Fail to convert a TF model to caffe2 or Pytorch

[bwang482]

Platform (like ubuntu 16.04/win10): Ubuntu 16.04

Python version: Python 3.5.5

Source framework with version (like Tensorflow 1.4.1 with GPU): Tensorflow 1.10.1 with GPU

Destination framework with version (like CNTK 2.3 with GPU): Caffe2 with CPU / Pytorch 0.4.1 with GPU

Pre-trained model path (webpath or webdisk path): No web path. It is a pre-trained cnn model in tensorflow

Error:

aaa@aaa:$ mmconvert -sf tensorflow -in cnn_tree.ckpt-51243.meta -iw cnn_tree.ckpt-51243 --dstNodeName output/Softmax -df pytorch -om tf2pytorch
/anaconda/envs/py35/lib/python3.5/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
Parse file [cnn_tree.ckpt-51243.meta] with binary format successfully.
Tensorflow model file [cnn_tree.ckpt-51243.meta] loaded successfully.
Tensorflow checkpoint file [cnn_tree.ckpt-51243] loaded successfully. [21] variables loaded.
Tensorflow has not supported operator [Cast] with name [network/conv_layer/conv_node/conv_step/coefficients/coef_r/Cast].
Tensorflow has not supported operator [Cast] with name [network/conv_layer/conv_node/conv_step/coefficients/coef_l/Cast].
Tensorflow has not supported operator [Less] with name [network/conv_layer/conv_node/conv_step/combine/Tensordot/Less].
Tensorflow has not supported operator [ExpandDims] with name [network/conv_layer/conv_node/conv_step/trees/children_tensor/ExpandDims].
Tensorflow has not supported operator [ExpandDims] with name [network/conv_layer/conv_node/conv_step/trees/ExpandDims].
Tensorflow has not supported operator [GreaterEqual] with name [network/conv_layer/conv_node/conv_step/combine/Tensordot/GreaterEqual].
Tensorflow has not supported operator [Range] with name [network/conv_layer/conv_node/conv_step/combine/Tensordot/range].
Tensorflow has not supported operator [NotEqual] with name [network/conv_layer/conv_node/conv_step/coefficients/coef_r/count_nonzero/NotEqual].
Tensorflow has not supported operator [Cast] with name [network/conv_layer/conv_node/conv_step/combine/Tensordot/Cast_1].
Tensorflow has not supported operator [Cast] with name [network/conv_layer/conv_node/conv_step/combine/Tensordot/Cast].
Tensorflow has not supported operator [Cast] with name [network/conv_layer/conv_node/conv_step/coefficients/coef_r/count_nonzero/ToInt64].
Tensorflow has not supported operator [Fill] with name [network/conv_layer/conv_node/conv_step/coefficients/coef_r/ones].
Tensorflow has not supported operator [Fill] with name [network/conv_layer/conv_node/conv_step/coefficients/coef_l/ones].
Traceback (most recent call last):
  File "/home/ccrmad/.local/bin/mmconvert", line 11, in <module>
    sys.exit(_main())
  File "/home/ccrmad/.local/lib/python3.5/site-packages/mmdnn/conversion/_script/convert.py", line 102, in _main
    ret = convertToIR._convert(ir_args)
  File "/home/ccrmad/.local/lib/python3.5/site-packages/mmdnn/conversion/_script/convertToIR.py", line 115, in _convert
    parser.run(args.dstPath)
  File "/home/ccrmad/.local/lib/python3.5/site-packages/mmdnn/conversion/common/DataStructure/parser.py", line 22, in run
    self.gen_IR()
  File "/home/ccrmad/.local/lib/python3.5/site-packages/mmdnn/conversion/tensorflow/tensorflow_parser.py", line 309, in gen_IR
    func(current_node)
  File "/home/ccrmad/.local/lib/python3.5/site-packages/mmdnn/conversion/tensorflow/tensorflow_parser.py", line 703, in rename_Mul
    shape = shape[-1]
IndexError: list index out of range

There are a lot of not supported operators, and thus the reason the failure?

[rainLiuplus]

Hi @bluemonk482 , could you please provide your model?

[bwang482]

@rainLiuplus Thanks for your reply! Here is the saved weights and graph of the model: https://www.dropbox.com/s/7j3d0zptjl8qo0h/cnn_tree.ckpt-51243.zip?dl=0

It is pretty small.

[chungyau97]

@rainLiuplus @bluemonk482 How to convert it to pytorch? I'm using original author pretrained-model. mmconvert -sf tensorflow -in model-20180402-114759.meta -iw model-20180402-114759.ckpt-275.index --dstNodeName embeddings -df pytorch -om 20180402-114759.pth

Error:

[spaul13]

Anyone successfully able to convert tensorflow to caffe2?

I am getting the following error while conversion. Can anyone plz tell me how to resolve this error?

_paul90@cloud8:/export1/paul90/new_tensorflow/tensorflow$ mmconvert -sf tensorflow -in sp_birnn_ckpt/sp_model.ckpt.meta -iw sp_birnn_ckpt/sp_model.ckpt.index --dstNodeName GradientDescent -df caffe2 -om tf2Caffe2 TensorflowEmitter has not supported operator [StopGradient] with name [softmax_cross_entropy_with_logits_sg/labels_stop_gradient]. TensorflowEmitter has not supported operator [NoOp] with name [GradientDescent]. Traceback (most recent call last): File "/home/paul90/.local/bin/mmconvert", line 11, in sys.exit(_main()) File "/home/paul90/.local/lib/python2.7/site-packages/mmdnn/conversion/_script/convert.py", line 102, in _main ret = convertToIR._convert(ir_args) File "/home/paul90/.local/lib/python2.7/site-packages/mmdnn/conversion/_script/convertToIR.py", line 115, in _convert parser.run(args.dstPath) File "/home/paul90/.local/lib/python2.7/site-packages/mmdnn/conversion/common/DataStructure/parser.py", line 22, in run self.gen_IR() File "/home/paul90/.local/lib/python2.7/site-packages/mmdnn/conversion/tensorflow/tensorflow_parser.py", line 424, in gen_IR func(current_node) File "/home/paul90/.local/lib/python2.7/site-packages/mmdnn/conversion/tensorflow/tensorflow_parser.py", line 967, in rename_Slice shape = tensor_util.MakeNdarray(shape).tolist() File "/home/paul90/.local/lib/python2.7/site-packages/tensorflow/python/framework/tensor_util.py", line 597, in MakeNdarray tensor_dtype = dtypes.as_dtype(tensor.dtype) File "/home/paul90/.local/lib/python2.7/site-packages/tensorflow/python/framework/dtypes.py", line 719, in as_dtype raise TypeError("Cannot convert value %r to a TensorFlow DType." % typevalue) TypeError: Cannot convert value 0 to a TensorFlow DType.

[grapefruitL]

@spaul13 hi ! i got the same err on my mtcnn-tf-model2caffe work did you solve the problem ?

[bin9000]

@grapefruitL I got the same problem and I solved. Try to install mmdnn again, following this website https://github.com/microsoft/MMdnn

[bitwangdan]

@spaul13 @grapefruitL hi, i got the same err, did you solve the problem ?