search

skhu101 / SHERF

Code for our ICCV'2023 paper "SHERF: Generalizable Human NeRF from a Single Image"
Other
297 stars 10 forks source link

Training Speed #39

Closed Lewis-lv0 closed 2 months ago

Lewis-lv0 commented 2 months ago

Hi @skhu101 , thanks for your great work! I am trying to train the model on RenderPeople dataset but I found the training will take ~4 days. May I ask what can be the main reason for the training to take such a long time? Here are my guesses: rendering the whole image before calculating the training loss; global feature extractor (StyleGAN); transformer-based feature fusion module; large training steps.

I would appreciate if I can hear your optinion on this question. Many thanks :)

skhu101 commented 2 months ago

Hi, thanks for your interest in our work. Based on my observation, the training converges for about 1~2 day. Longer training makes tiny differences in the final performance.