------------------------------------------------------------
HZ_PREPROCESS_FOR_data BPU id(0) HzSQuantizedPreprocess 0.999902 127.000000
/conv1/Conv BPU id(0) HzSQuantizedConv 0.995028 253.856461
/maxpool/MaxPool BPU id(0) HzQuantizedMaxPool 0.986078 1.761778
/layer1/layer1.0/conv1/Conv BPU id(0) HzSQuantizedConv 0.944165 1.761778
/layer1/layer1.0/conv2/Conv BPU id(0) HzSQuantizedConv 0.964870 1.422739
/layer1/layer1.1/conv1/Conv BPU id(0) HzSQuantizedConv 0.912611 2.758681
/layer1/layer1.1/conv2/Conv BPU id(0) HzSQuantizedConv 0.958991 1.346770
/layer2/layer2.0/conv1/Conv BPU id(0) HzSQuantizedConv 0.935197 3.533995
/layer2/layer2.0/conv2/Conv BPU id(0) HzSQuantizedConv 0.904247 1.255401
/layer2/layer2.0/downsample/downsample.0/Conv BPU id(0) HzSQuantizedConv 0.938371 3.533995
/layer2/layer2.1/conv1/Conv BPU id(0) HzSQuantizedConv 0.888894 2.211774
/layer2/layer2.1/conv2/Conv BPU id(0) HzSQuantizedConv 0.935048 1.245804
/layer3/layer3.0/conv1/Conv BPU id(0) HzSQuantizedConv 0.907809 2.861637
/layer3/layer3.0/conv2/Conv BPU id(0) HzSQuantizedConv 0.909788 1.452334
/layer3/layer3.0/downsample/downsample.0/Conv BPU id(0) HzSQuantizedConv 0.924617 2.861637
/layer3/layer3.1/conv1/Conv BPU id(0) HzSQuantizedConv 0.883150 1.643583
/layer3/layer3.1/conv2/Conv BPU id(0) HzSQuantizedConv 0.910612 0.954917
/layer4/layer4.0/conv1/Conv BPU id(0) HzSQuantizedConv 0.821044 2.078356
/layer4/layer4.0/conv2/Conv BPU id(0) HzSQuantizedConv 0.861210 1.876143
/layer4/layer4.0/downsample/downsample.0/Conv BPU id(0) HzSQuantizedConv 0.782653 2.078356
/layer4/layer4.1/conv1/Conv BPU id(0) HzSQuantizedConv 0.667947 2.044354
/layer4/layer4.1/conv2/Conv BPU id(0) HzSQuantizedConv 0.774661 2.900229
/Resize BPU id(0) HzQuantizedRoiResize 0.802121 16.436541
/layer3/layer3.1/relu_1/Relu_output_0_Requantize BPU id(0) HzRequantize
/Concat BPU id(0) Concat 0.801647 16.436541
/conv_up_level1/Conv BPU id(0) HzSQuantizedConv 0.810718 16.436541
/Resize_1 BPU id(0) HzQuantizedRoiResize 0.816626 33.241734
/Resize_1_output_0_Requantize BPU id(0) HzRequantize
/layer2/layer2.1/relu_1/Relu_output_0_Requantize BPU id(0) HzRequantize
/Concat_1 BPU id(0) Concat 0.816599 33.241734
/conv_up_level2/Conv BPU id(0) HzSQuantizedConv 0.836094 30.123194
/Resize_2 BPU id(0) HzQuantizedRoiResize 0.837536 44.009842
/Resize_2_output_0_Requantize BPU id(0) HzRequantize
/layer1/layer1.1/relu_1/Relu_output_0_Requantize BPU id(0) HzRequantize
/Concat_2 BPU id(0) Concat 0.838752 44.009842
/conv_up_level3/Conv BPU id(0) HzSQuantizedConv 0.870276 44.154037
/fpn0_hm_cen/fpn0_hm_cen.0/Conv BPU id(0) HzSQuantizedConv 0.924563 33.241734
/fpn0_hm_cen/fpn0_hm_cen.2/Conv BPU id(0) HzSQuantizedConv 0.943938 313.639984
/Resize_3 BPU id(0) HzQuantizedResizeUpsample 0.943938 312.455658
/fpn1_hm_cen/fpn1_hm_cen.0/Conv BPU id(0) HzSQuantizedConv 0.874081 44.009842
/fpn1_hm_cen/fpn1_hm_cen.2/Conv BPU id(0) HzSQuantizedConv 0.861351 500.141083
/fpn2_hm_cen/fpn2_hm_cen.0/Conv BPU id(0) HzSQuantizedConv 0.892319 38.183331
/fpn2_hm_cen/fpn2_hm_cen.2/Conv BPU id(0) HzSQuantizedConv 0.917691 220.486969
/Unsqueeze BPU id(0) Reshape
/Unsqueeze_1 BPU id(0) Reshape
/Unsqueeze_2 BPU id(0) Reshape
/Concat_4 BPU id(0) Concat 0.892625 312.455658
/Softmax CPU – Softmax 0.683802 312.455658
/Mul BPU id(1) HzSElementwiseMul 0.637066 312.455658
/ReduceSum CPU – ReduceSum 0.932151 91.498909
/ReduceSum_reshape CPU – Reshape
/fpn0_cen_offset/fpn0_cen_offset.0/Conv BPU id(0) HzSQuantizedConv 0.779345 33.241734
/fpn0_cen_offset/fpn0_cen_offset.2/Conv BPU id(0) HzSQuantizedConv 0.867806 267.370972
/Resize_4 BPU id(0) HzQuantizedResizeUpsample 0.867808 299.037598
/fpn1_cen_offset/fpn1_cen_offset.0/Conv BPU id(0) HzSQuantizedConv 0.782416 44.009842
/fpn1_cen_offset/fpn1_cen_offset.2/Conv BPU id(0) HzSQuantizedConv 0.665010 233.543671
/fpn2_cen_offset/fpn2_cen_offset.0/Conv BPU id(0) HzSQuantizedConv 0.765405 38.183331
/fpn2_cen_offset/fpn2_cen_offset.2/Conv BPU id(0) HzSQuantizedConv 0.604727 102.658096
/Unsqueeze_3 BPU id(0) Reshape
/Unsqueeze_4 BPU id(0) Reshape
/Unsqueeze_5 BPU id(0) Reshape
/Concat_6 BPU id(0) Concat 0.863411 299.037598
/Softmax_1 CPU – Softmax 0.865288 299.037598
/Mul_1 BPU id(2) HzSElementwiseMul 0.548331 299.037598
/ReduceSum_1 CPU – ReduceSum 0.621538 9.692646
/ReduceSum_1_reshape CPU – Reshape
/fpn0_direction/fpn0_direction.0/Conv BPU id(0) HzSQuantizedConv 0.755548 33.241734
/fpn0_direction/fpn0_direction.2/Conv BPU id(0) HzSQuantizedConv 0.717000 425.373810
/Resize_5 BPU id(0) HzQuantizedResizeUpsample 0.716995 73.949379
/fpn1_direction/fpn1_direction.0/Conv BPU id(0) HzSQuantizedConv 0.705615 44.009842
/fpn1_direction/fpn1_direction.2/Conv BPU id(0) HzSQuantizedConv 0.450235 41.534039
/fpn2_direction/fpn2_direction.0/Conv BPU id(0) HzSQuantizedConv 0.606136 38.183331
/fpn2_direction/fpn2_direction.2/Conv BPU id(0) HzSQuantizedConv 0.606858 138.430069
/Unsqueeze_6 BPU id(0) Reshape
/Unsqueeze_7 BPU id(0) Reshape
/Unsqueeze_8 BPU id(0) Reshape
/Concat_8 BPU id(0) Concat 0.606557 73.949379
/Softmax_2 CPU – Softmax 0.850093 73.949379
/Mul_2 BPU id(3) HzSElementwiseMul 0.620950 73.949379
/ReduceSum_2 CPU – ReduceSum 0.513161 1.124602
/ReduceSum_2_reshape CPU – Reshape
/fpn0_z_coor/fpn0_z_coor.0/Conv BPU id(0) HzSQuantizedConv 0.844084 33.241734
/fpn0_z_coor/fpn0_z_coor.2/Conv BPU id(0) HzSQuantizedConv 0.762663 382.411621
/Resize_6 BPU id(0) HzQuantizedResizeUpsample 0.762663 79.672722
/Resize_6_NHWC2NCHW_LayoutConvert_Output0_reshape BPU id(0) Reshape
/fpn1_z_coor/fpn1_z_coor.0/Conv BPU id(0) HzSQuantizedConv 0.869190 44.009842
/fpn1_z_coor/fpn1_z_coor.2/Conv BPU id(0) HzSQuantizedConv 0.836509 84.870514
…2/Conv_NHWC2NCHW_LayoutConvert_Output0_reshape BPU id(0) Reshape
/fpn2_z_coor/fpn2_z_coor.0/Conv BPU id(0) HzSQuantizedConv 0.915549 38.183331
/fpn2_z_coor/fpn2_z_coor.2/Conv BPU id(0) HzSQuantizedConv 0.958527 53.304226
…2/Conv_NHWC2NCHW_LayoutConvert_Output0_reshape BPU id(0) Reshape
/Concat_10 BPU id(0) Concat 0.760884 79.672722
/Softmax_3 CPU – Softmax 0.953106 79.672722
/Mul_3 BPU id(4) HzSElementwiseMul 0.849502 79.672722
/ReduceSum_3 CPU – ReduceSum 0.933492 2.750437
/ReduceSum_3_reshape CPU – Reshape
/fpn0_dim/fpn0_dim.0/Conv BPU id(0) HzSQuantizedConv 0.905381 33.241734
/fpn0_dim/fpn0_dim.2/Conv BPU id(0) HzSQuantizedConv 0.959605 139.523239
/Resize_7 BPU id(0) HzQuantizedResizeUpsample 0.959608 61.583912
/fpn1_dim/fpn1_dim.0/Conv BPU id(0) HzSQuantizedConv 0.856017 44.009842
/fpn1_dim/fpn1_dim.2/Conv BPU id(0) HzSQuantizedConv 0.675916 107.564522
/fpn2_dim/fpn2_dim.0/Conv BPU id(0) HzSQuantizedConv 0.823923 38.183331
/fpn2_dim/fpn2_dim.2/Conv BPU id(0) HzSQuantizedConv 0.463424 109.336662
/Unsqueeze_12 BPU id(0) Reshape
/Unsqueeze_13 BPU id(0) Reshape
/Unsqueeze_14 BPU id(0) Reshape
/Concat_12 BPU id(0) Concat 0.734860 61.583912
/Softmax_4 CPU – Softmax 0.974800 61.583912
/Mul_4 BPU id(5) HzSElementwiseMul 0.934541 61.583912
/ReduceSum_4 CPU – ReduceSum 0.956766 27.699253
/ReduceSum_4_reshape CPU – Reshape
2023-06-09 14:41:55,732 INFO The quantify model output:
===============================================================================
Node Cosine Similarity L1 Distance L2 Distance Chebyshev Distance
-------------------------------------------------------------------------------
/ReduceSum 0.932151 6.658581 0.038718 72.361755
/ReduceSum_1 0.621538 1.135461 0.006949 20.349892
/ReduceSum_2 0.513161 0.414121 0.002889 2.062102
/ReduceSum_3 0.933492 0.284830 0.002482 2.326995
/ReduceSum_4 0.956766 0.805929 0.006046 19.764456
2023-06-09 14:41:55,732 INFO [Fri Jun 9 14:41:55 2023] End to Horizon NN Model Convert.
2023-06-09 14:41:55,846 WARNING node: { Softmax : None} does not exist, please double check your input
2023-06-09 14:41:55,846 INFO start convert to *.bin file…
2023-06-09 14:41:55,883 INFO ONNX model output num : 5
2023-06-09 14:41:55,888 INFO ############# model deps info #############
2023-06-09 14:41:55,888 INFO hb_mapper version : 1.13.5
2023-06-09 14:41:55,888 INFO hbdk version : 3.41.5
2023-06-09 14:41:55,888 INFO hbdk runtime version: 3.15.8.0
2023-06-09 14:41:55,888 INFO horizon_nn version : 0.15.5
2023-06-09 14:41:55,888 INFO ############# model_parameters info #############
2023-06-09 14:41:55,888 INFO onnx_model : /mnt/sda/shiyucun/pcp_j5/larry_pcp/src/super_fast_object_detection/src/sfa/model2onnx/fpn_resnet0420.onnx
2023-06-09 14:41:55,888 INFO BPU march : bayes
2023-06-09 14:41:55,888 INFO layer_out_dump : False
2023-06-09 14:41:55,888 INFO log_level : DEBUG
2023-06-09 14:41:55,888 INFO working dir : /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs
2023-06-09 14:41:55,888 INFO output_model_file_prefix: fpn_resnet
2023-06-09 14:41:55,888 INFO ############# input_parameters info #############
2023-06-09 14:41:55,888 INFO ------------------------------------------
2023-06-09 14:41:55,888 INFO ---------input info : data ---------
2023-06-09 14:41:55,888 INFO input_name : data
2023-06-09 14:41:55,888 INFO input_type_rt : bgr
2023-06-09 14:41:55,888 INFO input_space&range : regular
2023-06-09 14:41:55,888 INFO input_layout_rt : NCHW
2023-06-09 14:41:55,888 INFO input_type_train : bgr
2023-06-09 14:41:55,889 INFO input_layout_train : NCHW
2023-06-09 14:41:55,889 INFO norm_type : no_preprocess
2023-06-09 14:41:55,889 INFO input_shape : 1x3x608x608
2023-06-09 14:41:55,889 INFO input_batch : 1
2023-06-09 14:41:55,889 INFO cal_data_dir : /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/calibration_data_bgr_f32
2023-06-09 14:41:55,889 INFO cal_data_type : float32
2023-06-09 14:41:55,889 INFO ---------input info : data end -------
2023-06-09 14:41:55,889 INFO ------------------------------------------
2023-06-09 14:41:55,889 INFO ############# calibration_parameters info #############
2023-06-09 14:41:55,889 INFO preprocess_on : False
2023-06-09 14:41:55,889 INFO calibration_type: : kl
2023-06-09 14:41:55,889 INFO max_percentile : 1.0
2023-06-09 14:41:55,889 INFO run_on_bpu : { Softmax : None};
2023-06-09 14:41:55,889 INFO ############# compiler_parameters info #############
2023-06-09 14:41:55,889 INFO hbdk_pass_through_params: --O3 --core-num 1 --fast
2023-06-09 14:41:55,889 INFO input-source : {‘data’: ‘ddr’, ‘_default_value’: ‘ddr’}
2023-06-09 14:41:55,931 INFO Convert to runtime bin file sucessfully!
2023-06-09 14:41:55,932 INFO End Model Convert
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs# clear
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs# hb_mapper makertbin --config /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/config.yaml --model-type onnx
2023-06-09 14:46:21,017 INFO log will be stored in /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs/hb_mapper_makertbin.log
2023-06-09 14:46:21,018 INFO Start hb_mapper…
2023-06-09 14:46:21,018 INFO hbdk version 3.41.5
2023-06-09 14:46:21,018 INFO horizon_nn version 0.15.5
2023-06-09 14:46:21,018 INFO hb_mapper version 1.13.5
2023-06-09 14:46:21,018 INFO Start Model Convert…
2023-06-09 14:46:21,021 INFO Using onnx model file: /mnt/sda/shiyucun/pcp_j5/larry_pcp/src/super_fast_object_detection/src/sfa/model2onnx/fpn_resnet0420.onnx
2023-06-09 14:46:21,048 INFO Model has 1 inputs according to model file
2023-06-09 14:46:21,049 INFO 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.
2023-06-09 14:46:21,049 INFO custom_op does not exist, skipped
2023-06-09 14:46:21,049 WARNING Input node data’s input_source not set, it will be set to ddr by default
2023-06-09 14:46:21,051 INFO *******************************************
2023-06-09 14:46:21,051 INFO First calibration picture name: 1589168947_0.pcd.bgr
2023-06-09 14:46:21,051 INFO First calibration picture md5:
6f5602903a2881239cdec090376ae782 /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/calibration_data_bgr_f32/1589168947_0.pcd.bgr
2023-06-09 14:46:21,056 INFO *******************************************
2023-06-09 14:46:21,281 INFO [Fri Jun 9 14:46:21 2023] Start to Horizon NN Model Convert.
2023-06-09 14:46:21,281 INFO Parsing the input parameter:{‘data’: {‘input_shape’: [1, 3, 608, 608], ‘input_batch’: 1, ‘expected_input_type’: ‘BGR_128’, ‘original_input_type’: ‘BGR’, ‘original_input_layout’: ‘NCHW’}}
2023-06-09 14:46:21,281 INFO Parsing the calibration parameter
2023-06-09 14:46:21,281 INFO There are 1 nodes designated to run on the bpu: [‘Softmax’].
2023-06-09 14:46:21,281 INFO Parsing the hbdk parameter:{‘hbdk_pass_through_params’: '–O3 --core-num 1 --fast ', ‘input-source’: {‘data’: ‘ddr’, ‘_default_value’: ‘ddr’}}
2023-06-09 14:46:21,281 INFO HorizonNN version: 0.15.5
2023-06-09 14:46:21,281 INFO HBDK version: 3.41.5
2023-06-09 14:46:21,281 INFO [Fri Jun 9 14:46:21 2023] Start to parse the onnx model.
2023-06-09 14:46:21,307 INFO Input ONNX model infomation:
ONNX IR version: 6
Opset version: [11]
Producer: pytorch2.0.1
Domain: none
Input name: data, [1, 3, 608, 608]
Output name: output, [1, 3, 152, 152]
Output name: 342, [1, 2, 152, 152]
Output name: 368, [1, 2, 152, 152]
Output name: 394, [1, 1, 152, 152]
Output name: 420, [1, 3, 152, 152]
2023-06-09 14:46:21,488 INFO [Fri Jun 9 14:46:21 2023] End to parse the onnx model.
2023-06-09 14:46:21,488 INFO Model input names parsed from model: [‘data’]
2023-06-09 14:46:21,488 INFO Create a preprocessing operator for input_name data with means=None, std=None, original_input_layout=NCHW, color convert from ‘BGR’ to ‘BGR’.
2023-06-09 14:46:21,810 INFO Saving the original float model: fpn_resnet_original_float_model.onnx.
2023-06-09 14:46:21,810 INFO [Fri Jun 9 14:46:21 2023] Start to optimize the model.
2023-06-09 14:46:22,125 INFO [Fri Jun 9 14:46:22 2023] End to optimize the model.
2023-06-09 14:46:22,311 INFO Saving the optimized model: fpn_resnet_optimized_float_model.onnx.
2023-06-09 14:46:22,311 INFO [Fri Jun 9 14:46:22 2023] Start to calibrate the model.
2023-06-09 14:46:22,312 INFO There are 102 samples in the calibration data set.
2023-06-09 14:46:22,510 INFO Run calibration model with kl method.
kl calibration in progress: 0%| | 0/13 [00:00<?, ?it/s]2023-06-09 14:46:25.515300646 [E:onnxruntime:, sequential_executor.cc:183 Execute] Non-zero status code returned while running Reshape node. Name:‘/Unsqueeze_14’ Status Message: /home/jenkins/agent/workspace/model_convert/onnxruntime/onnxruntime/core/providers/cpu/tensor/reshape_helper.h:43 onnxruntime::ReshapeHelper::ReshapeHelper(const onnxruntime::TensorShape&, std::vector&) gsl::narrow_cast<int64_t>(input_shape.Size()) == size was false. The input tensor cannot be reshaped to the requested shape. Input shape:{8,3,152,152}, requested shape:{1,3,152,152,1}
kl calibration in progress: 0%| | 0/13 [00:02<?, ?it/s]
2023-06-09 14:46:25,515 INFO Above info is caused by batch mode infer and can be ignored
2023-06-09 14:46:25,515 INFO Reset batch_size=1 and execute calibration again…
kl calibration in progress: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 102/102 [01:51<00:00, 1.09s/it]
2023-06-09 14:48:16,860 INFO [Fri Jun 9 14:48:16 2023] End to calibrate the model.
2023-06-09 14:48:16,860 INFO [Fri Jun 9 14:48:16 2023] Start to quantize the model.
2023-06-09 14:48:26,257 INFO [Fri Jun 9 14:48:26 2023] End to quantize the model.
2023-06-09 14:48:26,817 INFO Saving the quantized model: fpn_resnet_quantized_model.onnx.
2023-06-09 14:48:27,615 INFO [Fri Jun 9 14:48:27 2023] Start to compile the model with march bayes.
2023-06-09 14:48:27,985 INFO Compile submodel: torch_jit_subgraph_0
2023-06-09 14:48:28,224 WARNING Can not find the scale for node HZ_PREPROCESS_FOR_data_NCHW2NHWC_LayoutConvert_Input0
2023-06-09 14:48:28,550 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr’]
2023-06-09 15:00:36,069 INFO Compile submodel: torch_jit_subgraph_1
2023-06-09 15:00:36,109 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 15:00:36,286 INFO Compile submodel: torch_jit_subgraph_2
2023-06-09 15:00:36,325 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 15:00:36,530 INFO Compile submodel: torch_jit_subgraph_3
2023-06-09 15:00:36,573 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 15:00:36,697 INFO Compile submodel: torch_jit_subgraph_4
2023-06-09 15:00:36,736 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 15:00:36,842 INFO Compile submodel: torch_jit_subgraph_5
2023-06-09 15:00:36,881 INFO hbdk-cc parameters:[‘–O3’, ‘–core-num’, ‘1’, ‘–fast’, ‘–input-layout’, ‘NHWC’, ‘–output-layout’, ‘NHWC’, ‘–input-source’, ‘ddr,ddr’]
2023-06-09 15:00:37,414 INFO [Fri Jun 9 15:00:37 2023] End to compile the model with march bayes.
2023-06-09 15:00:37,415 INFO The converted model node information:
==============================================================================================================================================
Node ON Subgraph Type Cosine Similarity Threshold
-----------------------------------------------------------------------------------------------------------------------------------------------
HZ_PREPROCESS_FOR_data BPU id(0) HzSQuantizedPreprocess 0.999902 127.000000
/conv1/Conv BPU id(0) HzSQuantizedConv 0.995028 253.856461
/maxpool/MaxPool BPU id(0) HzQuantizedMaxPool 0.986078 1.761778
/layer1/layer1.0/conv1/Conv BPU id(0) HzSQuantizedConv 0.944165 1.761778
/layer1/layer1.0/conv2/Conv BPU id(0) HzSQuantizedConv 0.964870 1.422739
/layer1/layer1.1/conv1/Conv BPU id(0) HzSQuantizedConv 0.912611 2.758681
/layer1/layer1.1/conv2/Conv BPU id(0) HzSQuantizedConv 0.958991 1.346770
/layer2/layer2.0/conv1/Conv BPU id(0) HzSQuantizedConv 0.935197 3.533995
/layer2/layer2.0/conv2/Conv BPU id(0) HzSQuantizedConv 0.904247 1.255401
/layer2/layer2.0/downsample/downsample.0/Conv BPU id(0) HzSQuantizedConv 0.938371 3.533995
/layer2/layer2.1/conv1/Conv BPU id(0) HzSQuantizedConv 0.888894 2.211774
/layer2/layer2.1/conv2/Conv BPU id(0) HzSQuantizedConv 0.935048 1.245804
/layer3/layer3.0/conv1/Conv BPU id(0) HzSQuantizedConv 0.907809 2.861637
/layer3/layer3.0/conv2/Conv BPU id(0) HzSQuantizedConv 0.909788 1.452334
/layer3/layer3.0/downsample/downsample.0/Conv BPU id(0) HzSQuantizedConv 0.924617 2.861637
/layer3/layer3.1/conv1/Conv BPU id(0) HzSQuantizedConv 0.883150 1.643583
/layer3/layer3.1/conv2/Conv BPU id(0) HzSQuantizedConv 0.910612 0.954917
/layer4/layer4.0/conv1/Conv BPU id(0) HzSQuantizedConv 0.821044 2.078356
/layer4/layer4.0/conv2/Conv BPU id(0) HzSQuantizedConv 0.861210 1.876143
/layer4/layer4.0/downsample/downsample.0/Conv BPU id(0) HzSQuantizedConv 0.782653 2.078356
/layer4/layer4.1/conv1/Conv BPU id(0) HzSQuantizedConv 0.667947 2.044354
/layer4/layer4.1/conv2/Conv BPU id(0) HzSQuantizedConv 0.774661 2.900229
/Resize BPU id(0) HzQuantizedRoiResize 0.802121 16.436541
/layer3/layer3.1/relu_1/Relu_output_0_Requantize BPU id(0) HzRequantize
/Concat BPU id(0) Concat 0.801647 16.436541
/conv_up_level1/Conv BPU id(0) HzSQuantizedConv 0.810718 16.436541
/Resize_1 BPU id(0) HzQuantizedRoiResize 0.816626 33.241734
/Resize_1_output_0_Requantize BPU id(0) HzRequantize
/layer2/layer2.1/relu_1/Relu_output_0_Requantize BPU id(0) HzRequantize
/Concat_1 BPU id(0) Concat 0.816599 33.241734
/conv_up_level2/Conv BPU id(0) HzSQuantizedConv 0.836094 30.123194
/Resize_2 BPU id(0) HzQuantizedRoiResize 0.837536 44.009842
/Resize_2_output_0_Requantize BPU id(0) HzRequantize
/layer1/layer1.1/relu_1/Relu_output_0_Requantize BPU id(0) HzRequantize
/Concat_2 BPU id(0) Concat 0.838752 44.009842
/conv_up_level3/Conv BPU id(0) HzSQuantizedConv 0.870276 44.154037
/fpn0_hm_cen/fpn0_hm_cen.0/Conv BPU id(0) HzSQuantizedConv 0.924563 33.241734
/fpn0_hm_cen/fpn0_hm_cen.2/Conv BPU id(0) HzSQuantizedConv 0.943938 313.639984
/Resize_3 BPU id(0) HzQuantizedResizeUpsample 0.943938 312.455658
/fpn1_hm_cen/fpn1_hm_cen.0/Conv BPU id(0) HzSQuantizedConv 0.874081 44.009842
/fpn1_hm_cen/fpn1_hm_cen.2/Conv BPU id(0) HzSQuantizedConv 0.861351 500.141083
/fpn2_hm_cen/fpn2_hm_cen.0/Conv BPU id(0) HzSQuantizedConv 0.892319 38.183331
/fpn2_hm_cen/fpn2_hm_cen.2/Conv BPU id(0) HzSQuantizedConv 0.917691 220.486969
/Unsqueeze BPU id(0) Reshape
/Unsqueeze_1 BPU id(0) Reshape
/Unsqueeze_2 BPU id(0) Reshape
/Concat_4 BPU id(0) Concat 0.892625 312.455658
/Softmax CPU – Softmax 0.683802 312.455658
/Mul BPU id(1) HzSElementwiseMul 0.637066 312.455658
/ReduceSum CPU – ReduceSum 0.932151 91.498909
/ReduceSum_reshape CPU – Reshape
/fpn0_cen_offset/fpn0_cen_offset.0/Conv BPU id(0) HzSQuantizedConv 0.779345 33.241734
/fpn0_cen_offset/fpn0_cen_offset.2/Conv BPU id(0) HzSQuantizedConv 0.867806 267.370972
/Resize_4 BPU id(0) HzQuantizedResizeUpsample 0.867808 299.037598
/fpn1_cen_offset/fpn1_cen_offset.0/Conv BPU id(0) HzSQuantizedConv 0.782416 44.009842
/fpn1_cen_offset/fpn1_cen_offset.2/Conv BPU id(0) HzSQuantizedConv 0.665010 233.543671
/fpn2_cen_offset/fpn2_cen_offset.0/Conv BPU id(0) HzSQuantizedConv 0.765405 38.183331
/fpn2_cen_offset/fpn2_cen_offset.2/Conv BPU id(0) HzSQuantizedConv 0.604727 102.658096
/Unsqueeze_3 BPU id(0) Reshape
/Unsqueeze_4 BPU id(0) Reshape
/Unsqueeze_5 BPU id(0) Reshape
/Concat_6 BPU id(0) Concat 0.863411 299.037598
/Softmax_1 CPU – Softmax 0.865288 299.037598
/Mul_1 BPU id(2) HzSElementwiseMul 0.548331 299.037598
/ReduceSum_1 CPU – ReduceSum 0.621538 9.692646
/ReduceSum_1_reshape CPU – Reshape
/fpn0_direction/fpn0_direction.0/Conv BPU id(0) HzSQuantizedConv 0.755548 33.241734
/fpn0_direction/fpn0_direction.2/Conv BPU id(0) HzSQuantizedConv 0.717000 425.373810
/Resize_5 BPU id(0) HzQuantizedResizeUpsample 0.716995 73.949379
/fpn1_direction/fpn1_direction.0/Conv BPU id(0) HzSQuantizedConv 0.705615 44.009842
/fpn1_direction/fpn1_direction.2/Conv BPU id(0) HzSQuantizedConv 0.450235 41.534039
/fpn2_direction/fpn2_direction.0/Conv BPU id(0) HzSQuantizedConv 0.606136 38.183331
/fpn2_direction/fpn2_direction.2/Conv BPU id(0) HzSQuantizedConv 0.606858 138.430069
/Unsqueeze_6 BPU id(0) Reshape
/Unsqueeze_7 BPU id(0) Reshape
/Unsqueeze_8 BPU id(0) Reshape
/Concat_8 BPU id(0) Concat 0.606557 73.949379
/Softmax_2 CPU – Softmax 0.850093 73.949379
/Mul_2 BPU id(3) HzSElementwiseMul 0.620950 73.949379
/ReduceSum_2 CPU – ReduceSum 0.513161 1.124602
/ReduceSum_2_reshape CPU – Reshape
/fpn0_z_coor/fpn0_z_coor.0/Conv BPU id(0) HzSQuantizedConv 0.844084 33.241734
/fpn0_z_coor/fpn0_z_coor.2/Conv BPU id(0) HzSQuantizedConv 0.762663 382.411621
/Resize_6 BPU id(0) HzQuantizedResizeUpsample 0.762663 79.672722
/Resize_6_NHWC2NCHW_LayoutConvert_Output0_reshape BPU id(0) Reshape
/fpn1_z_coor/fpn1_z_coor.0/Conv BPU id(0) HzSQuantizedConv 0.869190 44.009842
/fpn1_z_coor/fpn1_z_coor.2/Conv BPU id(0) HzSQuantizedConv 0.836509 84.870514
…2/Conv_NHWC2NCHW_LayoutConvert_Output0_reshape BPU id(0) Reshape
/fpn2_z_coor/fpn2_z_coor.0/Conv BPU id(0) HzSQuantizedConv 0.915549 38.183331
/fpn2_z_coor/fpn2_z_coor.2/Conv BPU id(0) HzSQuantizedConv 0.958527 53.304226
…2/Conv_NHWC2NCHW_LayoutConvert_Output0_reshape BPU id(0) Reshape
/Concat_10 BPU id(0) Concat 0.760884 79.672722
/Softmax_3 CPU – Softmax 0.953106 79.672722
/Mul_3 BPU id(4) HzSElementwiseMul 0.849502 79.672722
/ReduceSum_3 CPU – ReduceSum 0.933492 2.750437
/ReduceSum_3_reshape CPU – Reshape
/fpn0_dim/fpn0_dim.0/Conv BPU id(0) HzSQuantizedConv 0.905381 33.241734
/fpn0_dim/fpn0_dim.2/Conv BPU id(0) HzSQuantizedConv 0.959605 139.523239
/Resize_7 BPU id(0) HzQuantizedResizeUpsample 0.959608 61.583912
/fpn1_dim/fpn1_dim.0/Conv BPU id(0) HzSQuantizedConv 0.856017 44.009842
/fpn1_dim/fpn1_dim.2/Conv BPU id(0) HzSQuantizedConv 0.675916 107.564522
/fpn2_dim/fpn2_dim.0/Conv BPU id(0) HzSQuantizedConv 0.823923 38.183331
/fpn2_dim/fpn2_dim.2/Conv BPU id(0) HzSQuantizedConv 0.463424 109.336662
/Unsqueeze_12 BPU id(0) Reshape
/Unsqueeze_13 BPU id(0) Reshape
/Unsqueeze_14 BPU id(0) Reshape
/Concat_12 BPU id(0) Concat 0.734860 61.583912
/Softmax_4 CPU – Softmax 0.974800 61.583912
/Mul_4 BPU id(5) HzSElementwiseMul 0.934541 61.583912
/ReduceSum_4 CPU – ReduceSum 0.956766 27.699253
/ReduceSum_4_reshape CPU – Reshape
2023-06-09 15:00:37,416 INFO The quantify model output:
===============================================================================
Node Cosine Similarity L1 Distance L2 Distance Chebyshev Distance
-------------------------------------------------------------------------------
/ReduceSum 0.932151 6.658581 0.038718 72.361755
/ReduceSum_1 0.621538 1.135461 0.006949 20.349892
/ReduceSum_2 0.513161 0.414121 0.002889 2.062102
/ReduceSum_3 0.933492 0.284830 0.002482 2.326995
/ReduceSum_4 0.956766 0.805929 0.006046 19.764456
2023-06-09 15:00:37,416 INFO [Fri Jun 9 15:00:37 2023] End to Horizon NN Model Convert.
2023-06-09 15:00:37,530 WARNING node: Softmax does not exist, please double check your input
2023-06-09 15:00:37,531 INFO start convert to *.bin file…
2023-06-09 15:00:37,566 INFO ONNX model output num : 5
2023-06-09 15:00:37,572 INFO ############# model deps info #############
2023-06-09 15:00:37,572 INFO hb_mapper version : 1.13.5
2023-06-09 15:00:37,572 INFO hbdk version : 3.41.5
2023-06-09 15:00:37,572 INFO hbdk runtime version: 3.15.8.0
2023-06-09 15:00:37,572 INFO horizon_nn version : 0.15.5
2023-06-09 15:00:37,572 INFO ############# model_parameters info #############
2023-06-09 15:00:37,572 INFO onnx_model : /mnt/sda/shiyucun/pcp_j5/larry_pcp/src/super_fast_object_detection/src/sfa/model2onnx/fpn_resnet0420.onnx
2023-06-09 15:00:37,572 INFO BPU march : bayes
2023-06-09 15:00:37,572 INFO layer_out_dump : False
2023-06-09 15:00:37,572 INFO log_level : DEBUG
2023-06-09 15:00:37,572 INFO working dir : /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs
2023-06-09 15:00:37,572 INFO output_model_file_prefix: fpn_resnet
2023-06-09 15:00:37,573 INFO ############# input_parameters info #############
2023-06-09 15:00:37,573 INFO ------------------------------------------
2023-06-09 15:00:37,573 INFO ---------input info : data ---------
2023-06-09 15:00:37,573 INFO input_name : data
2023-06-09 15:00:37,573 INFO input_type_rt : bgr
2023-06-09 15:00:37,573 INFO input_space&range : regular
2023-06-09 15:00:37,573 INFO input_layout_rt : NCHW
2023-06-09 15:00:37,573 INFO input_type_train : bgr
2023-06-09 15:00:37,573 INFO input_layout_train : NCHW
2023-06-09 15:00:37,573 INFO norm_type : no_preprocess
2023-06-09 15:00:37,573 INFO input_shape : 1x3x608x608
2023-06-09 15:00:37,573 INFO input_batch : 1
2023-06-09 15:00:37,573 INFO cal_data_dir : /mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/calibration_data_bgr_f32
2023-06-09 15:00:37,573 INFO cal_data_type : float32
2023-06-09 15:00:37,573 INFO ---------input info : data end -------
2023-06-09 15:00:37,573 INFO ------------------------------------------
2023-06-09 15:00:37,573 INFO ############# calibration_parameters info #############
2023-06-09 15:00:37,573 INFO preprocess_on : False
2023-06-09 15:00:37,573 INFO calibration_type: : kl
2023-06-09 15:00:37,573 INFO max_percentile : 1.0
2023-06-09 15:00:37,573 INFO run_on_bpu : Softmax;
2023-06-09 15:00:37,573 INFO ############# compiler_parameters info #############
2023-06-09 15:00:37,573 INFO hbdk_pass_through_params: --O3 --core-num 1 --fast
2023-06-09 15:00:37,573 INFO input-source : {‘data’: ‘ddr’, ‘_default_value’: ‘ddr’}
2023-06-09 15:00:37,621 INFO Convert to runtime bin file sucessfully!
2023-06-09 15:00:37,621 INFO End Model Convert
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
root@425524437de2:/mnt/sda/shiyucun/oe/oe1.1.37/zeron/model/model_outputs#
如果您只是希望加速softmax的计算，那其实并不需要使用dsp，softmax是支持在BPU上运行的，需要在模型编译前，在yaml文件里配置run_on_bpu，具体方式可以查看工具链手册的4.1.1.6 PTQ章节。不过运行在BPU上有一定可能性导致精度下降。
嗯嗯，这边建议先尝试把softmax运行在BPU上。
您好，参数是这样配的哈，run_on_bpu之后跟的是softmax那个算子的具体名称，比如说你打开转换之前的浮点onnx模型，选中那个算子，会显示softmax算子的具体名称，比如Softmax_103，Softmax_127等等，yaml配置的参考写法如下
run_on_bpu: ‘Softmax_103; Softmax_127’
可以再试一下
run_on_bpu: ‘/Softmax;/Softmax_1;/Softmax_2;/Softmax_3;/Softmax_4’
这样写就好了，我上板测试下，多谢~
另外，使用DSP模型报错的问题还得麻烦帮忙看下，后续Softmax层我们多半要裁掉，这里我们我们只是做个测试，我们是想把使用DSP的这个通路跑通。
目前的状态：
1. DSP侧镜像使用的是OE 1.1.37目录ddk/samples/vdsp_rpc_sample/nn/script/image下的编译好的镜像文件
2. J5 CPU侧程序在初始化时注册了Softmax算子：
ret = hbDNNRegisterLayerCreator(“Softmax”, hobot::dnn::DSPSoftmax_layer_creator);-
if (0 != ret) {-
ROS_ERROR(“hbDNNRegisterLayerCreator failed, ret = %d”, ret);
return ret;
}
一运行就会返回错误码 -6000002
了解，首先-6000002确实是模型初始化出问题了，其次用DSP跑softmax之前确保softmax属于CPU算子（即没有配置run_on_bpu）。另外我们的J5工具链已经更新到1.1.52版本了，您可以更新一下OE包及docker环境重新试一下看看，需要在1.1.52下重新编译模型，并且按照1.1.52手册的运行指导来运行。
好的好的~，run_on_bpu的确没有配置，我先升级下OE包试试~
嗯嗯，后续可以关注下手册这两个部分。
以1.1.52的手册为准。
OK，我试试，3Q~
