Closed zcong17huang closed 3 years ago
The disparity map is relative to the left image (i.e., left x coordinate - right x coordinate = disp (positive)), so to find the correponding pixel on the right image, we should do x - disp
.
The disparity map is relative to the left image (i.e., left x coordinate - right x coordinate = disp (positive)), so to find the correponding pixel on the right image, we should do
x - disp
.
Yes, you are right. I have found my mistake. Thank you very much.
Hello @haofeixu . Thank you for the paper and the code. In your code https://github.com/haofeixu/aanet/blob/master/nets/warp.py line 55
offset = torch.cat((-disp, torch.zeros_like(disp)), dim=1) # [B, 2, H, W]
Here you warp the right img to match the left, so the right one should do right shift operation. So I think here should bedisp
, not-disp
. Is it right?