Video Object Segmentation on YouTube-VOS 2019 (val)
89.3J-Score (Seen)BATMAN
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| BATMANConstant Cost=true2023.03 | 89.3 | 84.5 | 79 | 87.2 | 85 | — | — | — | — | |
| XMemConstant Cost=true, YouTube-VOS training stage=true, BL30K training stage=true2023.03 | 89.2 | 84.8 | 80.3 | 88.8 | 85.8 | — | — | — | — | |
| Cutie-base2024.12 | 86.8 | — | 85.9 | 90.4 | 81.6 | — | — | — | 89.3 | |
| ReVOS2024.12 | 86.8 | — | 85.3 | 89.8 | 82.1 | — | — | — | 89.9 | |
| DeAOTBackbone=SwinB-DeAOT-L2024.12 | 86.2 | — | 85.6 | 90.6 | 80 | — | — | — | 88.4 | |
| CutieVariant=base+2025.10 | 86.2 | 86.9 | 90.7 | 81.6 | 89.2 | — | — | — | — | |
| JointFormer2025.10 | 86.1 | 87 | 90.6 | 82 | 89.5 | — | — | — | — | |
| RMem + DeAOTBackbone=R50-DeAOT-L, Memory bank=restricted, Temporal position embedding=false2024.12 | 85.9 | — | 84.6 | 89.4 | 80.8 | — | — | — | 88.9 | |
| SAM 2++2025.10 | 85.8 | 87.1 | 92.5 | 80 | 90.3 | — | — | — | — | |
| Cutie-basewith STM=true, Backbone=ResNet-50, Training Data=YouTube VOS + DAVIS2024.11 | 85.5 | — | 90 | 80.6 | 88.3 | — | — | — | 86.1 | |
| DEVAtraining_data=with OVIS, FPS=25.32023.09 | 85.4 | 86.2 | 89.9 | 80.5 | 89.1 | — | — | — | — | |
| Cutie-basewith STM=true, Backbone=ResNet-50, Training Data=YouTube VOS + DAVIS + MOSE2024.11 | 85.4 | — | 90 | 81.3 | 89.3 | — | — | — | 86.5 | |
| XMem2024.12 | 85.4 | — | 84.3 | 88.6 | 80.3 | — | — | — | 88.5 | |
| DeAOTBackbone=SwinB2025.10 | 85.4 | 86.4 | 90.3 | 80.6 | 89.3 | — | — | — | — | |
| Cutie-smallwith STM=true, Backbone=ResNet-18, Training Data=YouTube VOS + DAVIS2024.11 | 85.3 | — | 89.6 | 80.9 | 89 | — | — | — | 86.2 | |
| DEVA2025.10 | 85.3 | 86.3 | 89.8 | 80.6 | 89.2 | — | — | — | — | |
| Cutie-smallwith STM=true, Backbone=ResNet-18, Training Data=YouTube VOS + DAVIS + MOSE2024.11 | 85.2 | — | 89.7 | 81.1 | 89.2 | — | — | — | 86.3 | |
| SAM22024.12 | 85.2 | — | 83.7 | 90.6 | 87.9 | — | — | — | 88.5 | |
| DEVAtraining_data=without OVIS, FPS=25.32023.09 | 85 | 85.5 | 89.4 | 79.7 | 88 | — | — | — | — | |
| SwinB-AOST (nl=3)†backbone=Swin-Base, nl=3, alpha=infinity, All-Frame (30FPS)=true, Hierarchical Structure=true2022.03 | 85 | — | 89.8 | 80.3 | 88.8 | — | 86 | 5.1 | — | |
| DEVAwith STM=true, Backbone=ResNet-502024.11 | 85 | — | 89.4 | 79.7 | 88 | — | — | — | 85.5 | |
| MITSInitialization format=Mask, Backbone=ResNet-502023.08 | 84.9 | 85.9 | 89.7 | 80.4 | 88.7 | — | — | — | — | |
| OneVOSFine-tuned on=MOSE2025.10 | 84.9 | 86.3 | 89.9 | 81.1 | 89.4 | — | — | — | — | |
| XMemAll-Frame (30FPS)=true2022.03 | 84.8 | — | 89.2 | 80.3 | 88.8 | — | 85.5 | 22.6 | — | |
| R50-AOST (nl=3)†backbone=ResNet-50, nl=3, alpha=infinity, All-Frame (30FPS)=true, Hierarchical Structure=true2022.03 | 84.7 | — | 89.6 | 79.2 | 87.6 | — | 85.3 | 6.3 | — | |
| DeAOTInitialization format=Mask2023.08 | 84.6 | 85.9 | 89.4 | 80.8 | 88.9 | — | — | — | — | |
| DeAOTMethod Category=Memory-based2023.12 | 84.6 | 85.9 | 89.4 | 80.8 | 88.9 | — | — | — | — | |
| DeAOTHierarchical Structure=true2022.03 | 84.6 | — | 89.4 | 80.8 | 88.9 | — | 85.9 | 22.4 | — | |
| AOTBackbone=SwinB2025.10 | 84.6 | 85.3 | 89.5 | 79.3 | 87.7 | — | — | — | — | |
| BATMAN2022.08 | 84.5 | — | 89.3 | 79 | 87.2 | — | 85 | — | — | |
| SwinB-AOST (nl=3)backbone=Swin-Base, nl=3, Hierarchical Structure=true2022.03 | 84.5 | — | 89.3 | 79 | 87.3 | — | 85 | 9.3 | — | |
| XMemLabeled data=100%2023.03 | 84.3 | 85.5 | 88.6 | 80.3 | 88.6 | — | — | — | — | |
| XMemConstant Cost=false2023.06 | 84.3 | — | 88.6 | 80.3 | 88.6 | — | 85.5 | — | — | |
| XMemInitialization format=Mask2023.08 | 84.3 | 85.5 | 88.6 | 80.3 | 88.6 | — | — | — | — | |
| XMemFPS=22.62023.09 | 84.3 | 85.5 | 88.6 | 80.3 | 88.6 | — | — | — | — | |
| XMemMethod Category=Memory-based2023.12 | 84.3 | 85.5 | 88.6 | 80.3 | 88.6 | — | — | — | — | |
| XMemwith STM=true, Backbone=ResNet-502024.11 | 84.3 | — | 88.6 | 80.3 | 88.6 | — | — | — | 85.5 | |
| XMem2025.10 | 84.3 | 85.5 | 88.6 | 80.5 | 88.7 | — | — | — | — | |
| SwinB-AOST (nl=3)†backbone=Swin-Base, nl=3, alpha=infinity, Hierarchical Structure=true2022.03 | 84.2 | — | 88.9 | 79.8 | 88 | — | 85.2 | 9.3 | — | |
| DeAOTwith STM=true, Backbone=ResNet-502024.11 | 84.2 | — | 89.2 | 80.2 | 88.8 | — | — | — | 85.6 | |
| SwinB-AOT-LBackbone=Swin-B2021.06 | 84 | 84.5 | 88.8 | 78.4 | 86.7 | — | — | — | — | |
| XMem*Labeled data=100%2023.03 | 84 | 85.3 | 88.2 | 80.4 | 88.4 | — | — | — | — | |
| R50-AOST (nl=3)†backbone=ResNet-50, nl=3, alpha=infinity, Hierarchical Structure=true2022.03 | 84 | — | 88.7 | 79.5 | 87.9 | — | 85 | 14.9 | — | |
| AOTInitialization format=Mask2023.08 | 83.9 | 85.3 | 88.8 | 79.9 | 88.5 | — | — | — | — | |
| AOT-R50Backbone=ResNet-50, FPS=6.42023.09 | 83.9 | 85.3 | 88.8 | 79.9 | 88.5 | — | — | — | — | |
| AOTwith STM=true, Backbone=ResNet-502024.11 | 83.9 | — | 88.8 | 79.9 | 88.5 | — | — | — | 85.3 | |
| R50-AOST (nl=3)backbone=ResNet-50, nl=3, Hierarchical Structure=true2022.03 | 83.8 | — | 88.7 | 79.3 | 87.7 | — | 84.9 | 14.9 | — | |
| XMemStage2 only training=true2024.09 | 83.6 | 84.3 | 78.5 | 88 | 87.1 | — | — | — | — | |
| LiVOSwith STM=false, Backbone=ResNet-50, Training Data=YouTube VOS + DAVIS, Training iterations=30k additional2024.11 | 83.6 | — | 87.6 | 73.3 | 80.6 | — | — | — | 81.3 | |
| R50-AOT-LBackbone=ResNet-502021.06 | 83.5 | 84.1 | 88.1 | 78.4 | 86.3 | — | — | — | — | |
| R50-AOTRegional Segmentation=false2022.07 | 83.5 | — | 88.1 | 78.4 | 86.3 | — | 84.1 | 14.9 | — | |
| AOT2022.08 | 83.5 | — | 88.1 | 78.4 | 86.3 | — | 84.1 | — | — | |
| R50-AOT-LLabeled data=100%2023.03 | 83.5 | 84.1 | 88.1 | 78.4 | 86.3 | — | — | — | — | |
| 2-shot XMem w/ OursLabeled data=7.3%2023.03 | 83.5 | 84.5 | 88 | 79.1 | 87.3 | — | — | — | — | |
| AOT-LStage2 only training=true2024.09 | 83.5 | 84.1 | 78.4 | 88.1 | 86.3 | — | — | — | — | |
| LiVOSwith STM=false, Backbone=ResNet-50, Training Data=YouTube VOS + DAVIS + MOSE2024.11 | 83.5 | — | 87.6 | 75.8 | 83.3 | — | — | — | 82.6 | |
| AOST (nl=3)nl=3, Hierarchical Structure=true2022.03 | 83.3 | — | 88 | 79 | 87.4 | — | 84.4 | 15.9 | — | |
| MobileVOSConstant Cost=true, Variant=Model soup, YouTube-VOS training stage=true2023.03 | 83.2 | 83.3 | 87.7 | 76.9 | 85.3 | — | — | — | — | |
| UniRef++-LMethod Category=Non-memory, Backbone=Large2023.12 | 83.1 | 83 | 87.7 | 77.3 | 84.1 | — | — | — | — | |
| RPCMmulti-scale testing=true2021.12 | 83 | — | 87.3 | 79.4 | 87.2 | — | 84.2 | — | — | |
| PCVOSAll-Frame (30FPS)=true2022.03 | 83 | — | 88 | 79.6 | 87.9 | — | 84.6 | 10.5 | — | |
| UniRef-LMethod Category=Non-memory, Backbone=Large2023.12 | 82.9 | 82.7 | 86.9 | 76.8 | 84.1 | — | — | — | — | |
| JIMDStage2 only training=true2024.09 | 82.9 | 84.6 | 79.7 | 87.8 | 87.9 | — | — | — | — | |
| AOT-L2021.06 | 82.8 | 83.7 | 87.5 | 78 | 86.7 | — | — | — | — | |
| AOT-LMethod Category=Memory-based, Variant=Large2023.12 | 82.8 | 83.7 | 87.5 | 78 | 86.7 | — | — | — | — | |
| AOT-LHierarchical Structure=true2022.03 | 82.8 | — | 87.5 | 78 | 86.7 | — | 83.7 | 16 | — | |
| Adaptive Object Calibration Networkvariant=MF, multiple historical frames=true, external data=true2022.07 | 82.7 | — | 86.9 | 79.4 | 87.2 | — | 84.1 | — | — | |
| Adaptive Object Calibration Networkvariant=BaseMS, multi-scale and flip testing=true, external data=true2022.07 | 82.7 | — | 87.1 | 80 | 87.8 | — | 84.4 | — | — | |
| STCNPre-training dataset=BL30K2021.06 | 82.6 | 84.2 | 87 | 79.4 | 87.7 | — | — | — | — | |
| RPCM2021.12 | 82.6 | — | 86.9 | 79.1 | 87.1 | — | 83.9 | — | — | |
| STCNConstant Cost (CC)=false, BL30K pre-training=true2022.05 | 82.6 | 84.2 | 87 | 79.4 | 87.7 | — | — | — | — | |
| STCNRegional Segmentation=false2022.07 | 82.6 | — | 87 | 79.4 | 87.7 | — | 84.2 | 11 | — | |
| RAVOSRegional Segmentation=true2022.07 | 82.6 | — | 87 | 79.4 | 87.7 | — | 84.2 | 20 | — | |
| PCVOSvariant=Ours-L52022.08 | 82.6 | 84.6 | 87.3 | 80 | 88.3 | — | — | — | — | |
| RPCMVOS2022.08 | 82.6 | — | 86.9 | 79.1 | 87.1 | — | 83.9 | — | — | |
| STCNConstant Cost=false, YouTube-VOS training stage=true, BL30K training stage=true2023.03 | 82.6 | 84.2 | 87 | 79.4 | 87.7 | — | — | — | — | |
| STCNConstant Cost=false, Additional training data (BL30K)=true2023.06 | 82.6 | — | 87 | 79.4 | 87.7 | — | 84.2 | — | — | |
| RPCMAll-Frame (30FPS)=false2022.03 | 82.6 | — | 86.9 | 79.1 | 87.1 | — | 83.9 | 7.7 | — | |
| STCN2024.09 | 82.6 | 84.2 | 79.4 | 87 | 87.7 | — | — | — | — | |
| LiVOSwith STM=false, Backbone=ResNet-50, Training Data=YouTube VOS + DAVIS2024.11 | 82.6 | — | 86.8 | 71.7 | 78.4 | — | — | — | 79.9 | |
| AOT-B2022.07 | 82.5 | — | 87 | 77.8 | 86 | — | 83.3 | — | — | |
| STCN+InstMoveintegration=InstMove2023.03 | 82.5 | 83.4 | 77.9 | 86.9 | 86 | — | — | — | — | |
| AOT-B2021.06 | 82.4 | 83.3 | 87.1 | 77.8 | 86 | — | — | — | — | |
| AOT-BMethod Category=Memory-based, Variant=Base2023.12 | 82.4 | 83.3 | 87.1 | 77.8 | 86 | — | — | — | — | |
| AOT-BHierarchical Structure=true2022.03 | 82.4 | — | 87.1 | 77.8 | 86 | — | 83.3 | 20.5 | — | |
| Adaptive Object Calibration Networkvariant=Base2022.07 | 82.3 | — | 86.6 | 79 | 86.9 | — | 83.7 | — | — | |
| AOT-Lmultiple historical frames=true, external data=true2022.07 | 82.2 | — | 86.9 | 78.3 | 86.9 | — | 83.6 | — | — | |
| AOTArchitecture=Transformer-based2022.08 | 82.2 | 83.6 | 86.9 | 78.3 | 86.9 | — | — | — | — | |
| MobileVOSConstant Cost=true, Backbone=ResNet18, YouTube-VOS training stage=true, BL30K training stage=true2023.03 | 82.1 | 82.8 | 86.4 | 77 | 85.6 | — | — | — | — | |
| STCN + TrickVOS (PT)Constant Cost=false, STCN re-training (no ASPP/BL30K)=true2023.06 | 82.1 | — | 86.4 | 77.2 | 85.5 | — | 82.8 | — | — | |
| SWEMInference speed=> 20 FPS2022.08 | 82 | 82.6 | 86.1 | 77.2 | 85.2 | — | — | — | — | |
| RDE-VOSConstant Cost (CC)=true, BL30K pre-training=true2022.05 | 81.9 | 83.3 | 86.3 | 78 | 86.9 | — | — | — | — | |
| RDE-VOSConstant Cost=true, YouTube-VOS training stage=true, BL30K training stage=true2023.03 | 81.9 | 83.3 | 86.3 | 78 | 86.9 | — | — | — | — | |
| MITSInitialization format=Box, Backbone=ResNet-502023.08 | 81.9 | 81.8 | 86.2 | 75.2 | 83.7 | — | — | — | — | |
| UniRef++-R50Method Category=Non-memory, Backbone=ResNet-502023.12 | 81.9 | 81.9 | 86.2 | 76.5 | 83.2 | — | — | — | — | |
| CFBImulti-scale and flip strategy=true2020.03 | 81.8 | 82.4 | 86.1 | 76.9 | 84.8 | — | — | — | — | |
| CFBIMSsynthetic data for offline training=false, online fine-tuning=false2021.08 | 81.8 | 82.4 | 86.1 | 76.9 | 84.8 | — | — | — | — | |
| CFBImulti-scale testing=true, flip testing=true2021.12 | 81.8 | — | 86.1 | 76.9 | 84.8 | — | 82.4 | — | — | |
| CFBImulti-scale and flip testing=true, external data=true2022.07 | 81.8 | — | 86.1 | 76.9 | 84.8 | — | 82.4 | — | — | |
| HMMN2021.09 | 81.7 | 82.5 | 86.1 | 77.3 | 85 | — | — | — | — |