For the DTDB inpainting dataset, we take one example from each of the 45 categories.
Only 48 frames are kept.
Images are cropped and padded to 432x240 similarly to other video inpainting datasets.

