Performance? about flownet2-pytorch

[fitsumreda]

we'll release scripts that downloads and converts flownet2-official (from authors) Caffe pre-trained models to PyTorch, which can be used with --resume argument in this repository
we will also provide information about model training from scratch.

[fitsumreda]

we've added instructions that downloads and converts official caffe-pre-trained models to PyTorch.
https://github.com/NVIDIA/flownet2-pytorch#convert-official-caffe-pre-trained-models-to-pytorch

you may use the converted checkpoints with the --resume argument in this repository

[PkuRainBow]

@fitsumreda I am also wondering the whether the performance of this code base can match the caffe version flownet2. It would be great if you could share us the related experimental numbers.

[Yuliang-Zou]

Hi @fitsumreda , thanks for the response. Since installing Caffe on cluster could be messy, can you provide download links for us to get the converted PyTorch models? Like Google Drive or Dropbox.

Also, I am still curious about the performance (training from scratch using this code, and using converted models with this code).