search

SHI-Labs / OneFormer

OneFormer: One Transformer to Rule Universal Image Segmentation, arxiv 2022 / CVPR 2023
https://praeclarumjj3.github.io/oneformer
MIT License
1.51k stars 135 forks source link

Cityscapes val SOTA results not found in the paper #112

Closed eldar closed 10 months ago

eldar commented 10 months ago

Dear all,

Thanks for your excellent paper and provided code. It is currently #1 on Cityscapes val instance segmentation leaderboard with AP 50.6 https://paperswithcode.com/sota/instance-segmentation-on-cityscapes-val however I could not find this result in the paper. On GitHub the best model seems to reach 48.9 (https://github.com/SHI-Labs/OneFormer#cityscapes) - 1.5% below than the number on PapersWithCode. Is there an updated config for the SOTA result reaching 50.6 AP?

praeclarumjj3 commented 10 months ago

Hi @eldar, thanks for your interest.

The model can be found in the readme on the internimage branch: https://github.com/SHI-Labs/OneFormer/blob/internimage/README.md#cityscapes

eldar commented 10 months ago

Thanks so much for the quick response! This is resolved then.