Open zhouilu opened 2 weeks ago
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
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.
rots = torch.nan_to_num(rot, nan=1.0)
My gpu is L20(cuda11.8), do you have any good suggestions?
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=0and 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?