Segmenter-based-on-OpenMMLab

"Segmenter: Transformer for Semantic Segmentation, arxiv 2105.05633." reproduced via mmsegmentation.

We reproduce Segmenter via mmsegmentation
based on official open-sourced code.

Environment

  • python=3.7

  • pytorch=1.7.1

  • torchvision=0.8.2

  • cudatoolkit=10.1

  • mmcv-full=1.3.10

  • mmsegmentation=0.16.0

Note: You should install pytorch with a version higher than 1.7, because the pretrained model of DeiT is saved via 1.7+
pytorch. Otherwise you may encounter some errors while loading the state_dict.

Results on ADE20K

The passwds of download links are all 'nopw'.

Exp Name backbone Our mIoU-SS mIoU in paper Resolution BS Download
4th line in Table3 Seg-B-Linear/16 DeiT-B 46.83 47.10 512x512 8 model config log
4th line in Table6 Seg-B-Mask/16 DeiT-B 48.41 47.67 512x512 8 model config log
6th line in Table3 Seg-B -Linear/16 ViT-B 45.70 45.69 512x512 8 model config log

GitHub

https://github.com/EricKani/Segmenter-Based-on-OpenMMLab