Fix minor bugs

[Ilya-Fradlin]

Pull Request Description:

Changes Made:

1. Fix in engine.py:

   • Resolved an issue in engine.py where evaluating the model led to a RuntimeError due to a device mismatch in the line sample_feats = feats[sample_mask]. This error occurred because the indices should be on the same device as the indexed tensor. The fix ensures proper handling of device compatibility.

2. Modification to lr_drop Parameter:

   • Adjusted the lr_drop parameter to accept a list of integers, addressing a discrepancy in the original formulation that expected only a single integer.

3. Correction in train_multi_scannet40.sh:

   • Added the necessary '\' to train_multi_scannet40.sh, without it given parameters were being ignored.

Additional Notes:

• This pull request has been tested locally and addresses the mentioned issues effectively.
• In the first bullet point, another solution can be completely removing the line sample_feats = feats[sample_mask] as sample_feats is not being used later.
• Please review and feel free to adjust if needed.

[ywyue]

Thanks @Ilya-Fradlin for this pull request!

Notes: For 1, it is interesting because it didn't lead to RuntimeError in my machine. For 2, I only use one step lr decay in my training though. This change makes the parameter format compatible with MultiStepLR. For 3, thanks for catching this typo.

Merged :)