Giters

zhaofuq / Instant-NSR

Human Performance Modeling and Rendering via Neural Animated Mesh (Published in SIGGRAPH Asia 2022)

Home Page:https://zhaofuq.github.io/NeuralAM/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The running speed is relatively slow

qiuyuenheng opened this issue · comments

commented

I used the following command to train the model, but the speed remained at 1-2 it/s:
python train_nerf.py "${INPUTS}/dance" --workspace "${WORKSPACE}" --downscale 1 --network sdf

then, i try to use:
python train_nerf.py "${INPUTS}/dance" --workspace "${WORKSPACE}" --downscale 1 --network enc
The speed has increased, but it can only be maintained at 6-9 it/s

when i try：
python train_nerf.py "${INPUTS}/dance" --workspace "${WORKSPACE}" --downscale 1 --network enc --cuda_ray True
The speed is very fast, reaching 50it/s, but after training for a period of time, the loss becomes nan.

It seems that you mentioned in the code that the current project does not support cuda_ ray. So, the third method seems to be beyond your scope of use. So, if only the second method is used, is it reasonable that the training speed on my device can only reach 6-9it/s?