search

xinntao / EDVR

Winning Solution in NTIRE19 Challenges on Video Restoration and Enhancement (CVPR19 Workshops) - Video Restoration with Enhanced Deformable Convolutional Networks. EDVR has been merged into BasicSR and this repo is a mirror of BasicSR.
https://github.com/xinntao/BasicSR
1.48k stars 320 forks source link

about DCNv2 compiling #99

Open dawnlh opened 4 years ago

dawnlh commented 4 years ago

In the Dependencies and Installation part, you mentioned that:

Deformable Convolution. We use Charles Shang's DCNv2 implementation. Please first compile it.

cd ./codes/models/modules/DCNv2
bash make.sh

but there is no "modules" directory in this path, so do you mean I should download it from 'Charles Shang's DCNv2' repository and put it in this path then compile it myself? (if I run the code directly, I will encounter the error--"Failed to import DCNv2 module")

Thanks a lot

townblack commented 4 years ago

In the Dependencies and Installation part, you mentioned that:

Deformable Convolution. We use Charles Shang's DCNv2 implementation. Please first compile it.

cd ./codes/models/modules/DCNv2
bash make.sh

but there is no "modules" directory in this path, so do you mean I should download it from 'Charles Shang's DCNv2' repository and put it in this path then compile it myself? (if I run the code directly, I will encounter the error--"Failed to import DCNv2 module")

Thanks a lot

Hi, you can change to the 'dev' branch, then you will find "modules" directory.

xinntao commented 4 years ago

@dawnlh We have updated the repo to use dcn in mmdetection. But we did not update the readme, so it raised the confusion. Now we have updated the readme.

dawnlh commented 4 years ago

@xinntao Is there any difference between using DCNv2 and DCN? DCNv2 performs better？

dawnlh commented 4 years ago

@townblack oh, I got it, thanks!