spla-tam
CUDA OUT OF MEMORY
Thank you very much for sharing your excellent work!
I am trying to run splatam on my computer with GeForce RTX 4060 （8 GB). I tried on Repica/office0 dataset by command python scripts/splatam.py configs/replica/splatam.py. The code could run for a while but got the error "CUDA OUT OF MEMORY" at "Tracking Time Step: 1050".
I have tried to decrease the number of "num_frames" and the code could run well. However, I want to reconstruct the whole scene. I wonder if I could change other parameters to run the code on a complete sequence, or is my graphics card configuration unable to support splatam to run a complete Replica sequence?
Looking forward to your reply!
Same error (My setup is RTX 4070 8GB vRAM). I cannot understand how they rendered the Replica Dasaset using 3080 Ti. (They did not mention the vRAM. Maybe it should be higher.)
I will try to decrease the number of frames like you, but I have a question. Is there any way to render all the datasets separating half by half?
Hi, the 3080Ti has 12 GB of vRAM.
If you are running into CUDA OOM issues, you can run the reconstruction at a lower resolution. For example, see the following lines of this config: https://github.com/spla-tam/SplaTAM/blob/da6bbcd24c248dc884ac7f49d62e91b841b26ccc/configs/replica/splatam_s.py#L52
