Getting weird output on using a different correlation implementation

NVlabs / PWC-Net

PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume, CVPR 2018 (Oral)

Other

1.63k stars 357 forks source link

Getting weird output on using a different correlation implementation #59

Open pranaypratyush opened 5 years ago

pranaypratyush commented 5 years ago

Hi there, I tried using https://github.com/ClementPinard/Pytorch-Correlation-extension for the correlation with the arguments (dilation_patch=1, kernel_size=1, stride=1, patch_size=(md * 2 + 1), padding=0). But I keep getting weird output for the pre-trained models. I am on pytorch 1. Would that be of any concern? download download 1

jrenzhile commented 5 years ago

You may follow the readme page for environment setup: https://github.com/NVlabs/PWC-Net/tree/master/PyTorch#installation I'm not able to help with issues with very diverse implementation details other than this one.

xianshunw commented 5 years ago

have you solved this problem? I also have such a problem.

pranaypratyush commented 5 years ago

I still haven't been able to solve this problem.

lelelexxx commented 5 years ago

@pranaypratyush Is the value of optical flow extremely large? I have reimplement pwcnet using mxnet, and when i loaded the pytorch pre-trained weights into my mxnet model, the EPE in MPI-Sintel and flyingchairs around 1000. Doubting about the different implement of correlation layer.