search

TuSimple / TuSimple-DUC

Understanding Convolution for Semantic Segmentation
https://arxiv.org/abs/1702.08502
Apache License 2.0
605 stars 118 forks source link

Failed to load the BatchNorm Block in ResNet_DUC_HDC_CityScapes-symbol.json #13

Closed Serian1992 closed 6 years ago

Serian1992 commented 6 years ago

Hi, I'm very interested in your work. But when I tried to repeat your experiment, a MXNetError appeared that Cannot find argument 'fix_linear_trans' when loading the ResNet_DUC_HDC_CityScapes-symbol.json. My current environment is Python 2.7, Mxnet version 1.0.0 under Win10. Is this error because of the operating system or the different Mxnet version ?

GrassSunFlower commented 6 years ago

It was due to MxNet version mismatch. The symbol file was uploaded like 1 year ago. There are three ways of fixing this problem:

  1. Try using MxNet from our fork: https://github.com/TuSimple/mxnet
  2. Or you can re-download the updated symbol file later. I'll update the file by tomorrow.
Serian1992 commented 6 years ago

Thanks a lot! I may need the updated symbol file. Also, to verify the efficiency of HDC and DUC, I would like to reinitialize and train the ResNet_DUC_HDC_CityScapes right from the baseline model ResNet-101 just like you, but seems like that only the well-reinited model is provided. Could you please also provide the source code on initializing the baseline model? Much appreciate.

GrassSunFlower commented 6 years ago

The symbol is already updated, you can check the download link. As for the pretrained model, please see #1