Video Object Segmentation on MOSE (val)
77.9J&F ScoreSAM 2 (Hiera-L)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SAM 2 (Hiera-L)Backbone=Hiera-L, Prompting protocol=first-frame ground-truth mask2024.08 | 77.9 | — | — | |
| G2TAM2026.07 | 77.8 | — | — | |
| SAM 2 (Hiera-B+)Backbone=Hiera-B+, Prompting protocol=first-frame ground-truth mask2024.08 | 76.6 | — | — | |
| SAM 2.1-B+2025.01 | 76.6 | — | — | |
| SAM 2-B+2025.01 | 75.8 | — | — | |
| SAM 22026.07 | 75.2 | — | — | |
| OriginalMemory interval (delta_t)=52026.02 | 74.7 | — | — | |
| SMRIE (Image Encoder Optimization)=false, MA (Memory Attention Optimization)=true, Speedup=1.78x, Memory interval (delta_t)=52026.02 | 74.6 | — | — | |
| SAM 2++2025.10 | 74.6 | 70.6 | 78.6 | |
| OriginalMemory interval (delta_t)=12026.02 | 74.5 | — | — | |
| SMR-uniformIE (Image Encoder Optimization)=false, MA (Memory Attention Optimization)=true, Speedup=1.76x, Memory interval (delta_t)=52026.02 | 74.5 | — | — | |
| SMR-randomIE (Image Encoder Optimization)=false, MA (Memory Attention Optimization)=true, Speedup=1.75x, Memory interval (delta_t)=52026.02 | 74.4 | — | — | |
| SMRIE (Image Encoder Optimization)=false, MA (Memory Attention Optimization)=true, Speedup=1.78x, Memory interval (delta_t)=12026.02 | 74.2 | — | — | |
| SMR-uniformIE (Image Encoder Optimization)=false, MA (Memory Attention Optimization)=true, Speedup=1.76x, Memory interval (delta_t)=12026.02 | 73.9 | — | — | |
| SMR-randomIE (Image Encoder Optimization)=false, MA (Memory Attention Optimization)=true, Speedup=1.75x, Memory interval (delta_t)=12026.02 | 73.8 | — | — | |
| SWRIE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=false, Speedup=1.83x, Memory interval (delta_t)=12026.02 | 73.4 | — | — | |
| SWRIE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=false, Speedup=1.83x, Memory interval (delta_t)=52026.02 | 73.4 | — | — | |
| Efficient-SAM2IE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=true, Speedup=1.80x, Memory interval (delta_t)=52026.02 | 73.1 | — | — | |
| Efficient-SAM2IE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=true, Speedup=1.80x, Memory interval (delta_t)=12026.02 | 72.6 | — | — | |
| Cutie-base+Backbone=base+, Prompting protocol=first-frame ground-truth mask2024.08 | 71.7 | — | — | |
| Cutie-base+2025.01 | 71.7 | — | — | |
| Cutie-base+Variant=base+2026.07 | 71.7 | — | — | |
| MemPoolIE (Image Encoder Optimization)=false, MA (Memory Attention Optimization)=true, Speedup=2.04x, Memory interval (delta_t)=12026.02 | 71.3 | — | — | |
| EdgeTAM2025.01 | 70 | — | — | |
| Cutie-baseBackbone=base, Prompting protocol=first-frame ground-truth mask2024.08 | 69.9 | — | — | |
| Cutie-base2025.01 | 69.9 | — | — | |
| Cutie-baseVariant=base2026.07 | 69.9 | — | — | |
| MemPoolIE (Image Encoder Optimization)=false, MA (Memory Attention Optimization)=true, Speedup=2.04x, Memory interval (delta_t)=52026.02 | 69.8 | — | — | |
| JointFormer2025.10 | 69.7 | 65.8 | 73.6 | |
| ALGMIE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=false, Speedup=1.76x, Memory interval (delta_t)=12026.02 | 67 | — | — | |
| DEVA2025.10 | 66.4 | 62.2 | 70.6 | |
| CutieVariant=base+2025.10 | 66.2 | 62.3 | 70.1 | |
| ALGMIE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=false, Speedup=1.76x, Memory interval (delta_t)=52026.02 | 66.1 | — | — | |
| ToMe4DMIE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=false, Speedup=1.14x, Memory interval (delta_t)=12026.02 | 66.1 | — | — | |
| DEVAPrompting protocol=first-frame ground-truth mask2024.08 | 66 | — | — | |
| DEVA2025.01 | 66 | — | — | |
| DEVA2026.07 | 66 | — | — | |
| ToMe4DMIE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=false, Speedup=1.14x, Memory interval (delta_t)=52026.02 | 65.2 | — | — | |
| OneVOSFine-tuned on=MOSE2025.10 | 62.2 | 57.9 | 66.6 | |
| DeAOTBackbone=SwinB2025.10 | 61.7 | 57.6 | 65.9 | |
| AOTBackbone=SwinB2025.10 | 60.2 | 56.2 | 64.1 | |
| SwinB-DeAOTBackbone=SwinB, Prompting protocol=first-frame ground-truth mask2024.08 | 59.9 | — | — | |
| SwinB-DeAOT2025.01 | 59.9 | — | — | |
| SwinB-DeAOTBackbone=SwinB2026.07 | 59.9 | — | — | |
| XMemPrompting protocol=first-frame ground-truth mask2024.08 | 59.6 | — | — | |
| XMem2025.01 | 59.6 | — | — | |
| XMem2026.07 | 59.6 | — | — | |
| SwinB-AOTBackbone=SwinB, Prompting protocol=first-frame ground-truth mask2024.08 | 59.4 | — | — | |
| SwinB-AOT2025.01 | 59.4 | — | — | |
| SwinB-AOTBackbone=SwinB2026.07 | 59.4 | — | — | |
| XMemMemory-based=true2023.12 | 57.6 | 53.3 | 62 | |
| XMem2025.10 | 56.3 | 52.1 | 60.5 | |
| GLEE-LiteMemory-based=false2023.12 | 56.1 | 51.8 | 60.4 | |
| STCNPrompting protocol=first-frame ground-truth mask2024.08 | 52.5 | — | — | |
| STCN2025.01 | 52.5 | — | — | |
| STCN2026.07 | 52.5 | — | — | |
| SWEMMemory-based=true2023.12 | 50.9 | 46.8 | 64.9 | |
| STCNMemory-based=true2023.12 | 50.8 | 46.6 | 55 | |
| STCN2025.10 | 50.1 | 46.1 | 54.1 | |
| RDEPrompting protocol=first-frame ground-truth mask2024.08 | 46.8 | — | — | |
| RDE2025.01 | 46.8 | — | — | |
| RDE2026.07 | 46.8 | — | — | |
| ToMeIE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=false, Speedup=1.43x, Memory interval (delta_t)=12026.02 | 44.7 | — | — | |
| ToMeIE (Image Encoder Optimization)=true, MA (Memory Attention Optimization)=false, Speedup=1.43x, Memory interval (delta_t)=52026.02 | 42.2 | — | — |