megvii-research / aaai2023-pvd Goto Github PK

Hi, thanks for your excellent work, it really benefits me a lot.
I have a question about self.clip_loss() which is used in train_step() function located in file distill_mutual/utils.py. I can not find where it is defined. Could you please tell me where it is defined and how it works?

Thank you in advance. ^_^

你好，请问这个算法在数据格式之间的转化都是基于模型保存文件为.pth吗

Sorry for bothering again ,
does this project support multiple GPU training ?
if not , what arguments can I modify to avoid OOM ?
(I just find --max_ray_batch in main_just_train_tea.py) ,
thanks a lot !

Hi, my test results on the synthetic dataset are good. But I tested on the official llff dataset and found that the effect is not good, in the llff_data/flower directory, my result is only 2.93.

Hello , first thanks for your great team work ! I wonder if it supports custom dataset in form of colmap or other format . Also I see that your team plans to release some guide in AAAI2023-PVD/tools/details.md , is it on the fly ?