HDMapNet
HDMapNet copied to clipboard
offset in perspective
Hello, great work!
According to my understanding, we project coordinate from camera to ego and bev mask to ego: pix_coords = proj_mat @ cam_coords to mask: pix_coords[:, 0] += offset[0] / 2 pix_coords[:, 2] -= offset[1] / 8 pix_coords = torch.stack([pix_coords[:, 2], pix_coords[:, 0]], axis=1)
But, I am confused about -offset[1]/8, why not +offset[1]/2. Looking forward to your reply.
Thanks in advance!