Domain Generalization in Semantic Segmentation using CLIP Style Prompts
Junghyeon Seo, Sungsik Kim, Seungheon Song, Jaekoo Lee, Korea Computer Congress 2024 (KCC 2024)
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 |
We refer on the following paper
PromptStyler (ICCV 2023), PØDA (ICCV 2023)