Question about the color input.

[linxuewu]

This is a very meaningful job!

1. I would like to ask if all training samples are input with color during the pre-training process?
2. In the zero-shot classification experiment, for ModelNet40, color input was not used. Should these dimensions be deleted or replaced with a fixed value (0.4)?
3. If color is not used as input in pre-training, will there be any impact on downstream tasks?

[junshengzhou]

Hi,

Thanks for your interest in Uni3D!

1. We set a ratio for each sample in a batch to determine whether to use its color or not. Please refer to the dataset file:https://github.com/baaivision/Uni3D/blob/8c7bb7a1daa6ac816451ef107e6d0f0441b213a8/data/datasets.py#L384C1-L384C1
2. These dimensions are replaced with a fixed value 0.4 for inference. This value keeps the same as the one we set in training.
3. We do include samples with color and without color during pre-training, which lead to robust performance in downstream tasks with colored/no color point clouds as input.