Question about modulated deform convolution using

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

Question about modulated deform convolution using #93

Open llp1996 opened 4 years ago

llp1996 commented 4 years ago

In paper "we ignore the modulation _delta m_k " .Is it meaning that use deform conv as dcn_v1? In code ,i find the mask is used in ModulatedDeformConv. thx~

TouqeerAhmad commented 4 years ago

I think they are omitting it in the paper to make the description less cumbersome -- in code it is in fact using DCNv2.

xinntao commented 4 years ago

Thanks @TouqeerAhmad In the paper, we omit it for a simpler description For simplicity, we only consider learnable offsets ∆pk and ignore modulation ∆mk in the descriptions and figures. We actually use dcnv2, the modulated deformable conv.