IndexError: too many indices for tensor of dimension 4

[shi3z]

Hi!

I tried your code. but error occurred.
Could you help me?
Please give me advice.

$ python train_stage1.py 2023-12-27 09:32:44,487 INFO [human_loader.py:133] Using local data in /home/shi3z/git/GPS-Gaussian/render_data/rectified_local/train ... 2023-12-27 09:32:44,583 INFO [human_loader.py:133] Using local data in /home/shi3z/git/GPS-Gaussian/render_data/rectified_local/val ... 0%| | 0/40000 [00:14<?, ?it/s] Traceback (most recent call last): File "/home/shi3z/git/GPS-Gaussian/train_stage1.py", line 175, in <module> trainer.train() File "/home/shi3z/git/GPS-Gaussian/train_stage1.py", line 56, in train _, flow_loss, metrics = self.model(data) File "/home/shi3z/.pyenv/versions/miniconda3-4.7.12/envs/gps_gaussian/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, **kwargs) File "/home/shi3z/git/GPS-Gaussian/lib/network.py", line 37, in forward flow_loss, metrics = sequence_loss(flow_predictions, flow, valid) File "/home/shi3z/git/GPS-Gaussian/lib/loss.py", line 15, in sequence_loss assert not torch.isinf(flow_gt[valid.bool()]).any() IndexError: too many indices for tensor of dimension 4

[ShunyuanZheng]

Sorry, there exists a bug in the early version of human_loader.py . Please pull the latest repository, or rectify the read_img function in human_loader.py's L88-L90.

Also, we re-upload the real-world test data in netdisk since some masks have 4 channels in the previous zip file which could arouse another error using the modified dataloader. We post-processed those masks to ensure all masks have 3 channels.