gaussian-splatting icon indicating copy to clipboard operation
gaussian-splatting copied to clipboard

Published 20 hours ago • verified publisher icon graphdeco-inria

Training Time Breakdown

Open AmeYatoro opened this issue 9 months ago • 0 comments

In the paper, the author mentioned that "The majority (∼80%) of our training time is spent in Python code, since...". However, I did a runtime breakdown on training of 3DGS using torch.cuda.Event(), and it turns out that most of the time is spent on the backward propagation(which is implemented by CUDA), can anyone please explain if there is a misunderstanding?

AmeYatoro avatar May 09 '24 13:05 AmeYatoro