运行mobilenetssd-v2 无法转换 rknn 模型

[复制链接] · 发表于 2019-7-19 14:06:27

系统：fedora28
toolkit版本： 1.0.0
tensorflow版本： 1.10.1

按照官方的训练步骤重新训练了pb文件
用mobilenetssd-v2 step1.py 无法转换
不知道什么原因谢谢指导~~~

step1.py 代码：

import numpy as np
import cv2
from rknn.api import RKNN

if __name__ == '__main__':

# Create RKNN object
rknn = RKNN(verbose = True, verbose_file =  './Error.log')

# pre-process config
print('--> config model')
rknn.config(channel_mean_value='127.5 127.5 127.5 128', reorder_channel='0 1 2', quantized_dtype='asymmetric_quantized-u8')
print('done')

# Load tflite model
print('--> Loading model')
ret = rknn.load_tensorflow(tf_pb='./frozen_inference_graph.pb',
                           inputs=['FeatureExtractor/MobilenetV2/MobilenetV2/input'],
                           outputs=['concat_1', 'concat'],
                           input_size_list=[[300,300,3]],
                           predef_file=None)
if ret != 0:
      print('Load model failed!')
      exit(ret)
print('done')

if ret != 0:
      print('Load model failed!')
      exit(ret)
print('done')

# Build model
print('--> hybrid_quantization_step1')
ret = rknn.hybrid_quantization_step1(dataset='./dataset.txt')
if ret != 0:
      print('hybrid_quantization_step1 failed!')
      exit(ret)
print('done')

# Tips
print('Please modify ssd_mobilenet_v2.quantization.cfg!')
print('==================================================================================================')
print('Modify method:')
print('1. delete @FeatureExtractor/MobilenetV2/expanded_conv/depthwise/depthwise_227:weight and its value')
print('2. delete @FeatureExtractor/MobilenetV2/expanded_conv/depthwise/depthwise_227:bias and its value')
print('3. delete @FeatureExtractor/MobilenetV2/Conv/Relu6_228

ut0 and its value')
print('==================================================================================================')
print('Original quantization profile and modified quantization profile diff shows in file quantization_profile.diff')

rknn.release()

报错信息:

E Catch exception when loading tensorflow model: ./frozen_inference_graph.pb!
T Traceback (most recent call last):
T File "/usr/local/lib/python3.6/site-packages/tensorflow/python/framework/importer.py", line 418, in import_graph_def
T    graph._c_graph, serialized, options)  # pylint: disable=protected-access
T tensorflow.python.framework.errors_impl.InvalidArgumentError: NodeDef mentions attr 'explicit_paddings' not in Op<name=Conv2D; signature=input:T, filter:T -> output:T; attr=T:type,allowed=[DT_HALF, DT_BFLOAT16, DT_FLOAT, DT_DOUBLE]; attr=strides:list(int); attr=use_cudnn_on_gpu:bool,default=true; attr=padding:string,allowed=["SAME", "VALID"]; attr=data_format:string,default="NHWC",allowed=["NHWC", "NCHW"]; attr=dilations:list(int),default=[1, 1, 1, 1]>; NodeDef: FeatureExtractor/MobilenetV2/Conv/Conv2D = Conv2D[T=DT_FLOAT, data_format="NHWC", dilations=[1, 1, 1, 1], explicit_paddings=[], padding="SAME", strides=[1, 2, 2, 1], use_cudnn_on_gpu=true](FeatureExtractor/MobilenetV2/MobilenetV2/input, FeatureExtractor/MobilenetV2/Conv/weights). (Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary.).
T During handling of the above exception, another exception occurred:
T Traceback (most recent call last):
T File "rknn/api/rknn_base.py", line 137, in rknn.api.rknn_base.RKNNBase.load_tensorflow
T File "rknn/base/RKNNlib/converter/convert_tf.py", line 482, in rknn.base.RKNNlib.converter.convert_tf.convert_tf.pre_process
T File "rknn/base/RKNNlib/converter/tensorflowloader.py", line 102, in rknn.base.RKNNlib.converter.tensorflowloader.TF_Graph_Preprocess.pre_proces
T File "rknn/base/RKNNlib/converter/tensorflowloader.py", line 627, in rknn.base.RKNNlib.converter.tensorflowloader.TF_Graph_Preprocess.calc_2_const
T File "rknn/base/RKNNlib/converter/tf_util.py", line 371, in rknn.base.RKNNlib.converter.tf_util.TFProto_Util.query_tensor
T File "rknn/base/RKNNlib/converter/tf_util.py", line 372, in rknn.base.RKNNlib.converter.tf_util.TFProto_Util.query_tensor
T File "/usr/local/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py", line 454, in new_func
T    return func(*args, **kwargs)
T File "/usr/local/lib/python3.6/site-packages/tensorflow/python/framework/importer.py", line 422, in import_graph_def
T    raise ValueError(str(e))
T ValueError: NodeDef mentions attr 'explicit_paddings' not in Op<name=Conv2D; signature=input:T, filter:T -> output:T; attr=T:type,allowed=[DT_HALF, DT_BFLOAT16, DT_FLOAT, DT_DOUBLE]; attr=strides:list(int); attr=use_cudnn_on_gpu:bool,default=true; attr=padding:string,allowed=["SAME", "VALID"]; attr=data_format:string,default="NHWC",allowed=["NHWC", "NCHW"]; attr=dilations:list(int),default=[1, 1, 1, 1]>; NodeDef: FeatureExtractor/MobilenetV2/Conv/Conv2D = Conv2D[T=DT_FLOAT, data_format="NHWC", dilations=[1, 1, 1, 1], explicit_paddings=[], padding="SAME", strides=[1, 2, 2, 1], use_cudnn_on_gpu=true](FeatureExtractor/MobilenetV2/MobilenetV2/input, FeatureExtractor/MobilenetV2/Conv/weights). (Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary.).

只看该作者 · 发表于 2019-7-26 08:42:39

是不是你的pb文件是高版本的TensorFlow生成的，上面日志，报错的点在TensorFlow里面。