Instance Segmentation on COCO 2017 (test-dev)
53AP (Overall)Soft Teacher
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Soft Teacherdetector=HTC++, backbone=Swin-L, scale=multi-scale, pre-training=Object365 [24]2021.06 | 53 | — | — | — | — | — | — | — | — | — | |
| Soft Teacherdetector=HTC++, backbone=Swin-L, scale=multi-scale2021.06 | 52.4 | — | — | — | — | — | — | — | — | — | |
| Mask2Former + Instance-Unique QueryingBackbone=Swin-L, #Epoch=1002022.10 | 51.8 | 76 | 56.8 | 29.9 | 55.1 | 73.3 | — | — | — | — | |
| Superviseddetector=HTC++, backbone=Swin-L, scale=multi-scale2021.06 | 51.2 | — | — | — | — | — | — | — | — | — | |
| Mask2FormerBackbone=Swin-L, #Epoch=1002022.10 | 50.2 | 74.8 | 54.7 | 29.2 | 53.8 | 71.1 | — | — | — | — | |
| QueryInstBackbone=Swin-L, Epochs=50, Training scale=multi-scale [400, 1200], Testing scale=single-scale2021.06 | 49.1 | — | — | 31.5 | 51.8 | 63.2 | — | — | — | — | |
| QueryInst (300 Queries)Backbone=Swin-L, Aug.=400 ~ 1200 w/ crop, Epochs=50, FPS=3.3 T, Stages=6, Queries=3002021.05 | 49.1 | 74.2 | 53.8 | 31.5 | 51.8 | 63.2 | — | — | — | — | |
| Hybrid Task CascadeDCN=true, SyncBN=true, Multi-scale training=true, Backbone=SENet-154, GA-RPN=true, Multi-scale testing=true, Ensemble=true2019.01 | 49 | 73 | 53.9 | 33.9 | 52.3 | 61.2 | — | — | — | — | |
| Hybrid Task CascadeDCN=true, SyncBN=true, Multi-scale training=true, Backbone=SENet-154, GA-RPN=true, Multi-scale testing=true2019.01 | 47.4 | 70.6 | 52.1 | 30.2 | 50.1 | 61.8 | — | — | — | — | |
| Mask2Former + Instance-Unique QueryingBackbone=Swin-B, #Epoch=502022.10 | 47.3 | 71.3 | 51 | 28.3 | 50.9 | 69.4 | — | — | — | — | |
| COCO 2017 Winner2019.01 | 46.7 | 69.5 | 51.3 | 26 | 49.1 | 64 | — | — | — | — | |
| SOLQBackbone=Swin-L, Epochs=50, Training scale=multi-scale [400, 1200], Testing scale=single-scale2021.06 | 46.7 | — | — | 29.2 | 50.1 | 60.9 | — | — | — | — | |
| SOLQBackbone=Swin-L, #Epoch=502022.10 | 46.7 | 72.7 | 50.6 | 29.2 | 50.1 | 60.9 | — | — | — | — | |
| Mask2Former + Instance-Unique QueryingBackbone=Swin-S, #Epoch=502022.10 | 46.7 | 70.7 | 50.1 | 27 | 50.9 | 68.8 | — | — | — | — | |
| GCNetBackbone=ResNeXt-101 64x4d, Cascade strategy=true, Deformable Convolution (DCN)=true, Head type=4conv1fc, Multiscale training=true, Training schedule=3x schd, GC ratio=r4, FLOPs(G)=1040.6G2020.12 | 45.4 | 68.9 | 49.6 | — | — | — | — | — | — | — | |
| Hybrid Task CascadeDCN=true, SyncBN=true, Multi-scale training=true, Backbone=SENet-154, GA-RPN=true2019.01 | 45.3 | 68.9 | 49.4 | 27 | 48.3 | 59.6 | — | — | — | — | |
| Mask2FormerBackbone=Swin-B, #Epoch=502022.10 | 44.9 | 68.5 | 48.5 | 24.6 | 48.7 | 67.4 | — | — | — | — | |
| Transformer-based BCNetBackbone=R50-FPN, Type=Query-based2022.08 | 44.6 | 68.1 | 48.7 | 24.1 | 47.7 | 66.7 | — | — | — | — | |
| QueryInst (300 Queries)Backbone=ResNeXt-101-FPN w/ DCN, Aug.=480 ~ 800 w/ crop, Epochs=36, FPS=3.1, Stages=6, Queries=3002021.05 | 44.6 | 68.1 | 48.7 | 26.6 | 46.9 | 57.7 | — | — | — | — | |
| Mask2FormerBackbone=Swin-S, #Epoch=502022.10 | 44.5 | 68.2 | 47.7 | 24.3 | 49 | 66.6 | — | — | — | — | |
| Hybrid Task CascadeDCN=true, SyncBN=true, Multi-scale training=true, Backbone=SENet-1542019.01 | 44.3 | 67.5 | 48.3 | 25 | 47.5 | 58.9 | — | — | — | — | |
| BoxeR-2DEpoch=36, end-to-end=true, Backbone=R-101, Schedule=3×2021.11 | 43.8 | — | — | 25 | 46.5 | 57.9 | — | — | — | — | |
| Mask2FormerBackbone=R50-FPN, Type=Query-based2022.08 | 43.6 | 66.5 | 47.9 | 23.5 | 47.4 | 64.1 | — | — | — | — | |
| QueryInstBackbone=R101-FPN, Epochs=36, Training scale=multi-scale, Testing scale=single-scale2021.06 | 42.8 | — | — | 24.6 | 45 | 55.5 | — | — | — | — | |
| QueryInstEpoch=36, end-to-end=false, Backbone=R-1012021.11 | 42.8 | — | — | 24.6 | 45 | 55.5 | — | — | — | — | |
| QueryInst (300 Queries)Backbone=ResNet-101-FPN, Aug.=480 ~ 800 w/ crop, Epochs=36, FPS=6.1, Stages=6, Queries=3002021.05 | 42.8 | 65.6 | 46.7 | 24.6 | 45 | 55.5 | — | — | — | — | |
| SOTR + Instance-Unique QueryingBackbone=ResNet-101, #Epoch=242022.10 | 42.6 | 64.1 | 45.8 | 11.2 | 61.2 | 75.3 | — | — | — | — | |
| Hybrid Task CascadeDCN=true, SyncBN=true, Multi-scale training=true2019.01 | 42.5 | 64.8 | 46.4 | 23.7 | 45.3 | 56.7 | — | — | — | — | |
| SOTR + Instance-Unique QueryingBackbone=ResNet-50, #Epoch=242022.10 | 42.2 | 61.9 | 43.9 | 11 | 60.5 | 73.5 | — | — | — | — | |
| PANetBackbone=ResNeXt-101, multi-scale training=true2018.03 | 42 | 65.1 | 45.7 | 22.4 | 44.7 | 58.1 | — | — | — | — | |
| PANetbackbone=ResNeXt-101, stage type=Two-stage2020.07 | 42 | 65.1 | 45.7 | 22.4 | 44.7 | 58.1 | — | — | — | — | |
| DynaMaskBackbone=ResNeXt-101-FPN, Sched.=3x, Multi-scale training=true2023.03 | 42 | 62.9 | 46.3 | 23.4 | 44 | 54.8 | — | — | — | — | |
| CenterMask*Backbone=V-99-FPN, epochs=36, GPU=V100, Implementation base=Detectron22019.11 | 41.8 | — | — | 24.4 | 44.4 | 54.3 | 77 | 12.9 | — | — | |
| QueryInstBackbone=ResNet-101-FPN, Sched.=3x, Multi-scale training=true2023.03 | 41.7 | 64.4 | 45.3 | 24.2 | 43.9 | 53.9 | — | — | — | — | |
| QueryInst (300 Queries)Backbone=ResNet-101-FPN, Aug.=640 ~ 800, Epochs=36, FPS=6.1, Stages=6, Queries=3002021.05 | 41.7 | 64.4 | 45.3 | 24.2 | 43.9 | 53.9 | — | — | — | — | |
| Mask TransfinerBackbone=R50-FPN, Type=Query-based2022.08 | 41.6 | 63.9 | 45.5 | 24.2 | 44.6 | 55.2 | — | — | — | — | |
| GCNet (r=16)Backbone=ResNeXt-101, Deformable Convolution=true, FLOPs=547.8G2019.04 | 41.5 | 64.6 | 45 | — | — | — | — | — | — | — | |
| GCNet (r=4)Backbone=ResNeXt-101, Deformable Convolution=true, FLOPs=547.8G2019.04 | 41.5 | 64.6 | 45 | — | — | — | — | — | — | — | |
| GCNetBackbone=ResNeXt-101, Cascade strategy=true, Deformable Convolution (DCN)=true, GC ratio=r16, FLOPs(G)=547.8G2020.12 | 41.5 | 64.6 | 45 | — | — | — | — | — | — | — | |
| GCNetBackbone=ResNeXt-101, Cascade strategy=true, Deformable Convolution (DCN)=true, GC ratio=r4, FLOPs(G)=547.8G2020.12 | 41.5 | 64.6 | 45 | — | — | — | — | — | — | — | |
| PointRendBackbone=ResNeXt-101-FPN, Sched.=3x, Multi-scale training=false2023.03 | 41.4 | 63.3 | 44.8 | 24.2 | 43.9 | 53.2 | — | — | — | — | |
| DynaMaskBackbone=ResNet-101-FPN, Sched.=3x, Multi-scale training=true2023.03 | 41.3 | 62.5 | 45.2 | 22.8 | 43.4 | 54 | — | — | — | — | |
| Cascade R-CNN BaselineBackbone=ResNeXt-101, Deformable Convolution=true, FLOPs=547.5G2019.04 | 41 | 63.9 | 44.3 | — | — | — | — | — | — | — | |
| Cascade R-CNNBackbone=ResNeXt-101, Cascade strategy=true, Deformable Convolution (DCN)=true, FLOPs(G)=547.5G2020.12 | 41 | 63.9 | 44.3 | — | — | — | — | — | — | — | |
| QueryInstBackbone=ResNet-101, #Epoch=362022.10 | 41 | 63.3 | 44.5 | 21.7 | 44.4 | 60.7 | — | — | — | — | |
| SOLQBackbone=R101, Epochs=50, Training scale=multi-scale, Testing scale=single-scale2021.06 | 40.9 | — | — | 22.5 | 43.8 | 54.6 | — | — | — | — | |
| SOLQEpoch=50, end-to-end=true, Backbone=R-1012021.11 | 40.9 | — | — | 22.5 | 43.8 | 54.6 | — | — | — | — | |
| HTCBackbone=R101-FPN, Epochs=36, Training scale=multi-scale, Testing scale=single-scale2021.06 | 40.8 | — | — | 23 | 43.5 | 52.6 | — | — | — | — | |
| HTCBackbone=ResNet-101-FPN, Aug.=480 ~ 800 w/ crop, Epochs=36, FPS=2.52021.05 | 40.8 | 62.6 | 44.3 | 23 | 43.5 | 52.6 | — | — | — | — | |
| Hybrid Task CascadeDCN=true, SyncBN=true2019.01 | 40.7 | 62.8 | 44.2 | 22.2 | 43.1 | 54.4 | — | — | — | — | |
| CenterMask-LiteBackbone=VoVNet-39-FPN, Size=600x, FPS=35.7 (P)2020.04 | 40.7 | — | — | 22.4 | 43.2 | 53.5 | — | — | — | — | |
| Mask TransfinerBackbone=ResNet-101-FPN, Sched.=3x, Multi-scale training=true2023.03 | 40.7 | — | — | 23.1 | 42.8 | 53.8 | — | — | — | — | |
| HTCBackbone=ResNet-101-FPN, Aug.=640 ~ 800, Epochs=36, FPS=2.52021.05 | 40.7 | 62.7 | 44.2 | 23.1 | 43.4 | 52.7 | — | — | — | — | |
| CenterMaskBackbone=V-99-FPN, epochs=36, GPU=V1002019.11 | 40.6 | — | — | 20.1 | 42.8 | 57 | 84 | 11.9 | — | — | |
| QueryInstBackbone=R50-FPN, Epochs=36, Training scale=multi-scale, Testing scale=single-scale2021.06 | 40.6 | — | — | 23.4 | 42.5 | 52.8 | — | — | — | — | |
| QueryInst (300 Queries)Backbone=ResNet-50-FPN, Aug.=640 ~ 800, Epochs=36, FPS=7, Stages=6, Queries=3002021.05 | 40.6 | 63 | 44 | 23.4 | 42.5 | 52.8 | — | — | — | — | |
| DCNv2backbone=ResNext-101, architecture=Mask R-CNN2018.11 | 40.5 | — | — | — | — | — | — | — | — | — | |
| SOTRBackbone=ResNet-101, #Epoch=242022.10 | 40.2 | 61.2 | 43.4 | 10.2 | 59 | 73.1 | — | — | — | — | |
| SIMbackbone=Swin-B-FPN, schedule=3x, supervision=box-supervised2023.03 | 40.2 | 66.9 | 41.3 | 21.1 | 43.5 | 56 | — | — | — | — | |
| GCNet (r=4)Backbone=ResNeXt-101, Deformable Convolution=false, FLOPs=537.3G2019.04 | 40.1 | 62.9 | 43.3 | — | — | — | — | — | — | — | |
| GCNetBackbone=ResNeXt-101, Cascade strategy=true, GC ratio=r4, FLOPs(G)=537.3G2020.12 | 40.1 | 62.9 | 43.3 | — | — | — | — | — | — | — | |
| DCT-MaskBackbone=R101-FPN, Epochs=36, Training scale=multi-scale, Testing scale=single-scale2021.06 | 40.1 | — | — | 22.7 | 42.7 | 51.8 | — | — | — | — | |
| K-NetBackbone=ResNet-101, #Epoch=362022.10 | 40.1 | 62.8 | 43.1 | 18.7 | 42.7 | 58.8 | — | — | — | — | |
| DCT-MaskBackbone=ResNet-101-FPN, Sched.=3x, Multi-scale training=true2023.03 | 40.1 | 61.2 | 43.6 | 22.7 | 42.7 | 51.8 | — | — | — | — | |
| QueryInst (100 Queries)Backbone=ResNet-50-FPN, Aug.=640 ~ 800, Epochs=36, FPS=10.5, Stages=6, Queries=1002021.05 | 40.1 | 62.3 | 43.4 | 23.3 | 42.1 | 52 | — | — | — | — | |
| PANetBackbone=ResNeXt-101, multi-scale training=false2018.03 | 40 | 62.8 | 43.1 | 18.8 | 42.3 | 57.2 | — | — | — | — | |
| GCNet (r=16)Backbone=ResNeXt-101, Deformable Convolution=false, FLOPs=537.2G2019.04 | 40 | 62.9 | 43.1 | — | — | — | — | — | — | — | |
| GCNetBackbone=ResNeXt-101, Cascade strategy=true, GC ratio=r16, FLOPs(G)=537.2G2020.12 | 40 | 62.9 | 43.1 | — | — | — | — | — | — | — | |
| Cascade Mask R-CNNBackbone=R101-FPN, Epochs=36, Training scale=multi-scale, Testing scale=single-scale2021.06 | 40 | — | — | 22.5 | 42.5 | 51.2 | — | — | — | — | |
| Cascade Mask R-CNNBackbone=ResNet-101-FPN, Aug.=480 ~ 800 w/ crop, Epochs=36, FPS=8.72021.05 | 40 | 61.7 | 43.5 | 22.5 | 42.5 | 51.2 | — | — | — | — | |
| ISTRBackbone=R101-FPN, Epochs=36, Training scale=multi-scale, Testing scale=single-scale2021.06 | 39.9 | — | — | 22.8 | 41.9 | 52.3 | — | — | — | — | |
| CondInst + Instance-Unique QueryingBackbone=ResNet-101, #Epoch=122022.10 | 39.9 | 62.7 | 42.4 | 20.8 | 42.3 | 55.7 | — | — | — | — | |
| QueryInstBackbone=R50-FPN, Type=Query-based2022.08 | 39.9 | 62.2 | 43 | 22.9 | 41.7 | 51.9 | — | — | — | — | |
| QueryInst (5 Stage, 100 Queries)Backbone=ResNet-50-FPN, Aug.=640 ~ 800, Epochs=36, FPS=13.5, Stages=5, Queries=1002021.05 | 39.9 | 62.2 | 43 | 22.9 | 41.7 | 51.9 | — | — | — | — | |
| CenterMask*Backbone=R-101-FPN, epochs=36, GPU=V100, Implementation base=Detectron22019.11 | 39.8 | — | — | 21.7 | 42.5 | 52 | 66 | 15.2 | — | — | |
| BCNet + Faster R-CNNBackbone=ResNet-1012021.03 | 39.8 | 61.5 | 43.1 | 22.7 | 42.4 | 51.1 | — | — | — | — | |
| BCNet + Faster R-CNNBackbone=R101-FPN, Type=Two-stage2022.08 | 39.8 | 61.5 | 43.1 | 22.7 | 42.4 | 51.1 | — | — | — | — | |
| Cascade Mask R-CNNBackbone=ResNet-101-FPN, Aug.=640 ~ 800, Epochs=36, FPS=8.72021.05 | 39.8 | 61.6 | 43 | 22.4 | 42.2 | 50.8 | — | — | — | — | |
| DCNv2backbone=ResNet-101, architecture=Mask R-CNN2018.11 | 39.7 | — | — | — | — | — | — | — | — | — | |
| HTCbackbone=ResNet-101, stage type=Two-stage2020.07 | 39.7 | 61.8 | 43.1 | 21 | 42.2 | 53.5 | — | — | — | — | |
| HTCBackbone=ResNet-1012021.03 | 39.7 | 61.8 | 43.1 | 21 | 42.2 | 53.5 | — | — | — | — | |
| HTCBackbone=R50-FPN, Epochs=36, Training scale=multi-scale, Testing scale=single-scale2021.06 | 39.7 | — | — | 22.6 | 42.2 | 50.6 | — | — | — | — | |
| SOLQBackbone=R50, Epochs=50, Training scale=multi-scale, Testing scale=single-scale2021.06 | 39.7 | — | — | 21.5 | 42.5 | 53.1 | — | — | — | — | |
| SOLOv2Backbone=R101-FPN, Epochs=72, Training scale=multi-scale, Testing scale=single-scale2021.06 | 39.7 | — | — | 17.3 | 42.9 | 57.4 | — | — | — | — | |
| SOLOv2backbone=ResNet-101-FPN, schedule=3x, supervision=fully-supervised2023.03 | 39.7 | 60.7 | 42.9 | 17.3 | 42.9 | 57.4 | — | — | — | — | |
| SOLOv2Backbone=ResNet-101-FPN, Sched.=6x, Multi-scale training=false2023.03 | 39.7 | 60.7 | 42.9 | 17.3 | 42.9 | 57.4 | — | — | — | — | |
| HTCBackbone=ResNet-101-FPN, Sched.=20e, Multi-scale training=false2023.03 | 39.7 | 61.8 | 43.1 | 21 | 42.2 | 53.5 | — | — | — | — | |
| HTCBackbone=R101-FPN, Type=Two-stage2022.08 | 39.7 | 61.8 | 43.1 | 21 | 42.2 | 53.5 | — | — | — | — | |
| SOLQBackbone=R50-FPN, Type=Query-based2022.08 | 39.7 | — | — | 21.5 | 42.5 | 53.1 | — | — | — | — | |
| HTCBackbone=ResNet-50-FPN, Aug.=640 ~ 800, Epochs=36, FPS=3.12021.05 | 39.7 | 61.4 | 43.1 | 22.6 | 42.2 | 50.6 | — | — | — | — | |
| MS R-CNNBackbone=ResNet-101 DCN+FPN2019.03 | 39.6 | 60.7 | 43.1 | 18.8 | 41.5 | 56.2 | — | — | — | — | |
| CenterMaskBackbone=X-101-FPN, epochs=36, GPU=V1002019.11 | 39.6 | — | — | 19.7 | 42 | 55.2 | 123 | 8.1 | — | — | |
| BCNet + FCOSBackbone=ResNet-1012021.03 | 39.6 | 61.2 | 42.7 | 22.3 | 42.3 | 51 | — | — | — | — | |
| BlendMaskBackbone=R101-FPN, Epochs=36, Training scale=multi-scale, Testing scale=single-scale2021.06 | 39.6 | — | — | 22.4 | 42.2 | 51.4 | — | — | — | — | |
| HTCBackbone=ResNet-101, #Epoch=202022.10 | 39.6 | 61 | 42.8 | 21.3 | 42.9 | 55 | — | — | — | — | |
| SOLOv2 + Instance-Unique QueryingBackbone=ResNet-101, #Epoch=122022.10 | 39.6 | 60.6 | 43.1 | 20 | 43.3 | 56.9 | — | — | — | — | |
| SOTRBackbone=ResNet-50, #Epoch=242022.10 | 39.6 | 60.7 | 42.6 | 10.3 | 58.7 | 72.1 | — | — | — | — | |
| DynaMaskBackbone=ResNet-101-FPN, Sched.=3x, Multi-scale training=false2023.03 | 39.6 | 61.4 | 42.9 | 21.2 | 42.4 | 53.2 | — | — | — | — | |
| BCNet + FCOSBackbone=R101-FPN, Type=One-stage2022.08 | 39.6 | 61.2 | 42.7 | 22.3 | 42.3 | 51 | — | — | — | — | |
| Hybrid Task CascadeDCN=true2019.01 | 39.5 | 61.3 | 42.8 | 20.9 | 41.8 | 52.7 | — | — | — | — |