Closed ootts closed 3 years ago
Hi, this function is using the disparity_change
to compute 3D flow. In sceneflow.py
depth1 = torch.from_numpy(intrinsics[0] / disp1).float()
depth2 = torch.from_numpy(intrinsics[0] / disp2).float()
# transformed depth
depth12 = torch.from_numpy(intrinsics[0] / (disp1 + disparity_change)).float()
flow3d = pops.backproject_flow3d(flow2d, depth1, depth12, intrinsics)
depth12 is already represented in the coordinates of the first frame, so no interpolation is necessary.
Hi, how do you fine disparity_change in the above code snippet? Thanks
I have a question about here. When we are generating scene flow using optical flow and depth maps, shouldn't we interpolate on depth1 using x1 and y1, and then backproject using interpolated depths? Thank you!