Replies: 2 comments
-
By default, Another aspect is that saving the images means encoding them on the CPU. On my machine, this is slower than the actual rendering part. You might benefit from multi-threading that through Python. (Speed-of-light, my 3090 needs ~100ms for a 1440x1080 frame, so the best case for 500 images would be 50 seconds there.) Cheers! |
Beta Was this translation helpful? Give feedback.
-
That helps a lot, thanks!! |
Beta Was this translation helpful? Give feedback.
-
Hi!
Instant-NGP is great, thanks for sharing!
I'm successfully training it in "nerf" mode, using the python binding provided by you in
scripts/run.py
.However, when I use the same code to render 500 images (1440x1080) it takes almost 10 minutes on a 3090 gpu. I disabled the metrics computation to be sure to focus only on rendering.
Is it the expected rendering speed?
Thanks in advance!
Beta Was this translation helpful? Give feedback.
All reactions