Hi,
I read the code in projection.py. However, I wonder why we need to normalize the pixels after reprojecting 3D points to the image plane.
def normalize(self, pixel_locations, h, w):
resize_factor = torch.tensor([w-1., h-1.]).to(pixel_locations.device)[None, None, :]
normalized_pixel_locations = 2 * pixel_locations / resize_factor - 1. # [n_views, n_points, 2]
return normalized_pixel_locations
Hi,
I read the code in projection.py. However, I wonder why we need to normalize the pixels after reprojecting 3D points to the image plane.