search

JingyunLiang / RVRT

Recurrent Video Restoration Transformer with Guided Deformable Attention (NeurlPS2022, official repository)
https://arxiv.org/abs/2206.02146
Other
354 stars 33 forks source link

why flows need flip in x-y dim? #8

Open yanglibo0512 opened 1 year ago

yanglibo0512 commented 1 year ago

code : offset1 = offset1 + flows[0].flip(2).repeat(1, 1, offset1.size(2) // 2, 1, 1) why flows need flip ?