Unsupervised Open-vocabulary Semantic Segmentation on Pascal VOC 2012 (val)
87.9mIoUFreeDA (ViT-L)
Evaluation Results
| Method | Links | |
|---|---|---|
| FreeDA (ViT-L)PAMR=false, Training Dataset=COCO Captions, Total Parameters (M)=732, Trainable Parameters (M)=0, Similarity=Visual2024.04 | 87.9 | |
| FreeDA (ViT-B)PAMR=false, Training Dataset=COCO Captions, Total Parameters (M)=236.1, Trainable Parameters (M)=0, Similarity=Visual2024.04 | 85.6 | |
| TCLPAMR=true, Training Dataset=CC3M+CC12M, Total Parameters (M)=178.3, Trainable Parameters (M)=21.7, Similarity=Visual2024.04 | 83.2 | |
| OVDiffPAMR=false, Total Parameters (M)=1226.4, Trainable Parameters (M)=0, Similarity=Visual2024.04 | 81.7 | |
| GroupViTPAMR=true, Training Dataset=CC12M+YFCC, Total Parameters (M)=55.8, Trainable Parameters (M)=55.8, Similarity=Visual2024.04 | 81.5 | |
| GroupViTPAMR=false, Training Dataset=CC12M+RedCaps, Total Parameters (M)=55.8, Trainable Parameters (M)=55.8, Similarity=Textual2024.04 | 79.7 | |
| TCLPAMR=false, Training Dataset=CC3M+CC12M, Total Parameters (M)=178.3, Trainable Parameters (M)=21.7, Similarity=Textual2024.04 | 77.5 | |
| MaskCLIPPAMR=false, Total Parameters (M)=291, Trainable Parameters (M)=0, Similarity=Textual2024.04 | 74.9 | |
| MaskCLIPPAMR=true, Total Parameters (M)=291, Trainable Parameters (M)=0, Similarity=Textual2024.04 | 72.1 | |
| ReCoPAMR=true, Training Dataset=ImageNet1k, Total Parameters (M)=313, Trainable Parameters (M)=0, Similarity=Visual2024.04 | 62.4 | |
| ReCoPAMR=false, Training Dataset=ImageNet1k, Total Parameters (M)=313, Trainable Parameters (M)=0, Similarity=Visual2024.04 | 57.7 |