请问下 negative 支不支持啊？

[复制链接] · 发表于 2019-12-2 21:04:13

请问neg支不支持啊，这个简单的操作

导入onnx模型

导入代码：

from rknn.api import RKNN  

 

INPUT_SIZE = 64

 

if __name__ == '__main__':

    # 创建RKNN执行对象

    rknn = RKNN(verbose=True)

    # 配置模型输入，用于NPU对数据输入的预处理

    # 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

    # (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

    # 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

    # 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

    #图像通道顺序不做调整

    rknn.config(channel_mean_value='0 0 0 255', reorder_channel='0 1 2')



    #model_file = 'XMC2-Cls_face_emotion_classifier.onnx'

    model_file = 'XMC2-Det_student_detector.onnx'

    #model_file = 'XMC2-Det_student_detector.pth.tar.onnx'

    #model_file = 'XMC2-Det_student_detector.pth.tar_op-v8.onnx'



    # 加载TensorFlow模型

    # tf_pb='digital_gesture.pb'指定待转换的TensorFlow模型

    # inputs指定模型中的输入节点

    # outputs指定模型中输出节点

    # input_size_list指定模型输入的大小

    print('--> Loading model', model_file)

    # rknn.load_tensorflow(tf_pb='digital_gesture.pb',

    #                      inputs=['input_x'],

    #                      outputs=['probability'],

    #                      input_size_list=[[INPUT_SIZE, INPUT_SIZE, 3]])

    rknn.load_onnx(model=model_file)



    print('done')



    is_quantization = False



    # 创建解析pb模型

    # do_quantization=False指定不进行量化

    # 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

    print('--> Building model')

    rknn.build(do_quantization=is_quantization)

    print('done')

 

    # 导出保存rknn模型文件

    if is_quantization:

        quant_suff = '_quant'

    else:

        quant_suff = '_un-quant'

    rknn.export_rknn( model_file + quant_suff + '.rknn')

 

    # Release RKNN Context

    rknn.release()





复制代码

报错信息：

I Match r_maxpool [['MaxPool_9']] [['MaxPool']] to [['pooling']]

I Try match AveragePool_13:out0

I Match r_avgpool [['AveragePool_13']] [['AveragePool']] to [['pooling']]

I Try match Relu_17:out0

I Match r_relu [['Relu_17']] [['Relu']] to [['relu']]

I Try match Conv_22:out0

I Match r_conv [['Conv_22', 'Initializer_26', 'Initializer_27']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]

I Try match Relu_8:out0

I Match r_relu [['Relu_8']] [['Relu']] to [['relu']]

I Try match Pad_12:out0

I Match pad_g1 [['Pad_12']] [['Pad']] to [['pad']]

I Try match Conv_16:out0

I Match r_conv [['Conv_16', 'Initializer_20', 'Initializer_21']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]

I Try match Relu_21:out0

I Match r_relu [['Relu_21']] [['Relu']] to [['relu']]

I Try match Concat_7:out0

I Match concat_2 [['Concat_7']] [['Concat']] to [['concat']]

I Try match Neg_6:out0

W Not match tensor Neg_6:out0 

E Catch exception when loading onnx model: XMC2-Det_student_detector.onnx!

E Traceback (most recent call last):

E   File "rknn\api\rknn_base.py", line 345, in rknn.api.rknn_base.RKNNBase.load_onnx

E   File "rknn\base\RKNNlib\converter\convert_onnx.py", line 1072, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx.match_paragraph_and_param

E   File "rknn\base\RKNNlib\converter\convert_onnx.py", line 985, in rknn.base.RKNNlib.converter.convert_onnx.convert_onnx._onnx_push_ready_tensor

E TypeError: 'NoneType' object is not iterable

Traceback (most recent call last):

  File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 2073, in <module>

    main()

  File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 2067, in main

    globals = debugger.run(setup['file'], None, None, is_module)

  File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 1418, in run

    return self._exec(is_module, entry_point_fn, module_name, file, globals, locals)

  File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 1425, in _exec

    pydev_imports.execfile(file, globals, locals)  # execute the script

  File "D:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\_pydev_imps\_pydev_execfile.py", line 18, in execfile

    exec(compile(contents+"\n", file, 'exec'), glob, loc)

  File "E:/Work/_Code/dmax/tools/covert_onnx/main.py", line 42, in <module>

    rknn.build(do_quantization=is_quantization)

  File "D:\Program Files\Python\Python36\lib\site-packages\rknn\api\rknn.py", line 148, in build

    inputs = self.rknn_base.net.get_input_layers()

AttributeError: 'NoneType' object has no attribute 'get_input_layers'

复制代码

只看该作者 · 发表于 2019-12-3 09:40:00

W Not match tensor Neg_6

ut0

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）。

请问下 negative 支不支持啊？

本帖子中包含更多资源