caffe模型转tensorflow模型

caffe转tensorflow的github地址，但是该仓库年久失修，好久不更新。在pull requests中我找到了这个版本。该版本增加了Python3.6和tensorflow1.10的支持。

将alexnet.prototxt文件（deploy.prototxt）和alexnet_cvgj_iter_320000.caffemodel文件放到caffe-tensorflow文件夹的根目录。

当使用命令./convert.py alexnet.prototxt --code-output-path=zdaiot.py转换的时候

遇到了如下错误：

1 2	Error encountered: Cannot determine dimensions of data layer. See comments in function shape_data for more info.

解决方法为：

将原本的文件

name: "AlexNet"
input: "data"
layer {
  name: "input"
  type: "Input"
  top: "data"
  input_param {
    shape {
      dim: 10
      dim: 3
      dim: 227
      dim: 227
    }
  }
}

中的input: "data"删除

然后遇到了下面错误

Converting data...
Saving source...
Traceback (most recent call last):
  File "convert.py", line 60, in <module>
    main()
  File "convert.py", line 56, in main
    args.phase)
  File "convert.py", line 37, in convert
    src_out.write(transformer.transform_source().encode('utf-8'))
  File "/home/zdaiot/Code/caffe-tensorflow/kaffe/tensorflow/transformer.py", line 282, in transform_source
    chains = mapper.map()
  File "/home/zdaiot/Code/caffe-tensorflow/kaffe/graph.py", line 288, in map
    mapped_chains.append(self.map_chain(chain))
  File "/home/zdaiot/Code/caffe-tensorflow/kaffe/graph.py", line 292, in map_chain
    return [self.map_node(node) for node in chain]
  File "/home/zdaiot/Code/caffe-tensorflow/kaffe/graph.py", line 292, in <listcomp>
    return [self.map_node(node) for node in chain]
  File "/home/zdaiot/Code/caffe-tensorflow/kaffe/graph.py", line 296, in map_node
    mapped_node = map_func(node)
  File "/home/zdaiot/Code/caffe-tensorflow/kaffe/tensorflow/transformer.py", line 149, in map_batch_norm
    scale_offset = len(node.data) == 4
TypeError: object of type 'NoneType' has no len()

解决方法为
I got it to work by replacing scale_offset= len(node.data) == 4 with scale_offset = len(node.output_shape) == 4 in transformer.py

使用Python2的时候，报错

1	TypeError: __new__() got an unexpected keyword argument 'serialized_options'

解决方法，出现该问题，很有可能是，终端上的 protoc 版本与python库内的protobuf版本不一样。

1	sudo pip2 install -U protobuf

接着使用./convert.py alexnet.prototxt --caffemodel alexnet_cvgj_iter_320000.caffemodel --data-output-path=zdaiot.npy将权重文件转为npy格式。