search

vuvko / mxnet2tf

MxNet to TensorFlow converter
MIT License
8 stars 5 forks source link

KeyError: u'_mulscalar0' #1

Open Achhhe opened 6 years ago

Achhhe commented 6 years ago

if name == 'main': parser = argparse.ArgumentParser() parser.add_argument('--mx_prefix',default='/home/xxx/anaconda2/envs/mobilefacenet/insightface_root/insightface/models/model-y1-arcface', help='Prefix of pretrained mxnet model (with path)') parser.add_argument('--tf_prefix',default='/home/xxx/anaconda2/envs/mobilefacenet/mxnet2tf-master', help='Name of TensorFlow model to save to (will be saved as binary protobuf)') parser.add_argument('--input_size', help='Size of network input', type=int, default=112) args = parser.parse_args() main(args.mx_prefix, args.tf_prefix, args.input_size)

hi, I change the mx_prefix and tf_prefix, and run the code, but there is an error KeyError: u'_mulscalar0' i use tensorflow 1.10.1 and mxnet-cu80 1.3.0 could you please give me some advice?

vuvko commented 6 years ago

Hi, it's probably a broadcasted multiplication. Sadly, I didn't add this type of elementwise operation.

vuvko commented 6 years ago

This converter was done for specific networks in mind (Resnext architecture). Now MxNet supports ONNX format so you should use it for transition between frameworks.