Skip to content

Domain Generalization in Semantic Segmentation using CLIP Style Prompts

Notifications You must be signed in to change notification settings

junghyeon0427/CLIP-Style-Prompts-in-DG

Repository files navigation

CLIP Style Prompts in DG

Domain Generalization in Semantic Segmentation using CLIP Style Prompts
Junghyeon Seo, Sungsik Kim, Seungheon Song, Jaekoo Lee, Korea Computer Congress 2024 (KCC 2024)

Paper Link

Method

image

Experimental Results (mIoU)

Method Cityscapes Night Rain Snow Fog
PØDA (Cityscapes → Night) 63.85 24.58 41.54 44.34 50.93
PØDA (Cityscapes → Rain) 63.87 22.92 40.78 43.49 50.41
PØDA (Cityscapes → Snow) 64.31 21.38 40.21 43.69 50.43
PØDA (Cityscapes → Fog) 63.25 22.44 38.04 41.31 48.44
Ours 64.78 23.26 43.60 44.69 52.44

Semantic Map Result

image

Reference Paper

We refer on the following paper
PromptStyler (ICCV 2023), PØDA (ICCV 2023)

About

Domain Generalization in Semantic Segmentation using CLIP Style Prompts

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published