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

Error converting from PyTorch to IR in Docker mmdnn/mmdnn:cpu.small #906

Open fprotopapa opened 4 years ago

fprotopapa commented 4 years ago

Platform (like ubuntu 16.04/win10): Ubuntu 16.04.4 LTS, Docker image: mmdnn/mmdnn:cpu.small Python version: 3.5 Source framework with version (like Tensorflow 1.4.1 with GPU): PyTorch 0.4 Destination framework with version (like CNTK 2.3 with GPU): IR Pre-trained model path (webpath or webdisk path): resnet101 Running scripts: mmtoir -f pytorch -d resnet101 --inputShape 3,224,224 -n imagenet_resnet101.pth

Running following commands:

root@291e2e238d7b:/mmdnn/example# mmdownload -f pytorch -n resnet101 -o ./ Downloading: "https://download.pytorch.org/models/resnet101-5d3b4d8f.pth" to /root/.torch/models/resnet101-5d3b4d8f.pth 100.0% PyTorch pretrained model is saved as [./imagenet_resnet101.pth].

root@291e2e238d7b:/mmdnn/example# mmtoir -f pytorch -d resnet101 --inputShape 3,224,224 -n imagenet_resnet101.pth Traceback (most recent call last): File "/usr/local/bin/mmtoir", line 11, in sys.exit(_main()) File "/usr/local/lib/python3.5/dist-packages/mmdnn/conversion/_script/convertToIR.py", line 184, in _main ret = _convert(args) File "/usr/local/lib/python3.5/dist-packages/mmdnn/conversion/_script/convertToIR.py", line 77, in _convert inputshape = [int(x) for x in args.inputShape] File "/usr/local/lib/python3.5/dist-packages/mmdnn/conversion/_script/convertToIR.py", line 77, in inputshape = [int(x) for x in args.inputShape] ValueError: invalid literal for int() with base 10: '3,224,224'

Resulting in error through wrong arg formating.

Changing file:

root@291e2e238d7b:/mmdnn/example# nano /usr/local/lib/python3.5/dist-packages/mmdnn/conversion/_script/convertToIR.py

From line 75 till " parser = PytorchParser(args.network, inputshape)" with:

elif args.srcFramework == 'pytorch':
    assert args.inputShape != None
    inputshape = []
    for x in args.inputShape:
        shape = x.split(',')
    inputshape = [int(x) for x in shape]
    from mmdnn.conversion.pytorch.pytorch_parser import PytorchParser
    parser = PytorchParser(args.network, inputshape)

Eliminates error:

root@291e2e238d7b:/mmdnn/example# mmtoir -f pytorch -d resnet101 --inputShape 3,224,224 -n imagenet_resnet101.pth IR network structure is saved as [resnet101.json]. IR network structure is saved as [resnet101.pb]. IR weights are saved as [resnet101.npy].

cookieli commented 3 years ago

Thank you for your issues and your fix. We will fix it in our next release.