load_onnx报错

rknn 1.0.0和0.9.9都试过。

模型文件：https://s3.amazonaws.com/onnx-model-zoo/arcface/resnet100.onnx


转换程序：

1. from rknn.api import RKNN
   
2. 
   
3. rknn = RKNN()
   
4. 
   
5. rknn.config(channel_mean_value='127.5 127.5 127.5 128', reorder_channel='0 1 2')
   
6. 
   
7. print('--> Loading model')
   
8. rknn.load_onnx(model='./resnet100.onnx')
   
9. print('done')
   
10. 
    
11. print('--> Building model')
    
12. rknn.build(do_quantization=False)
    
13. print('done')
    
14. 
    
15. rknn.export_rknn('./resnet100.rknn')
    
16. 
    
17. rknn.release()

复制代码

报错:

1. --> Loading model
   
2. W Not match tensor MatMul_464:out0
   
3. E Catch exception when loading onnx model: ./resnet100.onnx!
   
4. T Traceback (most recent call last):
   
5. T   File "rknn/api/rknn_base.py", line 344, in rknn.api.rknn_base.RKNNBase.load_onnx
   
6. T   File "rknn/base/RKNNlib/converter/convert_onnx.py", line 1037, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx.match_paragraph_and_param
   
7. T   File "rknn/base/RKNNlib/converter/convert_onnx.py", line 950, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx._onnx_push_ready_tensor
   
8. T TypeError: 'NoneType' object is not iterable
   

复制代码

rknn = RKNN(verbose=True)
请把verbose打开，不然看不到什么有用的 log

目前rknn版本1.0.0

1. W verbose file path is invalid, debug info will not dump to file.
   
2. --> Loading model
   
3. I Current ONNX Model use ir_version 3 opset_version 8
   
4. D import clients finished
   
5. I build output layer attach_BatchNormalization_466:out0
   
6. I Try match BatchNormalization_466:out0
   
7. I Match r_bn_v6 [['BatchNormalization_466', 'Initializer_770', 'Initializer_771', 'Initializer_772', 'Initializer_773']] [['BatchNormalization', 'Constant_0', 'Constant_1', 'Constant_2', 'Constant_3']] to [['batchnormalize']]
   
8. I Try match Add_465:out0
   
9. I Match r_add [['Add_465']] [['Add']] to [['add']]
   
10. I Try match MatMul_464:out0
    
11. W Not match tensor MatMul_464:out0
    
12. E Catch exception when loading onnx model: ./resnet100.onnx!
    
13. T Traceback (most recent call last):
    
14. T   File "rknn/api/rknn_base.py", line 344, in rknn.api.rknn_base.RKNNBase.load_onnx
    
15. T   File "rknn/base/RKNNlib/converter/convert_onnx.py", line 1037, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx.match_paragraph_and_param
    
16. T   File "rknn/base/RKNNlib/converter/convert_onnx.py", line 950, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx._onnx_push_ready_tensor
    
17. T TypeError: 'NoneType' object is not iterable
    
18. done
    
19. --> Building model
    
20. E Model or data is None, please load model first.
    
21. done
    
22. E RKNN model data is None, please load model first!
    

复制代码

感谢版版还记得这个帖子，现在onnx版本1.5.0，我降到过1.2.0，也是有问题

请问如何解决 load_onnx 报错.

T TypeError: 'NoneType' object is not iterable

使用rknn v1.0.0
load_tflite OK

完整LOG

1. --> config model
   
2. done
   
3. --> Loading ONNX model
   
4. I Current ONNX Model use ir_version 4 opset_version 9
   
5. D import clients finished
   
6. I build output layer attach_BatchNormalization_31:out0
   
7. I build output layer attach_Div_39:out0
   
8. I Try match Div_39:out0
   
9. I Match r_div [['Div_39']] [['Div']] to [['divide']]
   
10. I Try match BatchNormalization_31:out0
    
11. I Match r_bn_v6 [['BatchNormalization_31', 'Initializer_83', 'Initializer_79', 'Initializer_81', 'Initializer_82']] [['BatchNormalization', 'Constant_0', 'Constant_1', 'Constant_2', 'Constant_3']] to [['batchnormalize']]
    
12. I Try match BatchNormalization_36:out0
    
13. I Match r_bn_v6 [['BatchNormalization_36', 'Initializer_69', 'Initializer_65', 'Initializer_67', 'Initializer_68']] [['BatchNormalization', 'Constant_0', 'Constant_1', 'Constant_2', 'Constant_3']] to [['batchnormalize']]
    
14. I Try match Unsqueeze_38:out0
    
15. I Match r_unsqueeze [['Unsqueeze_38']] [['Unsqueeze']] to [['reshape']]
    
16. I Try match Conv_30:out0
    
17. I Match r_conv [['Conv_30', 'Initializer_78', 'Initializer_77']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
    
18. I Try match Conv_35:out0
    
19. I Match r_conv [['Conv_35', 'Initializer_64', 'Initializer_63']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
    
20. I Try match ReduceL2_37:out0
    
21. W Not match tensor ReduceL2_37:out0
    
22. E Catch exception when loading onnx model: /N/share/build/rknn/rknn-toolkit/example/mobilenet_v1_onnx/superpoint_v1_ws.onnx!
    
23. T Traceback (most recent call last):
    
24. T   File "rknn/api/rknn_base.py", line 344, in rknn.api.rknn_base.RKNNBase.load_onnx
    
25. T   File "rknn/base/RKNNlib/converter/convert_onnx.py", line 1037, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx.match_paragraph_and_param
    
26. T   File "rknn/base/RKNNlib/converter/convert_onnx.py", line 950, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx._onnx_push_ready_tensor
    
27. T TypeError: 'NoneType' object is not iterable
    
28. Load superpoint_v1.onnx failed!
    
29. 
    
30. Process finished with exit code 255

复制代码

yaowei 发表于 2019-5-17 14:20
目前rknn版本1.0.0

Hi @jefferyzhang， @yaowei，先谢谢指点！

yaowei 发表于 2019-5-17 14:20
目前rknn版本1.0.0

您好，请问W verbose file path is invalid, debug info will not dump to file.这个报错你是怎么解决的？

wschong 发表于 2019-6-24 14:57
Hi @jefferyzhang， @yaowei，先谢谢指点！

rknn toolkit可以升级到1.3.2比较好定位问题。

wschong 发表于 2019-6-24 14:57
Hi @jefferyzhang， @yaowei，先谢谢指点！

rknn toolkit可以升级到1.3.2比较好定位问题。