Generating HO3D dataset

[alakhag]

If I directly use the poses and segmentations from HO3D dataset, and generate point clouds close to the 3D model, how are the results expected? Because when I do that, although the geometry is better (because of better segmentation), but the texture takes a huge toll in the training result. Any reason why could that be, and how can I improve this?