Skip to content

Latest commit

 

History

History

semantic_segmentation

README.md

Usage

Please refer to get_started.md for installation and dataset preparation.

Training

Wave-ViT

# Wave-ViT-S
tools/dist_train.sh configs/wavevit/upernet_wavevit_s_512x512_160k_ade20k.py 8 --options model.pretrained=Pretrained/wavevit_s.pth

# Wave-ViT-B
tools/dist_train.sh configs/wavevit/upernet_wavevit_b_512x512_160k_ade20k.py 8 --options model.pretrained=Pretrained/wavevit_b.pth

Wave-ViT pre-trained models for semantic segmentation can be downloaded here

Access code for Baidu is nets

Dual-ViT

# Dual-ViT-S
tools/dist_train.sh configs/dualvit/upernet_dualvit_s_512x512_160k_ade20k.py 8 --options model.pretrained=Pretrained/dualvit_s.pth

# Dual-ViT-B
tools/dist_train.sh configs/dualvit/upernet_dualvit_b_512x512_160k_ade20k.py 8 --options model.pretrained=Pretrained/dualvit_b.pth

Dual-ViT pre-trained models for semantic segmentation can be downloaded here

Access code for Baidu is nets

Results and Models

Wave-ViT Results on ADE20K

Method Backbone mIoU config file
UPerNet Wave-ViT-S 49.6 config
UPerNet Wave-ViT-B 51.5 config

Dual-ViT Results on ADE20K

Method Backbone mIoU config file
UPerNet Dual-ViT-S 49.2 config
UPerNet Dual-ViT-B 51.7 config