MaureenZOU / TSAM

[CVPR2021] Official Implementation for Progressive Temporal Feature Alignment Network for Video Inpainting
90 stars 11 forks source link

Flow Validity Mask calculation #14

Open Huihui1002 opened 3 years ago

Huihui1002 commented 3 years ago

Dear Xueyan,

May I ask how do you calculate the Flow Validity Mask that you described in Section 3.3? Is this part of code in src/model/tsm/temperal_shift.py?

validity

But there is no cycle consistency here.

And how do you calculate the left_flowmask and right_flowmask?

Another question is: In line 109, torch.clamp(right_flowmask + (torch.sum(left_feature, dim=2, keepdim=True)==0).float(), 0, 1).int().float(). Why left_feature is used here?

Many thanks for your help in advance!