I found that datasets/scannet_dataset.py and interior_dataset.py did not use CROP. Why is CROP used only in NYU dataset?