search

microsoft / CSWin-Transformer

CSWin Transformer: A General Vision Transformer Backbone with Cross-Shaped, CVPR 2022
MIT License
539 stars 78 forks source link

the memory is nearly twice that of swin transformer #23

Open cjt222 opened 2 years ago

cjt222 commented 2 years ago

Hi，Thanks for great jobs, I try to infer a image ， I tried to use the segmentation model by cswin transformer small to test some pictures, and found that the memory is nearly twice that of swin transformer small. Is this normal?

vidit98 commented 2 years ago

Hi, you can try to set use_chk param of CSwin to True for reducing memory footprint. I suppose you are already using checkpointing in Swin model.