LiheYoung / UniMatch

[CVPR 2023] Revisiting Weak-to-Strong Consistency in Semi-Supervised Semantic Segmentation
https://arxiv.org/abs/2208.09910
MIT License
476 stars 59 forks source link

Question about the evaluation time #96

Closed Hugo-cell111 closed 1 year ago

Hugo-cell111 commented 1 year ago

Hi! Why the evaluation process in the first epoch cost so much time, while the speed is normal in the following epochs? The log is here and here. In the first epoch, it takes 91s; in the other epochs, it just takes 19-20s. Thanks!

LiheYoung commented 1 year ago

I am not sure the reason. It may be related to my used disk. But I think the 90s test time is acceptable.

Hugo-cell111 commented 1 year ago

No. When I reproduce UniMatch and other variants(such as UniPerb and DusPerb) on my own server, the same problem occurs to, so I think it might not just relate to the disk personally. Anyway, it is not a big problem since after the 2nd epoch everything goes normal. Thank you!