lhoyer / DAFormer

[CVPR22] Official Implementation of DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation
Other
470 stars 93 forks source link

why use cityscapes_half_512x512.py need more GPU memory? #73

Closed zjuxiaobaiq closed 7 months ago

zjuxiaobaiq commented 1 year ago

Hi , when i use cityscapes_half_512x512.py to do target-only experiment,i find it need more GPU memory than uda experiment? i can not figure out ,can you give me some advise?

miladsoltany commented 1 year ago

Hi. I've noticed that running the DAFormer code on an RTX8000 uses up 39gbs while it only uses up 10gbs on RTX6000 and 3090. It's weird but it is what it is =)