Weakly Supervised Segmentation on MS COCO
46.7APDETR
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DETRIoU threshold=0.22025.07 | 46.7 | 52.6 | 53.9 | 64.4 | 76.2 | 68.2 | 30.5 | |
| CLIP SimilarityIoU threshold=0.22025.07 | 24.8 | 37 | 28.2 | 42 | 42.2 | 62.5 | 15.7 | |
| Cross-Modal-SimVariant=SPARC, TopK=Global, Backbone=CLIP, IoU threshold=0.22025.07 | 22.3 | 28.3 | 26.2 | 36.9 | 31.9 | 56.9 | 13.8 | |
| Cross-Modal-SimVariant=SPARC, TopK=Global, Backbone=DINO, IoU threshold=0.22025.07 | 22.2 | 25.3 | 27.4 | 37.3 | 29.7 | 58.1 | 14.3 | |
| Concept-Latent-SumVariant=SPARC, TopK=Global, Backbone=DINO, IoU threshold=0.22025.07 | 22.2 | 24.4 | 26.4 | 35.2 | 28.4 | 51.6 | 13.7 | |
| Concept-Latent-SumVariant=SPARC, TopK=Global, Backbone=CLIP, IoU threshold=0.22025.07 | 22.2 | 27.6 | 25.4 | 34.7 | 31 | 49.3 | 13.1 | |
| Concept-Latent-SumVariant=SPARC, TopK=Local, Backbone=DINO, IoU threshold=0.22025.07 | 21.2 | 24.3 | 25.1 | 34.9 | 28.3 | 51.8 | 13.6 | |
| Concept-Latent-SumVariant=SPARC, TopK=Local, Backbone=CLIP, IoU threshold=0.22025.07 | 20.3 | 23.8 | 23.4 | 32.5 | 25.8 | 47.4 | 12.3 | |
| Cross-Modal-SimVariant=SPARC, TopK=Local, Backbone=DINO, IoU threshold=0.22025.07 | 19.4 | 22.1 | 24.2 | 34.3 | 26.5 | 54.6 | 12.9 | |
| Cross-Modal-SimVariant=SPARC, TopK=Local, Backbone=CLIP, IoU threshold=0.22025.07 | 17.6 | 24.2 | 20.4 | 30.4 | 26.6 | 47.2 | 10.2 | |
| Cross-Modal-SimVariant=USAE, Backbone=DINO, IoU threshold=0.22025.07 | 12.6 | 11.8 | 16.6 | 25.3 | 13.9 | 41.5 | 10 | |
| Concept-Latent-SumVariant=USAE, Backbone=DINO, IoU threshold=0.22025.07 | 10.8 | 8.9 | 14 | 22.4 | 10.5 | 36.4 | 9.6 | |
| Concept-Latent-SumVariant=USAE, Backbone=CLIP, IoU threshold=0.22025.07 | 0.6 | 2.7 | 0.7 | 2.7 | 2.8 | 3.6 | 3.5 | |
| Cross-Modal-SimVariant=USAE, Backbone=CLIP, IoU threshold=0.22025.07 | 0.4 | 1.9 | 0.4 | 2.2 | 2.1 | 2.9 | 2.3 |