fix pixel sampling with masks issue when data is list

[anc2001]

PixelSampler sample_method expects mask argument to be batched but mask is unbatched when the images / masks are given as list. Batches the mask in this case.

[anc2001]

For more context, when training on a dataset with images and masks of different sizes I get the following the following error

  File "/src/external/nerfstudio/nerfstudio/data/pixel_samplers.py", line 222, in collate_image_dataset_batch_list
    all_images.append(batch["image"][i][indices[:, 1], indices[:, 2]])
IndexError: index 2 is out of bounds for dimension 1 with size 2

This is because the expected shape from sample_method is [batch_size, 3] but in this case the returned indices are of shape [batch_size, 2]. Batching the mask fixes this error

[drewgilliam]

@brentyi & @tancik - hoping you could help here since you've merged some of my recent PRs. Is there more information or more changes needed to get this PR merged? Thanks!

[brentyi]

Thanks for the ping @drewgilliam and the fix @anc2001; I just checked and this makes sense to me!

[drewgilliam]

@brentyi - thanks for the assist, much appreciated!