Question about the memory bottleneck

Haochen-Wang409 / U2PL

[CVPR'22] Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels

Apache License 2.0

426 stars 59 forks source link

Question about the memory bottleneck #136

Closed Hugo-cell111 closed 1 year ago

Hugo-cell111 commented 1 year ago

Hi! I just set batch size as 1 and the memory comes to around 16 GB. I wonder where is the memory bottleneck and which part brings so much memory burden? Thanks!

Haochen-Wang409 commented 1 year ago

Which dataset did you use? It might be the input resolution.

Hugo-cell111 commented 1 year ago

I use Cityscapes and the input resolution is set as 769. Have you done the ablation study of memory bank and seen whether there exists difference of memory cost?