onnx模型转换checker报错 No quantifiable nodes were found

编辑：

之前没有发现工具链有更新，现在用新版的工具链（hbdk version 3.18.2 horizon_nn version 0.9.10 hb_mapper version 1.2.6）尝试了一下，简单的模型已经能够顺利转换了（op10），但是我们想要部署的模型（op11）仍然会出现以下报错：

ERROR HorizonRT not support these cpu operators: Where Gather Expand ScatterND

原始内容：

我在使用模型转换工具链转换onnx模型时出现报错No quantifiable nodes were found。起初以为是模型的问题，就先生成了一个更为简单的模型进行测试，但还是出现同样的报错，我在论坛上也没有搜索到类似的情况。请问可能是什么原因？

另外，请问，现在版本的模型转换工具链是否支持onnx opset_version=11？我们想要部署的模型在转换成onnx opset_version=10会有一些问题。

报错信息

log记录更详细信息：

2021-05-07 11:36:48,904 INFO hb_mapper_checker 67 Model type: onnx

2021-05-07 11:36:48,905 DEBUG hb_mapper_checker 68 march: bernoulli2

2021-05-07 11:36:48,905 INFO hb_mapper_checker 69 output file: ./playplay_checker.log

2021-05-07 11:36:48,907 INFO hb_mapper_checker 74 input names []

2021-05-07 11:36:48,907 INFO hb_mapper_checker 75 input shapes {}

2021-05-07 11:36:48,911 INFO hb_mapper_checker 81 Begin model checking....

2021-05-07 11:36:48,914 INFO build 36 [Fri May 7 11:36:48 2021] Start to Horizon NN Model Convert.

2021-05-07 11:36:48,915 INFO build 162 The input parameter is not specified, convert with default parameters.

2021-05-07 11:36:48,917 INFO build 346 The hbdk parameter is not specified, and the submodel will be compiled with the default parameter.

2021-05-07 11:36:48,921 INFO build 137 HorizonNN version: 0.9.7

2021-05-07 11:36:48,922 INFO build 141 HBDK version: 3.16.6

2021-05-07 11:36:48,923 INFO build 36 [Fri May 7 11:36:48 2021] Start to parse the onnx model.

2021-05-07 11:36:48,942 INFO onnx_parser 101 ONNX model info:

ONNX IR version: 6

Opset version: 10

Input name: 0, [25, 3]

2021-05-07 11:36:48,957 INFO build 39 [Fri May 7 11:36:48 2021] End to parse the onnx model.

2021-05-07 11:36:48,959 INFO build 457 Model input names: ['0']

2021-05-07 11:36:48,986 INFO build 821 Saving the original float model: ./.hb_check/original_float_model.onnx.

2021-05-07 11:36:48,987 INFO build 36 [Fri May 7 11:36:48 2021] Start to optimize the model.

2021-05-07 11:36:49,006 INFO build 39 [Fri May 7 11:36:49 2021] End to optimize the model.

2021-05-07 11:36:49,022 INFO build 832 Saving the optimized model: ./.hb_check/optimized_float_model.onnx.

2021-05-07 11:36:49,023 INFO build 36 [Fri May 7 11:36:49 2021] Start to calibrate the model.

2021-05-07 11:36:49,030 DEBUG tool_utils 95 exception in checker

2021-05-07 11:36:49,037 DEBUG tool_utils 96 Traceback (most recent call last):

File "/root/.local/lib/python3.6/site-packages/x3_tc_ui/utils/tool_utils.py", line 93, in __decorator

func(*args, **kargs)

File "/root/.local/lib/python3.6/site-packages/x3_tc_ui/hb_mapper.py", line 69, in checker

input_shape).run(version)

File "/root/.local/lib/python3.6/site-packages/x3_tc_ui/hb_mapper_checker.py", line 97, in run

**addtional_param)

File "/root/.local/lib/python3.6/site-packages/horizon_nn/build.py", line 1167, in check_onnx

return builder.build_model(original_model, save_model)

File "/root/.local/lib/python3.6/site-packages/horizon_nn/build.py", line 835, in build_model

calibration_table = self.model_calibration(optimized_model)

File "/root/.local/lib/python3.6/site-packages/horizon_nn/build.py", line 710, in model_calibration

optimized_model, self.march, "", self.node_dict)

File "/root/.local/lib/python3.6/site-packages/horizon_nn/horizon_onnx/quantizer.py", line 206, in create_pre_calibrated_model

node_dict)

horizon_nn.horizon_onnx.onnx_cpp2py_export.checker.ValidationError: No quantifiable nodes were found, and the model is not supported.

2021-05-07 11:36:49,037 ERROR tool_utils 97 No quantifiable nodes were found, and the model is not supported.

2021-05-07 11:36:49,038 INFO build 39 [Fri May 7 11:36:49 2021] End to calibrate the model.

2021-05-07 11:36:49,046 INFO build 39 [Fri May 7 11:36:49 2021] End to Horizon NN Model Convert.

使用Netron查看onnx模型信息如下：

使用Netron查看checker优化后onnx模型信息如下（./.hb_check/optimized_float_model.onnx.）：