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.
I have not tried to add BN layers only to the alignment module.
What I have tried is to add BN layers to all the model (including the alignment and the reconstruction modules). And the results are not as good as that without BN.
You may have a try. From my experience, I think it will only bring marginal improvement if it improves.
I have not tried to add BN layers only to the alignment module. What I have tried is to add BN layers to all the model (including the alignment and the reconstruction modules). And the results are not as good as that without BN.
You may have a try. From my experience, I think it will only bring marginal improvement if it improves.