Open
Description
Despite solving similar problems, I still have Illegal memory access CUDA issues.
Train scene is garden, command is python3 train.py -s data/garden/ -m garden -r 4 --data_device=cpu --sh_degree=0
and add rots = torch.nan_to_num(rot, nan=1.0) before render. I am sure that the input data of render does not have nan and type is torch.float32.
My gpu is L20(cuda11.8), do you have any good suggestions?
Metadata
Metadata
Assignees
Labels
No labels