Difference between nerfstudio's instant-ngp and the original implementation #2395
Unanswered
balintfodor
asked this question in
Q&A
Replies: 1 comment
-
Also want to know |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
When I played on my dataset I noticed that the original instant-ngp code runs faster and produces better results (less foggy, maybe a bit a sharper) with the default parameters than nerfstudio's instant-ngp implementation which I believe has similar default parameters (?), except the number of steps 30k vs 35k.
I tried to reduce the extra evaluation computation to get a better performance for the nerfstudio's implementation. I also tried to spot any subtle differences in the parameters without any luck. Used
images_on_gpu
too. I'm with a T4 (16GB) GPU, as I understand this is not the most ideal card for these purposes (in the original instant-ngp paper they use an RTX 3090) but that's what I have right now.Can you point me to directions about this performance difference? Either parameters, code lines, or some explanation would be super helpful.
Beta Was this translation helpful? Give feedback.
All reactions