RuntimeError: indices should be either on cpu or on the same device as the indexed tensor (cpu)

[Chris004869]

Hi,

I encountered an issue when running the following command:
python train.py -s $gs_data_path/m4c_processed -m output/m4c_processed --train_stage 1
The error message is:
RuntimeError: indices should be either on cpu or on the same device as the indexed tensor (cpu)
Could someone please help me understand what might be causing this error? My level of expertise is somewhat limited, and I’m not quite sure how to resolve this issue. Any guidance would be greatly appreciated.

Thank you!

[Suanfacayou]

Hello，

Modify the last line of error code and move all related variables to run on the GPU (using the statement tensor=tensor. to ('cpu '), if unsure, you can move all variables in the last error code to run on the GPU using this statement.)
I forgot the specific question, and if it's convenient, you can upload the entire error message.

Hope to help you.

[Chris004869]

Hello，

Modify the last line of error code and move all related variables to run on the GPU (using the statement tensor=tensor. to ('cpu '), if unsure, you can move all variables in the last error code to run on the GPU using this statement.) I forgot the specific question, and if it's convenient, you can upload the entire error message.

Hope to help you.

Hello,

Thank you for your help. Here is the complete error message I encountered:
Optimizing output/m4c_processed
Output folder: output/m4c_processed [25/07 06:33:29]
loading smpl data /data_share/gaoyuan/GaussianAvatar/gs_data/data/m4c_processed/train/smpl_parms.pth [25/07 06:33:29]
total pose length 110 [25/07 06:33:29]
Traceback (most recent call last):
File "/data_share/gaoyuan/GaussianAvatar/train.py", line 179, in
train(lp.extract(args), np.extract(args), op.extract(args), args.save_epochs, args.checkpoint_epochs)
File "/data_share/gaoyuan/GaussianAvatar/train.py", line 24, in train
avatarmodel = AvatarModel(model, net, opt, train=True)
File "/data_share/gaoyuan/GaussianAvatar/model/avatar_model.py", line 76, in init
query_points = query_map[valid_idx, :].cuda().contiguous()
RuntimeError: indices should be either on cpu or on the same device as the indexed tensor (cpu)
I appreciate your assistance.

[Suanfacayou]

Hello,

I think just modify the last line of your error message.
Change
“'query_points = query_map[valid_idx, :].cuda().contiguous()'”
to
“‘query_map = torch.from_numpy(np.load(query_map_path)['posmap' + str(self.model_parms.query_posmap_size)]).reshape(-1,3)
query_map = query_map.to('cuda')
query_points = query_map[valid_idx, :].contiguous()’”
You just need to compare your code and add the missing parts.

Hoping This Helps

[suzoosuagr]

The valid_idx was load as GPU tensor at previous step. At the error point, query_map is a cpu tensor. To fix it. You can modified to

query_points = query_map[valid_idx.cpu(), :].cuda().contiguous()

Good Luck

[nagexiaochengzi]

Thank you for your excellent work, but I have a new question:
Traceback (most recent call last):
File "/home/jkx/projects/mjz/GaussianAvatar/train.py", line 179, in
train(lp.extract(args), np.extract(args), op.extract(args), args.save_epochs, args.checkpoint_epochs)
File "/home/jkx/projects/mjz/GaussianAvatar/train.py", line 24, in train
avatarmodel = AvatarModel(model, net, opt, train=True)
File "/home/jkx/projects/mjz/GaussianAvatar/model/avatar_model.py", line 90, in init
self.query_lbs = query_lbs[valid_idx, :][None].expand(self.batch_size, -1, -1).cuda().contiguous()
RuntimeError: indices should be either on cpu or on the same device as the indexed tensor (cpu)