search

microsoft / MMdnn

MMdnn is a set of tools to help users inter-operate among different deep learning frameworks. E.g. model conversion and visualization. Convert models between Caffe, Keras, MXNet, Tensorflow, CNTK, PyTorch Onnx and CoreML.
MIT License
5.8k stars 965 forks source link

conver mxnet model to tensorflow, meet some errors!!! #849

Closed visonpon closed 4 years ago

visonpon commented 4 years ago

Platform (like ubuntu 16.04/win10): ubuntu16.04 Python version: python3.6 Source framework with version (like Tensorflow 1.4.1 with GPU): tensorflow1.12 mxnet.1.12, Destination framework with version (like CNTK 2.3 with GPU):

Pre-trained model path (webpath or webdisk path):

Running scripts: python3 -m mmdnn.conversion._scripts.convertToIR -f mxnet -n ./.json -w ./.params -d net --inputShape 3,224,224

error messages like follows: segmentation fault :11 Stack trace returned 10 entries: [bt] (0) /home/csf/anaconda3/envs/tf18/lib/python3.6/site-packages/mxnet/libmxnet.so(+0x31f81a) [0x7f03434c581a] .... Hope can get some help~
thanks~

linmajia commented 4 years ago

@visonpon , thank you very much for the feedback. Would you please provide the complete stack traces, and your model files if possible?

visonpon commented 4 years ago

trans

Above is the error message, thanks for your quick response.

linmajia commented 4 years ago

@visonpon , it seems to be an internal error of MXNet. Would you please provide your model files and corresponding source code?

visonpon commented 4 years ago

@linmajia Sorry that it's not convenient for me to upload the model and code, but I test my trained model on my current mxnet, and I also try to convert other simple mxnet models to TensorFlow, everything is normal. Only when I try to convert the new model, it fails.

linmajia commented 4 years ago

@visonpon , you may try a different version of MXNet (e.g., 1.2.0).