search

vt-vl-lab / FGVC

[ECCV 2020] Flow-edge Guided Video Completion
Other
1.55k stars 263 forks source link

without nonlocal frame flow? #13

Closed laoyoutiaotiao closed 3 years ago

laoyoutiaotiao commented 3 years ago

in your code, videoNonLocalFlowF and videoNonLocalFlowB is None? get_flowNN_gradient(args, gradient_x_filled, gradient_y_filled, mask, mask_gradient, videoFlowF, videoFlowB, None, None)

zhangqijun commented 3 years ago

in code comment

videoNonLocalFlowF: imgH x imgW x 2 x 3 x nFrame

# videoNonLocalFlowB: imgH x imgW x 2 x 3 x nFrame

in papaer said For every frame, we compute non-local flow to three additional frames using the homography-aligned method (Equation 2) For simplicity, we always select the first, middle, and last frames of the video as non-local neighbors

Is it mean for frame_i,calculate [first_frame->frame_i,middle_frame->frame_i,last_frame->frame_i] as videoNonLocalFlowF, [frame_i,->first_frame,frame_i->middle_frame,frame_i->last_frame] as videoNonLocalFlowB?

gaochen315 commented 3 years ago

For the beta version, I disable this function for a faster speed. I will update the code with the non-local flow support. Sorry about that.

21pl commented 3 years ago

did you ever manage to update the code for non-local flow support as ' --Nonlocal ' still produces a type error?