Semi-supervised Video Object Segmentation on SAM2 VOS Evaluation Suite (ES, PU, LV, SV) (test)
90.2ES (EntitySeg Score)SAM 2.1
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SAM 2.1Prompt Type=ground-truth mask, Para. (M)=38.9, Cost=2.2m, Note=directly use masks as inputs into VOS model without using SAM2025.06 | 90.2 | 80.9 | 82.2 | 76.5 | — | |
| SAM + XMem++Prompt Type=ground-truth mask, Para. (M)=157, Note=directly use masks as inputs into VOS model without using SAM2025.06 | 90 | 67.9 | 80.7 | 61.5 | — | |
| TinySAM + XMem++Prompt Type=ground-truth mask, Para. (M)=72.3, Note=directly use masks as inputs into VOS model without using SAM2025.06 | 90 | 67.9 | 80.7 | 61.5 | — | |
| SAM + CutiePrompt Type=ground-truth mask, Para. (M)=129.8, Note=directly use masks as inputs into VOS model without using SAM2025.06 | 89.7 | 62.5 | 81.4 | 64.8 | — | |
| TinySAM + CutiePrompt Type=ground-truth mask, Para. (M)=45.1, Note=directly use masks as inputs into VOS model without using SAM2025.06 | 89.7 | 62.5 | 81.4 | 64.8 | — | |
| SAM 2.1Prompt Type=Average, Para. (M)=38.9, Cost=2.2m2025.06 | 88.2 | 74.4 | 79.3 | 75.3 | 79.3 | |
| SAM 2.1Prompt Type=bounding box, Para. (M)=38.9, Cost=2.2m2025.06 | 87.8 | 75.4 | 77.7 | 75.1 | — | |
| TinySAM + SAM-I2VPrompt Type=ground-truth mask, Para. (M)=18.9, Cost=4.6k, Note=directly use masks as inputs into VOS model without using SAM2025.06 | 87.8 | 68.9 | 79.3 | 59.3 | — | |
| SAM + XMem++Prompt Type=Average, Para. (M)=1572025.06 | 86.9 | 61.2 | 75.1 | 59.9 | 70.8 | |
| SAM + CutiePrompt Type=Average, Para. (M)=129.82025.06 | 86.8 | 57.8 | 75.6 | 62.9 | 70.8 | |
| SAM 2.1Prompt Type=3 click, Para. (M)=38.9, Cost=2.2m2025.06 | 86.6 | 66.8 | 78.1 | 74.4 | — | |
| SAM + XMem++Prompt Type=bounding box, Para. (M)=1572025.06 | 86.3 | 63 | 73 | 59.7 | — | |
| TinySAM + XMem++Prompt Type=Average, Para. (M)=72.32025.06 | 86.2 | 62.6 | 74 | 59.6 | 70.6 | |
| SAM + CutiePrompt Type=bounding box, Para. (M)=129.82025.06 | 86.1 | 60 | 73.4 | 62.8 | — | |
| TinySAM + CutiePrompt Type=Average, Para. (M)=45.12025.06 | 86 | 59.2 | 74.7 | 62.4 | 70.6 | |
| TinySAM + SAM-I2VPrompt Type=Average, Para. (M)=18.9, Cost=4.6k2025.06 | 85.2 | 64.4 | 73.7 | 57.6 | 70.2 | |
| TinySAM + SAM-I2VPrompt Type=3 click, Para. (M)=18.9, Cost=4.6k2025.06 | 84.7 | 58.1 | 72.3 | 57 | — | |
| SAM + CutiePrompt Type=3 click, Para. (M)=129.82025.06 | 84.6 | 51 | 71.9 | 61.1 | — | |
| SAM + XMem++Prompt Type=3 click, Para. (M)=1572025.06 | 84.5 | 52.7 | 71.6 | 58.4 | — | |
| TinySAM + XMem++Prompt Type=bounding box, Para. (M)=72.32025.06 | 84.5 | 60.2 | 70.1 | 58.4 | — | |
| TinySAM + CutiePrompt Type=bounding box, Para. (M)=45.12025.06 | 84.3 | 58.5 | 70.8 | 61.3 | — | |
| TinySAM + CutiePrompt Type=3 click, Para. (M)=45.12025.06 | 84.1 | 56.7 | 72 | 61.1 | — | |
| TinySAM + XMem++Prompt Type=3 click, Para. (M)=72.32025.06 | 84 | 59.7 | 71.3 | 58.8 | — | |
| TinySAM + SA + TAPrompt Type=ground-truth mask, Para. (M)=18.7, Cost=4.6k, Note=directly use masks as inputs into VOS model without using SAM2025.06 | 83.5 | 61.1 | 77.4 | 50.5 | — | |
| TinySAM + SAM-I2VPrompt Type=bounding box, Para. (M)=18.9, Cost=4.6k2025.06 | 83.2 | 66.1 | 69.5 | 56.5 | — | |
| TinySAM + TAPrompt Type=ground-truth mask, Para. (M)=17.8, Cost=4.6k, Note=directly use masks as inputs into VOS model without using SAM2025.06 | 83.1 | 52.5 | 77.3 | 49.6 | — | |
| TinySAM + TAPrompt Type=Average, Para. (M)=17.8, Cost=4.6k2025.06 | 81.1 | 48.3 | 72.6 | 48.9 | 62.7 | |
| TinySAM + SA + TAPrompt Type=Average, Para. (M)=18.7, Cost=4.6k2025.06 | 80.8 | 57.9 | 73.1 | 50 | 65.5 | |
| TinySAM + TAPrompt Type=bounding box, Para. (M)=17.8, Cost=4.6k2025.06 | 80.2 | 47.3 | 69.3 | 48.6 | — | |
| TinySAM + TAPrompt Type=3 click, Para. (M)=17.8, Cost=4.6k2025.06 | 80 | 45 | 71.2 | 48.6 | — | |
| TinySAM + SA + TAPrompt Type=bounding box, Para. (M)=18.7, Cost=4.6k2025.06 | 79.9 | 57.6 | 70.5 | 49.8 | — | |
| TinySAM + SA + TAPrompt Type=3 click, Para. (M)=18.7, Cost=4.6k2025.06 | 79 | 54.9 | 71.5 | 49.8 | — |