Referential Audio-Visual Segmentation on Ref-AVS (seen)
0.688J & F ScorePRIMED
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PRIMEDTask=Ours2026.05 | 0.688 | — | 0.715 | 0.66 | — | — | |
| AURORATask=Ref-AVS2026.05 | 0.68 | — | 0.728 | 0.632 | — | — | |
| Ours AuralSAM2Backbone=Hiera-l2025.06 | 0.5868 | — | — | — | 56.16 | 61.19 | |
| Ours AuralSAM2Backbone=Hiera-b+2025.06 | 0.56 | — | — | — | 53.16 | 58.83 | |
| OISA-1BTask=Omni-AVS2026.05 | 0.552 | — | 0.587 | 0.517 | — | — | |
| TGS-AgentTask=Ref-AVS2026.05 | 0.549 | — | 0.604 | 0.495 | — | — | |
| Ours SAM (w/ AuralFuser)Backbone=ViT-h2025.06 | 0.5427 | — | — | — | 48.26 | 60.28 | |
| SAMA-AVS†Backbone=Hiera-b+, reimplementation=true2025.06 | 0.531 | — | — | — | 49.5 | 56.7 | |
| GAVS†Backbone=Hiera-b+, reimplementation=true2025.06 | 0.513 | — | — | — | 48 | 54.6 | |
| VoCaTask=Ref-AVS2026.05 | 0.509 | — | 0.608 | 0.409 | — | — | |
| TSAMTask=Ref-AVS2026.05 | 0.501 | — | 0.568 | 0.434 | — | — | |
| TSAMBackbone=ViT-h2025.06 | 0.501 | — | — | — | 43.4 | 56.8 | |
| CrabTask=Ref-AVS2026.05 | 0.493 | — | 0.58 | 0.405 | — | — | |
| SAM2-LOVETask=Ref-AVS2026.05 | 0.477 | — | 0.519 | 0.435 | — | — | |
| SAMA-AVSBackbone=ViT-h2025.06 | 0.477 | — | — | — | 39.2 | 56.2 | |
| SAM2-LOVEBackbone=Hiera-l2025.06 | 0.477 | — | — | — | 43.5 | 51.9 | |
| Omni-R1Task=Omni-Reasoning2026.05 | 0.472 | — | 0.514 | 0.43 | — | — | |
| EEMCTask=Ref-AVS2026.05 | 0.428 | — | 0.513 | 0.342 | — | — | |
| EEMCBackbone=Swin-b2025.06 | 0.428 | — | — | — | 34.2 | 51.3 | |
| ReferFormerTask=Ref-VOS*2026.05 | 0.407 | — | 0.501 | 0.313 | — | — | |
| AVSegFormerTask=AVS*2026.05 | 0.402 | — | 0.47 | 0.335 | — | — | |
| AVSegFormerBackbone=PVT-v22025.06 | 0.402 | — | — | — | 33.47 | 47 | |
| GAVSTask=AVS*2026.05 | 0.394 | — | 0.498 | 0.289 | — | — | |
| GAVSBackbone=ViT-h2025.06 | 0.3935 | — | — | — | 28.9 | 49.8 | |
| AVSBenchTask=AVS*2026.05 | 0.372 | — | 0.511 | 0.232 | — | — | |
| TPAVIBackbone=PVT-v22025.06 | 0.372 | — | — | — | 23.2 | 51.1 | |
| GroundedSAM2*Backbone=Hiera-b+, utilising SAM's zero-shot capability=true2025.06 | 0.342 | — | — | — | 28.5 | 39.9 | |
| R2VOSTask=Ref-VOS*2026.05 | 0.33 | — | 0.41 | 0.25 | — | — | |
| Crab2025.03 | — | 0.4054 | 0.58 | — | — | — |