Toybrick

标题: 请问下 negative 支不支持啊？ [打印本页]

---

作者: xsky    时间: 2019-12-2 21:04
标题: 请问下 negative 支不支持啊？

请问neg支不支持啊，这个简单的操作

导入onnx模型
[attach]766[/attach]


导入代码：

1. from rknn.api import RKNN  
   
2. 
   
3. INPUT_SIZE = 64
   
4. 
   
5. if __name__ == '__main__':
   
6.     # 创建RKNN执行对象
   
7.     rknn = RKNN(verbose=True)
   
8.     # 配置模型输入，用于NPU对数据输入的预处理
   
9.     # channel_mean_value='0 0 0 255'，那么模型推理时，将会对RGB数据做如下转换
   
10.     # (R - 0)/255, (G - 0)/255, (B - 0)/255。推理时，RKNN模型会自动做均值和归一化处理
    
11.     # reorder_channel=’0 1 2’用于指定是否调整图像通道顺序，设置成0 1 2即按输入的图像通道顺序不做调整
    
12.     # reorder_channel=’2 1 0’表示交换0和2通道，如果输入是RGB，将会被调整为BGR。如果是BGR将会被调整为RGB
    
13.     #图像通道顺序不做调整
    
14.     rknn.config(channel_mean_value='0 0 0 255', reorder_channel='0 1 2')
    
15. 
    
16.     #model_file = 'XMC2-Cls_face_emotion_classifier.onnx'
    
17.     model_file = 'XMC2-Det_student_detector.onnx'
    
18.     #model_file = 'XMC2-Det_student_detector.pth.tar.onnx'
    
19.     #model_file = 'XMC2-Det_student_detector.pth.tar_op-v8.onnx'
    
20. 
    
21.     # 加载TensorFlow模型
    
22.     # tf_pb='digital_gesture.pb'指定待转换的TensorFlow模型
    
23.     # inputs指定模型中的输入节点
    
24.     # outputs指定模型中输出节点
    
25.     # input_size_list指定模型输入的大小
    
26.     print('--> Loading model', model_file)
    
27.     # rknn.load_tensorflow(tf_pb='digital_gesture.pb',
    
28.     #                      inputs=['input_x'],
    
29.     #                      outputs=['probability'],
    
30.     #                      input_size_list=[[INPUT_SIZE, INPUT_SIZE, 3]])
    
31.     rknn.load_onnx(model=model_file)
    
32. 
    
33.     print('done')
    
34. 
    
35.     is_quantization = False
    
36. 
    
37.     # 创建解析pb模型
    
38.     # do_quantization=False指定不进行量化
    
39.     # 量化会减小模型的体积和提升运算速度，但是会有精度的丢失
    
40.     print('--> Building model')
    
41.     rknn.build(do_quantization=is_quantization)
    
42.     print('done')
    
43. 
    
44.     # 导出保存rknn模型文件
    
45.     if is_quantization:
    
46.         quant_suff = '_quant'
    
47.     else:
    
48.         quant_suff = '_un-quant'
    
49.     rknn.export_rknn( model_file + quant_suff + '.rknn')
    
50. 
    
51.     # Release RKNN Context
    
52.     rknn.release()
    
53. 
    
54. 
    

复制代码



报错信息：

1. I Match r_maxpool [['MaxPool_9']] [['MaxPool']] to [['pooling']]
   
2. I Try match AveragePool_13:out0
   
3. I Match r_avgpool [['AveragePool_13']] [['AveragePool']] to [['pooling']]
   
4. I Try match Relu_17:out0
   
5. I Match r_relu [['Relu_17']] [['Relu']] to [['relu']]
   
6. I Try match Conv_22:out0
   
7. I Match r_conv [['Conv_22', 'Initializer_26', 'Initializer_27']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
   
8. I Try match Relu_8:out0
   
9. I Match r_relu [['Relu_8']] [['Relu']] to [['relu']]
   
10. I Try match Pad_12:out0
    
11. I Match pad_g1 [['Pad_12']] [['Pad']] to [['pad']]
    
12. I Try match Conv_16:out0
    
13. I Match r_conv [['Conv_16', 'Initializer_20', 'Initializer_21']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
    
14. I Try match Relu_21:out0
    
15. I Match r_relu [['Relu_21']] [['Relu']] to [['relu']]
    
16. I Try match Concat_7:out0
    
17. I Match concat_2 [['Concat_7']] [['Concat']] to [['concat']]
    
18. I Try match Neg_6:out0
    
19. W Not match tensor Neg_6:out0
    
20. E Catch exception when loading onnx model: XMC2-Det_student_detector.onnx!
    
21. E Traceback (most recent call last):
    
22. E   File "rknn\api\rknn_base.py", line 345, in rknn.api.rknn_base.RKNNBase.load_onnx
    
23. E   File "rknn\base\RKNNlib\converter\convert_onnx.py", line 1072, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx.match_paragraph_and_param
    
24. E   File "rknn\base\RKNNlib\converter\convert_onnx.py", line 985, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx._onnx_push_ready_tensor
    
25. E TypeError: 'NoneType' object is not iterable
    
26. Traceback (most recent call last):
    
27.   File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 2073, in <module>
    
28.     main()
    
29.   File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 2067, in main
    
30.     globals = debugger.run(setup['file'], None, None, is_module)
    
31.   File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 1418, in run
    
32.     return self._exec(is_module, entry_point_fn, module_name, file, globals, locals)
    
33.   File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 1425, in _exec
    
34.     pydev_imports.execfile(file, globals, locals)  # execute the script
    
35.   File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\_pydev_imps\_pydev_execfile.py", line 18, in execfile
    
36.     exec(compile(contents+"\n", file, 'exec'), glob, loc)
    
37.   File "E:/Work/_Code/dmax/tools/covert_onnx/main.py", line 42, in <module>
    
38.     rknn.build(do_quantization=is_quantization)
    
39.   File "D:\Program Files\Python\Python36\lib\site-packages\rknn\api\rknn.py", line 148, in build
    
40.     inputs = self.rknn_base.net.get_input_layers()
    
41. AttributeError: 'NoneType' object has no attribute 'get_input_layers'
    

复制代码

---

作者: jefferyzhang    时间: 2019-12-3 09:40
W Not match tensor Neg_6ut0

至少说明在ONNX上不支持，你可以试试：
1. 用tf、caffe写支不支持
2. 这么简单的操作自己写个自定义OP应该也不会很难（参看文档）。

---

欢迎光临 Toybrick (https://t.rock-chips.com/)

Powered by Discuz! X3.3