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.
For the temporal Attention, I use your pertained params EDVR_REDS_deblur_L.pth to do inference.
And get cor_prob which values are all 1.
It's due to the values in cor_l are very large values. So, after sigmoid it becomes 1.
Is that the right way to compute similarity for vecters?
If so, the value of cor_prob would be small for two vecters share the same value with small norm.
For the temporal Attention, I use your pertained params EDVR_REDS_deblur_L.pth to do inference. And get cor_prob which values are all 1.
It's due to the values in cor_l are very large values. So, after sigmoid it becomes 1.
Is that the right way to compute similarity for vecters? If so, the value of cor_prob would be small for two vecters share the same value with small norm.