Downsampling using Overlapping vs non-overlapping Convolutions

SHI-Labs / Neighborhood-Attention-Transformer

Neighborhood Attention Transformer, arxiv 2022 / CVPR 2023. Dilated Neighborhood Attention Transformer, arxiv 2022

MIT License

1.04k stars 85 forks source link

Downsampling using Overlapping vs non-overlapping Convolutions #4

Closed mmaaz60 closed 2 years ago

mmaaz60 commented 2 years ago

Hi,

Congratulations on the great work. I wanted to ask if you have done an ablation on using Overlapping vs Non-Overlapping convolutions for the downsampling layers in NAT. If so, may you share the results?

Thank you

alihassanijr commented 2 years ago

Hi, Thank you for your interest.

Yes we have, it's in the paper already, but it's with the Swin-T configuration. If you're referring to NAT-Tiny without overlapping convs, I don't believe we have that yet.

alihassanijr commented 2 years ago

I'll close this issue for now due to inactivity, but feel free to reopen it if you have follow-up questions.