Video Instance Segmentation on YouTube-VIS (val)
47.6APPropose-Reduce
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Propose-ReduceParadigm=Propose-Reduce, Backbone=ResNeXt-101, HR-Ref=false2021.03 | 47.6 | 71.6 | 51.8 | 46.3 | 56 | — | |
| MaskPropParadigm=Clip-Match, Backbone=STSN [3]-ResNeXt-101, HR-Ref=true2021.03 | 46.6 | — | 51.2 | 44 | 52.6 | — | |
| EnsembleVISParadigm=Track-by-Detect, Backbone=ResNeXt-101*, HR-Ref=false2021.03 | 44.8 | — | 48.9 | 42.7 | 51.7 | — | |
| MaskPropParadigm=Clip-Match, Backbone=STSN [3]-ResNeXt-101, HR-Ref=false2021.03 | 44.7 | — | — | — | — | — | |
| MaskPropParadigm=Clip-Match, Backbone=ResNeXt-101, HR-Ref=true2021.03 | 44.3 | — | 48.3 | — | — | — | |
| Propose-ReduceParadigm=Propose-Reduce, Backbone=ResNet-101, HR-Ref=false2021.03 | 43.8 | 65.5 | 47.4 | 43 | 53.2 | — | |
| MaskPropType=Two-stage methods, Backbone=R101-FPN, Frames=13, Anchors=3, Others=high-solution mask refinement, temporal information, multi-scale training2021.04 | 42.5 | — | 45.6 | — | — | — | |
| Propose-ReduceParadigm=Propose-Reduce, Backbone=ResNet-50, HR-Ref=false2021.03 | 40.4 | 63 | 43.8 | 41.1 | 49.7 | — | |
| MaskPropType=Two-stage methods, Backbone=R50-FPN, Frames=13, Anchors=3, Others=high-solution mask refinement, temporal information, multi-scale training2021.04 | 40 | — | 42.9 | — | — | — | |
| PCANBackbone=ResNet-101, Type=One-stage, Online=true2021.06 | 37.6 | 57.2 | 41.3 | 37.2 | 43.9 | — | |
| STMask (ada)Type=One-stage methods, Backbone=R101-DCN, Frames=2, Anchors=3, Others=temporal information, FPS=23.42021.04 | 36.8 | 56.8 | 38 | 34.8 | 41.8 | — | |
| SG-NetStage Type=One Stage, Backbone=ResNet-101, Time (ms)=50.62021.03 | 36.3 | 57.1 | 39.6 | 35.9 | 43 | — | |
| STMask (ali)Type=One-stage methods, Backbone=R101-DCN, Frames=2, Anchors=3, Others=temporal information, FPS=22.12021.04 | 36.3 | 55.2 | 39.9 | 33.7 | 42 | — | |
| STMask*Backbone=ResNet-101, Type=One-stage, Online=true2021.06 | 36.3 | 55.2 | 39.9 | 33.7 | 42 | — | |
| SG-Net*Backbone=ResNet-101, Type=One-stage, Online=true2021.06 | 36.3 | 57.1 | 39.6 | 35.9 | 43 | — | |
| QueryInst-VISBackbone=ResNet-50, Multi-scale data augmentation=true2021.05 | 36.2 | 56.7 | 39.7 | 36.1 | 42.9 | 32.3 | |
| PCANBackbone=ResNet-50, Type=One-stage, Online=true2021.06 | 36.1 | 54.9 | 39.4 | 36.3 | 41.6 | — | |
| ObjPropBackbone=ResNet-502021.11 | 36 | 59.4 | 39.2 | 39.1 | 47.7 | — | |
| SipMaskStage Type=One Stage, Backbone=ResNet-101, Time (ms)=41.72021.03 | 35.8 | 56 | 39 | 35.4 | 42.4 | — | |
| VisTr*Backbone=ResNet-50, Type=Transformer, Online=false2021.06 | 35.6 | 56.8 | 37 | 35.2 | 40.2 | — | |
| CompFeatpost-processing=true2020.12 | 35.3 | 56 | 38.6 | 33.1 | 40.3 | — | |
| CompFeatBackbone=ResNet-50, Multi-scale data augmentation=false2021.05 | 35.3 | 56 | 38.6 | 33.1 | 40.3 | — | |
| VisTRBackbone=ResNet-101, Multi-scale data augmentation=false2021.05 | 35.3 | 57 | 36.2 | 34.3 | 40.4 | 27.7 | |
| SipMask++Type=One-stage methods, Backbone=R101-DCN, Frames=2, FPS=27.82021.04 | 35 | 56.1 | 35.2 | 36 | 41.2 | — | |
| SG-NetStage Type=One Stage, Backbone=ResNet-50, Time (ms)=43.52021.03 | 34.8 | 56.1 | 36.8 | 35.8 | 40.8 | — | |
| SG-Net*Backbone=ResNet-50, Type=One-stage, Online=true2021.06 | 34.8 | 56.1 | 36.8 | 35.8 | 40.8 | — | |
| STEm-SegStage Type=One Stage, Backbone=ResNet-101, Time (ms)=1252021.03 | 34.6 | 55.8 | 37.9 | 34.4 | 41.6 | — | |
| STEm-SegType=VOS baseline, Backbone=R101-FPN, FPS=2.12021.04 | 34.6 | 55.8 | 37.9 | 34.4 | 41.6 | — | |
| STEm-SegParadigm=Clip-Match, Backbone=ResNet-101, HR-Ref=false2021.03 | 34.6 | 55.8 | 37.9 | 34.4 | 41.6 | — | |
| STEm-Seg2020.03 | 34.6 | 55.8 | 37.9 | 34.4 | 41.6 | — | |
| STEm-SegBackbone=ResNet-101, Multi-scale data augmentation=false2021.05 | 34.6 | 55.8 | 37.9 | 34.4 | 41.6 | 2.1 | |
| QueryInst-VISBackbone=ResNet-50, Multi-scale data augmentation=false2021.05 | 34.6 | 55.8 | 36.5 | 35.4 | 42.4 | 32.3 | |
| VisTRBackbone=ResNet-50, Multi-scale data augmentation=false2021.05 | 34.4 | 55.7 | 36.5 | 33.5 | 38.9 | 30 | |
| SipMaskcategory=track-by-detect, multi-scale training=true2020.07 | 33.7 | 54.1 | 35.8 | 35.4 | 40.1 | — | |
| SipMaskStage Type=One Stage, Backbone=ResNet-50, Time (ms)=35.72021.03 | 33.7 | 54.1 | 35.8 | 35.4 | 40.1 | — | |
| SipMaskParadigm=Track-by-Detect, Backbone=ResNet-50, HR-Ref=false2021.03 | 33.7 | 54.1 | 35.8 | 35.4 | 40.1 | — | |
| SipMask-VISBackbone=ResNet-50, Multi-scale data augmentation=true2021.05 | 33.7 | 54.1 | 35.8 | 35.4 | 40.1 | 30.9 | |
| STMask (ada)Type=One-stage methods, Backbone=R50-DCN, Frames=2, Anchors=9, Others=temporal information, FPS=28.62021.04 | 33.5 | 52.1 | 36.9 | 31.1 | 39.2 | — | |
| STMask*Backbone=ResNet-50, Type=One-stage, Online=true2021.06 | 33.5 | 52.1 | 36.9 | 31.1 | 39.2 | — | |
| TraDeS (Ours)Publication=CVPR'212021.03 | 32.6 | 52.6 | 32.8 | — | — | — | |
| SipMaskcategory=track-by-detect2020.07 | 32.5 | 53 | 33.3 | 33.5 | 38.9 | — | |
| SipMaskpost-processing=true2020.12 | 32.5 | 53 | 33.3 | 33.5 | 38.9 | — | |
| SipMaskPublication=ECCV'202021.03 | 32.5 | 53 | 33.3 | — | — | — | |
| SipMask++Type=One-stage methods, Backbone=R50-DCN, Frames=2, FPS=302021.04 | 32.5 | 53 | 33.3 | 33.5 | 38.9 | — | |
| SipMaskBackbone=ResNet-50, Type=One-stage, Online=true2021.06 | 32.5 | 53 | 33.3 | 33.5 | 38.9 | — | |
| SipMask2021.07 | 32.5 | — | — | — | — | — | |
| SipMaskBackbone=ResNet-502021.11 | 32.5 | 53 | 33.3 | 33.5 | 38.9 | — | |
| SipMask-VISBackbone=ResNet-50, Multi-scale data augmentation=false2021.05 | 32.5 | 53 | 33.3 | 33.5 | 38.9 | 30.9 | |
| MaskTrack R-CNN + BCNetBackbone=ResNet-502022.08 | 32.4 | 53.9 | 34 | 33.9 | 39.1 | — | |
| MaskTrack R-CNNExternal Training Data=MSCOCO, post-processing=true2020.12 | 32.2 | 52.1 | 34.6 | 31.8 | 37.2 | — | |
| MaskTrack R-CNNStage Type=Two Stage, Backbone=ResNet-101, Time (ms)=142.82021.03 | 31.8 | 53 | 33.6 | 33.2 | 37.6 | — | |
| STEm-SegStage Type=One Stage, Backbone=ResNet-50, Time (ms)=83.32021.03 | 30.6 | 50.7 | 33.5 | 31.6 | 37.1 | — | |
| STEm-SegType=VOS baseline, Backbone=R50-FPN2021.04 | 30.6 | 50.7 | 33.5 | 31.6 | 37.1 | — | |
| STEm-SegParadigm=Clip-Match, Backbone=ResNet-50, HR-Ref=false2021.03 | 30.6 | 50.7 | 33.5 | 31.6 | 37.1 | — | |
| STEm-SegBackbone=ResNet-50, Type=One-stage, Online=false2021.06 | 30.6 | 50.7 | 33.5 | 31.6 | 37.1 | — | |
| STEm-SegBackbone=ResNet-502021.11 | 30.6 | 50.7 | 33.5 | 31.6 | 37.1 | — | |
| STEm-SegBackbone=ResNet-502020.03 | 30.6 | 50.7 | 33.5 | 31.6 | 37.1 | — | |
| STEm-SegBackbone=ResNet-50, Multi-scale data augmentation=false2021.05 | 30.6 | 50.7 | 33.5 | 31.6 | 37.1 | 4.4 | |
| MaskTrack R-CNNCategory=Track-by-detect2019.05 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrack R-CNNcategory=track-by-detect2020.07 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrack R-CNNpost-processing=true2020.12 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrack R-CNNPublication=ICCV'192021.03 | 30.3 | 51.1 | 32.6 | — | — | — | |
| MaskTrackSupervision Type=Fully supervised2021.03 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrack R-CNNStage Type=Two Stage, Backbone=ResNet-50, Time (ms)=98.82021.03 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrackType=Two-stage methods, Backbone=R50-FPN, Frames=2, Anchors=3, FPS=18.42021.04 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrackParadigm=Track-by-Detect, Backbone=ResNet-50, HR-Ref=false2021.03 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrack R-CNNBackbone=ResNet-50, Type=Two-stage, Online=true2021.06 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MTRCNNuses same observations=true, observation details=segmentation masks in every single frames2021.07 | 30.3 | — | — | — | — | — | |
| MaskTrack R-CNNBackbone=ResNet-502021.11 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrack R-CNNBackbone=ResNet-502022.08 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrack R-CNNProposals/Detections=true, Optical Flow=true2020.03 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | — | |
| MaskTrack R-CNNBackbone=ResNet-50, Multi-scale data augmentation=false2021.05 | 30.3 | 51.1 | 32.6 | 31 | 35.5 | 22.1 | |
| UniTrackuses same observations=true, observation details=segmentation masks in every single frames2021.07 | 30.1 | — | — | — | — | — | |
| OSMNCategory=Track-by-detect2019.05 | 27.5 | 45.1 | 29.1 | 28.6 | 33.1 | — | |
| SeqTrackerCategory=Track-by-detect2019.05 | 27.5 | 45.7 | 28.7 | 29.7 | 32.5 | — | |
| OSMNcategory=track-by-detect2020.07 | 27.5 | 45.1 | 29.1 | 28.6 | 31.1 | — | |
| OSMNpost-processing=true2020.12 | 27.5 | 45.1 | 29.1 | 28.6 | 33.1 | — | |
| SeqTrackerpost-processing=true2020.12 | 27.5 | 45.7 | 28.7 | 29.7 | 32.5 | — | |
| OSMN (track-by-detect)Publication=CVPR'18, variant=track-by-detect2021.03 | 27.5 | 45.1 | 29.1 | — | — | — | |
| OSMNStage Type=Two Stage, Backbone=ResNet-502021.03 | 27.5 | 45.1 | 29.1 | 28.6 | 31.1 | — | |
| SeqTrackerStage Type=Two Stage, Backbone=ResNet-502021.03 | 27.5 | 45.7 | 28.7 | 29.7 | 32.5 | — | |
| OSMNType=VOS baseline, Backbone=R50-FPN, Frames=3, Anchors=32021.04 | 27.5 | 45.1 | 29.1 | 28.6 | 33.1 | — | |
| OSMNuses same observations=true, observation details=segmentation masks in every single frames2021.07 | 27.5 | — | — | — | — | — | |
| OSMNProposals/Detections=true2020.03 | 27.5 | 45.1 | 29.1 | 28.6 | 33.1 | — | |
| SeqTrackerProposals/Detections=true2020.03 | 27.5 | 45.7 | 28.7 | 29.7 | 32.5 | — | |
| FEELVOSCategory=Mask propagation2019.05 | 26.9 | 42 | 29.7 | 29.9 | 33.4 | — | |
| FEELVOScategory=mask propagation2020.07 | 26.9 | 42 | 29.7 | 29.9 | 33.4 | — | |
| FEELVOSPublication=CVPR'192021.03 | 26.9 | 42 | 29.7 | — | — | — | |
| FEELVOSStage Type=Two Stage, Backbone=ResNet-502021.03 | 26.9 | 42 | 29.7 | 29.9 | 33.4 | — | |
| FEELVOSType=VOS baseline, Backbone=R50-FPN, Frames=3, Anchors=32021.04 | 26.9 | 42 | 29.7 | 29.9 | 33.4 | — | |
| FEELVOSBackbone=ResNet-50, Type=Two-stage, Online=true2021.06 | 26.9 | 42 | 29.7 | 29.9 | 33.4 | — | |
| FEELVOS2021.07 | 26.9 | — | — | — | — | — | |
| FEELVOSBackbone=ResNet-502022.08 | 26.9 | 42 | 29.7 | 29.9 | 33.4 | — | |
| FEELVOSFirst Frame Proposals=true2020.03 | 26.9 | 42 | 29.7 | 29.9 | 33.4 | — | |
| Our Baseline2021.03 | 26.4 | 43.2 | 26.8 | — | — | — | |
| DeepSORTCategory=Track-by-detect2019.05 | 26.1 | 42.9 | 26.1 | 27.8 | 31.3 | — | |
| DeepSORTpost-processing=true2020.12 | 26.1 | 42.9 | 26.1 | 27.8 | 31.3 | — | |
| DeepSORTSupervision Type=Fully supervised2021.03 | 26.1 | 42.9 | 26.1 | 27.8 | 31.3 | — | |
| DeepSORTStage Type=Two Stage, Backbone=ResNet-502021.03 | 26.1 | 42.9 | 26.1 | 27.8 | 31.3 | — | |
| DeepSORTBackbone=ResNet-50, Type=Two-stage, Online=true2021.06 | 26.1 | 42.9 | 26.1 | 27.8 | 31.3 | — |