Instance Segmentation on COCO (test-dev)
58.4APMMask Frozen-DINO-DETR
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Mask Frozen-DINO-DETRbackbone=FocalNet-L, epochs=6, Object365 pre-training=true2023.08 | 58.4 | — | — | — | 37.8 | 70.4 | 55.3 | — | — | — | 79.3 | 61.4 | |
| FastInst-D3Backbone=R50-d-DCN, Epochs=50, Resolution=640, FPS=32.5, Scale=single-scale2023.03 | 57.9 | 40.5 | 62.6 | 42.9 | 11.9 | 76.7 | — | — | — | — | — | — | |
| FastInst-D3Backbone=R101, Epochs=50, Resolution=640, FPS=28, Scale=single-scale2023.03 | 57.1 | 39.9 | 61.5 | 42.3 | 11.4 | 76.6 | — | — | — | — | — | — | |
| Mask2FormerBackbone=R101, Epochs=50, Resolution=640, FPS=21.1, Scale=single-scale, Note=Light version with PPM-FPN2023.03 | 56.3 | 39.5 | 61.7 | 41.6 | 11.2 | 75.8 | — | — | — | — | — | — | |
| FastInst-D3Backbone=R50, Epochs=50, Resolution=640, FPS=35.5, Scale=single-scale2023.03 | 56.2 | 38.6 | 60.2 | 40.6 | 10.8 | 75.2 | — | — | — | — | — | — | |
| FastInst-D1Backbone=R101, Epochs=50, Resolution=640, FPS=35.3, Scale=single-scale2023.03 | 55.8 | 38.3 | 60.2 | 40.5 | 10.7 | 74.8 | — | — | — | — | — | — | |
| Mask2FormerBackbone=R50, Epochs=50, Resolution=640, FPS=25.3, Scale=single-scale, Note=Light version with PPM-FPN2023.03 | 54.9 | 38 | 60.3 | 39.8 | 10.8 | 74.3 | — | — | — | — | — | — | |
| FastInst-D1Backbone=R50-d-DCN, Epochs=50, Resolution=576, FPS=47.8, Scale=single-scale2023.03 | 54.9 | 38 | 59.7 | 39.9 | 10 | 74.5 | — | — | — | — | — | — | |
| Mask2Formerbackbone=Swin-L, epochs=100, Object365 pre-training=false2023.08 | 53.8 | — | — | — | 29.1 | 71.2 | 50.5 | — | — | — | 74.9 | 54.9 | |
| ViT-Adapter-LBackbone=ViT-L2022.05 | 53 | — | — | — | — | — | — | — | — | — | — | — | |
| FastInst-D1Backbone=R50, Epochs=50, Resolution=576, FPS=53.8, Scale=single-scale2023.03 | 53 | 35.6 | 56.7 | 37.2 | 8.8 | 72.8 | — | — | — | — | — | — | |
| SwinV2-LBackbone=SwinV2-L2022.05 | 52.7 | — | — | — | — | — | — | — | — | — | — | — | |
| SOITbackbone=Swin-L, epochs=36, Object365 pre-training=false2023.08 | 52.7 | — | — | — | 30.2 | 65.2 | 49.2 | — | — | — | 74.3 | 53.5 | |
| QueryInstBackbone=Swin-L2022.12 | 52.6 | 48.9 | — | — | 30.8 | 68.3 | — | — | — | — | — | — | |
| CB-Swin-LBackbone=Swin-L2022.05 | 52.3 | — | — | — | — | — | — | — | — | — | — | — | |
| QueryInstbackbone=Swin-L, epochs=50, Object365 pre-training=false2023.08 | 51.8 | — | — | — | 31.5 | 63.2 | 49.1 | — | — | — | 74.2 | 53.8 | |
| DiffusionInstBackbone=Swin-L, Denoise steps=one-step, Predefined filters=1002022.12 | 51.5 | 48.3 | — | — | 29.6 | 64 | — | — | — | — | — | — | |
| Focal-LDetection Method=HTC++, Params=265M, Multi-scale evaluation=true2021.07 | 51.3 | — | — | — | — | — | — | — | — | — | — | — | |
| Swin-LDetection Method=HTC++, Params=284M, Multi-scale evaluation=true2021.07 | 51.1 | — | — | — | — | — | — | — | — | — | — | — | |
| DetectoRSBackbone=ResNeXt-101-64x4d, TTA=true2020.06 | 50.9 | — | 72 | 53.3 | 31.6 | 61.5 | 48.5 | — | — | — | — | — | |
| DetectoRSCategory=Pixel-based, Backbone=X-101-64x4d, Epoch=40, multi-scale testing=true2021.04 | 50.9 | 48.5 | 72 | 53.3 | 31.6 | 61.5 | — | — | — | — | — | — | |
| DiffusionInstBackbone=Swin-B, Denoise steps=one-step, Predefined filters=1002022.12 | 50.5 | 47.6 | — | — | 28.3 | 63.4 | — | — | — | — | — | — | |
| Swin-LDetection Method=HTC++, Params=284M, FLOPs=1470G, Multi-scale evaluation=false2021.07 | 50.2 | — | — | — | — | — | — | — | — | — | — | — | |
| SOLQBackbone=Swin-L2022.12 | 50.1 | 46.7 | — | — | 29.2 | 60.9 | — | — | — | — | — | — | |
| SOLQbackbone=Swin-L, epochs=50, Object365 pre-training=false2023.08 | 50.1 | — | — | — | 29.2 | 60.9 | 46.7 | — | — | — | — | — | |
| DetectoRSBackbone=ResNeXt-101-32x4d, TTA=true2020.06 | 49.5 | — | 71.1 | 51.6 | 30.3 | 59.6 | 47.1 | — | — | — | — | — | |
| Our Team 2017ensemble=true2018.03 | 49.1 | 46.7 | 69.5 | 51.3 | 26 | 64 | — | — | — | — | — | — | |
| PANetSettings=baseline + DCN + testing tricks + larger model + ensemble2018.03 | 49.1 | 46.7 | 69.5 | 51.3 | 26 | 64 | — | — | — | — | — | — | |
| DetectoRSBackbone=ResNeXt-101-32x4d, TTA=false2020.06 | 48.7 | — | 69.2 | 50.1 | 27.4 | 59.6 | 45.8 | — | — | — | — | — | |
| DetectoRSBackbone=ResNet-50, TTA=true2020.06 | 48.3 | — | 69.8 | 50.1 | 29.2 | 58.2 | 45.8 | — | — | — | — | — | |
| DetectoRSBackbone=ResNet-50, TTA=false2020.06 | 47.5 | — | 67.7 | 48.3 | 25.6 | 58.3 | 44.4 | — | — | — | — | — | |
| Copy-pasteParams=185M, FLOPs=1440G, Multi-scale evaluation=false2021.07 | 47.4 | — | — | — | — | — | — | — | — | — | — | — | |
| HTC + DCN + multi-scale trainingBackbone=ResNeXt-101-64x4d, TTA=false2020.06 | 47.2 | — | 67.8 | 48.1 | 25.3 | 58.7 | 44.2 | — | — | — | — | — | |
| HTCBackbone=X-101-A2-FPN, Sch.=2x, Multi-scale training=true2021.05 | 47.1 | 44 | 67.5 | 47.9 | 25.7 | 57.6 | — | — | — | — | — | — | |
| ResNeSt-200Multi-scale evaluation=false2021.07 | 47.1 | — | — | — | — | — | — | — | — | — | — | — | |
| PANetSettings=baseline + DCN + testing tricks + larger model2018.03 | 46.5 | 44.4 | 67 | 48.5 | 23.6 | 62.2 | — | — | — | — | — | — | |
| Box2Mask-Tsupervision=box-supervised, backbone=Swin-L2022.12 | 45.9 | 42.4 | 70.2 | 43.3 | 22.1 | 62.9 | — | — | — | — | — | — | |
| GCNet*FLOPs=1041G, Multi-scale evaluation=false2021.07 | 45.4 | — | — | — | — | — | — | — | — | — | — | — | |
| Panoptic SegFormerBackbone=R50, random crops=true2021.09 | 45.3 | 41.7 | — | — | 21.9 | 56.3 | — | — | — | — | — | — | |
| Box2Mask-Tsupervision=box-supervised, backbone=Swin-B2022.12 | 45.1 | 41.5 | 69.2 | 42.1 | 20.8 | 61.7 | — | — | — | — | — | — | |
| SOLOv2category=box-free, backbone=Res-DCN-101-FPN2020.03 | 45 | 41.7 | 63.2 | 45.1 | 18 | 61.6 | — | — | — | — | — | — | |
| SOLOv2Backbone=ResNet-101-DCN, Type=One-stage, Arch=SOLOv2, Sup=Mask, FPS=10.32021.05 | 45 | 41.7 | 63.2 | 45.1 | 18 | 61.6 | — | — | — | — | — | — | |
| BoxTeacherBackbone=Swin-Base-FPN, Schedule=3x, Supervision=Box-supervised2022.10 | 44.9 | 40.6 | 65 | 42.5 | 23.4 | 54.2 | — | — | — | — | — | — | |
| HTCBackbone=X-101-A2-FPN, Sch.=20e2021.05 | 44.8 | 42.1 | 65.3 | 45.7 | 23.6 | 56 | — | — | — | — | — | — | |
| Mask TransfinerBackbone=R101-FPN, Type=Two-stage, DCN=true2021.11 | 44.8 | 42.2 | — | — | 24.1 | 55.4 | — | — | — | — | — | — | |
| Mask TransfinerBackbone=R101-FPN, Output Size=112x112, Hardware=Titan RTX GPU2021.11 | 44.8 | 42.2 | — | — | 24.1 | 55.4 | — | 6.1 | — | — | — | — | |
| PANetBackbone=X-101-PAFPN, Multi-scale training=true2021.05 | 44.7 | 42 | 65.1 | 45.7 | 22.4 | 58.1 | — | — | — | — | — | — | |
| Mask TransfinerBackbone=R50-FPN, Type=Query-based, DCN=false2021.11 | 44.6 | 41.6 | — | — | 24.2 | 55.2 | — | — | — | — | — | — | |
| PANetSettings=baseline + DCN + testing tricks2018.03 | 44.4 | 42 | 63.5 | 46 | 21.8 | 58.1 | — | — | — | — | — | — | |
| DiffusionInstBackbone=ResNet-101, Denoise steps=one-step, Predefined filters=1002022.12 | 44.3 | 41.5 | — | — | 22.9 | 55.2 | — | — | — | — | — | — | |
| HTCBackbone=R101-FPN, Output Size=28x28, DCN=true, Training Schedule=3x, Hardware=Titan RTX GPU2021.11 | 44.2 | 41.7 | — | — | 23.3 | 53.8 | — | 2.1 | — | — | — | — | |
| BlendMask* +deform convs (interval = 3)Backbone=R-101, Epochs=60, Aug.=true, Time (ms)=105.02020.01 | 44.1 | 41.3 | 63.1 | 44.6 | 22.7 | 54.5 | — | — | — | — | — | — | |
| HTCBackbone=ResNeXt-101-64x4d, TTA=false2020.06 | 44 | — | 63.9 | 44.8 | 22.7 | 54.7 | 41.3 | — | — | — | — | — | |
| PointRendBackbone=R101-FPN, Output Size=224x224, DCN=true, Training Schedule=3x, Hardware=Titan RTX GPU2021.11 | 44 | 41.3 | — | — | 20.6 | 55.3 | — | 7.2 | — | — | — | — | |
| HTCBackbone=ResNeXt-101-FPN, runtime (fps)=2.12019.01 | 43.9 | — | 63.9 | 44.7 | 22.8 | 54.6 | 41.2 | — | — | — | — | — | |
| HTCCategory=Pixel-based, Backbone=X-101-64x4d, Epoch=202021.04 | 43.9 | 41.2 | 63.9 | 44.7 | 22.8 | 54.6 | — | — | — | — | — | — | |
| HTCBackbone=X-101-FPN, Sch.=20e2021.05 | 43.9 | 41.2 | 63.9 | 44.7 | 22.8 | 54.6 | — | — | — | — | — | — | |
| HTCBackbone=X-101-FPN, Sch.=20e, Re-implementation status=true2021.05 | 43.9 | 41.3 | 64 | 44.8 | 22.7 | 54.8 | — | — | — | — | — | — | |
| BCNetBackbone=R101-FPN, Output Size=28x28, DCN=true, Training Schedule=3x, Hardware=Titan RTX GPU2021.11 | 43.9 | 41.2 | — | — | 23.6 | 52.8 | — | 8.9 | — | — | — | — | |
| QueryInstsupervision=mask-supervised, backbone=ResNet-1012022.12 | 43.9 | 41.7 | 64.4 | 45.3 | 24.2 | 53.9 | — | — | — | — | — | — | |
| QueryInstBackbone=ResNet-1012022.12 | 43.9 | 41.7 | — | — | 24.2 | 53.9 | — | — | — | — | — | — | |
| Panoptic SegFormerBackbone=R50, random crops=false2021.09 | 43.8 | 40.4 | — | — | 21.1 | 54.7 | — | — | — | — | — | — | |
| SOLQBackbone=ResNet-1012022.12 | 43.8 | 40.9 | — | — | 22.5 | 54.6 | — | — | — | — | — | — | |
| D-SOLObackbone=Res-DCN-101-FPN, framework_type=ours2019.12 | 43.6 | 40.5 | 62.4 | 43.7 | 17.7 | 59.3 | — | — | — | — | — | — | |
| HTCBackbone=R-101-A2-FPN, Sch.=20e2021.05 | 43.5 | 40.8 | 63.6 | 44.1 | 22.3 | 54.4 | — | — | — | — | — | — | |
| Standard TransformerBackbone=R101-FPN, Output Size=56x56, Hardware=Titan RTX GPU2021.11 | 43.5 | 41.3 | — | — | 23.4 | 53.2 | — | 1.4 | — | — | — | — | |
| SOLO +deform convsBackbone=R-101, Epochs=72, Aug.=true2020.01 | 43.3 | 40.4 | 62.7 | 43.3 | 17.6 | 58.9 | — | — | — | — | — | — | |
| HTCBackbone=ResNeXt-101-32x4d, TTA=false2020.06 | 43.3 | — | 63.2 | 44.1 | 22 | 54.2 | 40.7 | — | — | — | — | — | |
| K-Net-N256Backbone=R-101-FPN, Box-free=true, NMS-free=true, Epochs=36, FPS=15.5, P. (M)=56.292021.06 | 43.3 | 40.6 | 63.3 | 43.7 | 18.8 | 59 | — | — | — | — | — | — | |
| K-Net-N256backbone=R101, epochs=36, Object365 pre-training=false2023.08 | 43.3 | — | — | — | 18.8 | 59 | 40.6 | — | — | — | 63.3 | 43.7 | |
| D2DetBackbone=ResNet101-FPN, Input Size=~ 1333 x 800, Time=168, Stage Category=Two-Stage2020.07 | 43 | 40.2 | 61.5 | 43.7 | 21.7 | 54 | — | — | — | — | — | — | |
| SOLOv2Backbone=R-101-FPN, Box-free=true, Epochs=36, FPS=14.3, P. (M)=65.362021.06 | 43 | 39.5 | 60.8 | 42.6 | 16.7 | 57.4 | — | — | — | — | — | — | |
| SOLOv2category=box-free, backbone=Res-101-FPN2020.03 | 42.9 | 39.7 | 60.7 | 42.9 | 17.3 | 57.4 | — | — | — | — | — | — | |
| SOLOv2supervision=mask-supervised, backbone=ResNet-1012022.12 | 42.9 | 39.7 | 60.7 | 42.9 | 17.3 | 57.4 | — | — | — | — | — | — | |
| SOLOv2Backbone=ResNet-1012022.12 | 42.9 | 39.7 | — | — | 17.3 | 57.4 | — | — | — | — | — | — | |
| SOLOv2Backbone=R-101-FPN, Schedule=6x, Supervision=Mask-supervised2022.10 | 42.9 | 39.7 | 60.7 | 42.9 | 17.3 | 57.4 | — | — | — | — | — | — | |
| SOLOv2Backbone=R101, Epochs=36, Resolution=800, FPS=15.7, Scale=single-scale2023.03 | 42.9 | 39.7 | 60.7 | 42.9 | 17.3 | 57.4 | — | — | — | — | — | — | |
| Cascade Mask R-CNNBackbone=ResNeXt-101-FPN, runtime (fps)=2.52019.01 | 42.8 | — | 62.7 | 43.4 | 22 | 52.9 | 40.1 | — | — | — | — | — | |
| Mask TransfinerBackbone=R101-FPN, Type=Two-stage, DCN=false2021.11 | 42.8 | 40.7 | — | — | 23.1 | 53.8 | — | — | — | — | — | — | |
| Mask R-CNNBackbone=R101-FPN, Output Size=28x28, DCN=true, Training Schedule=3x, Hardware=Titan RTX GPU2021.11 | 42.8 | 39.4 | — | — | 18.6 | 54.5 | — | 9.6 | — | — | — | — | |
| K-NetBackbone=R-101-FPN, Box-free=true, NMS-free=true, Epochs=36, FPS=16.2, P. (M)=56.252021.06 | 42.7 | 40.1 | 62.8 | 43.1 | 18.7 | 58.8 | — | — | — | — | — | — | |
| MS R-CNNBackbone=R101-FPN, Type=Two-stage, DCN=true2021.11 | 42.7 | 39.6 | — | — | 18.9 | 55.1 | — | — | — | — | — | — | |
| Mask Scoring R-CNNBackbone=R101-FPN, Output Size=28x28, DCN=true, Training Schedule=3x, Hardware=Titan RTX GPU2021.11 | 42.7 | 39.6 | — | — | 18.9 | 55.1 | — | 9.2 | — | — | — | — | |
| K-Netsupervision=mask-supervised, backbone=ResNet-1012022.12 | 42.7 | 40.1 | 62.8 | 43.1 | 18.7 | 58.8 | — | — | — | — | — | — | |
| PointRendBackbone=R101-FPN, Type=Two-stage, DCN=false2021.11 | 42.6 | 39.6 | — | — | 19.8 | 53.7 | — | — | — | — | — | — | |
| Cascade Mask R-CNNBackbone=R101-FPN, Output Size=28x28, DCN=true, Training Schedule=3x, Hardware=Titan RTX GPU2021.11 | 42.6 | 41.5 | — | — | 22.1 | 54.2 | — | 4.8 | — | — | — | — | |
| Cascade R-CNN + InstaBoost (map guided)Backbone=Res-101-FPN, Augmentation Strategy=appearance consistency heatmap guided InstaBoost (map guided)2019.08 | 42.5 | 39.5 | 61.4 | 42.9 | 21.2 | 52.1 | — | — | — | — | — | — | |
| SOLQBackbone=R50-FPN, Type=Query-based, DCN=false2021.11 | 42.5 | 39.7 | — | — | 21.5 | 53.1 | — | — | — | — | — | — | |
| SOLQBackbone=R502021.09 | 42.5 | 39.7 | — | — | 21.5 | 53.1 | — | — | — | — | — | — | |
| QueryInstBackbone=R502021.09 | 42.5 | 40.6 | — | — | 23.4 | 52.8 | — | — | — | — | — | — | |
| SOLQBackbone=ResNet-502022.12 | 42.5 | 39.7 | — | — | 21.5 | 53.1 | — | — | — | — | — | — | |
| HTCBackbone=R-50-A2-FPN, Sch.=20e2021.05 | 42.4 | 39.8 | 62.3 | 43 | 21.6 | 52.8 | — | — | — | — | — | — | |
| BCNetBackbone=R101-FPN, Type=Two-stage, DCN=false2021.11 | 42.4 | 39.8 | — | — | 22.7 | 51.1 | — | — | — | — | — | — | |
| PANetBackbone=X-101-PAFPN2021.05 | 42.3 | 40 | 62.8 | 43.1 | 18.8 | 57.2 | — | — | — | — | — | — | |
| Mask TransfinerBackbone=R50-FPN, Type=Two-stage, DCN=true2021.11 | 42.3 | 40.5 | — | — | 22.8 | 52.5 | — | — | — | — | — | — | |
| PANetSettings=baseline + DCN2018.03 | 42.2 | 39.5 | 62 | 42.8 | 19.8 | 54.7 | — | — | — | — | — | — | |
| HTCBackbone=ResNet-101-FPN, runtime (fps)=2.42019.01 | 42.2 | — | 61.8 | 43.1 | 21 | 53.5 | 39.7 | — | — | — | — | — | |
| BlendMask*Backbone=R-101, Epochs=36, Aug.=true, Time (ms)=94.12020.01 | 42.2 | 39.6 | 61.6 | 42.6 | 22.1 | 51.4 | — | — | — | — | — | — | |
| HTCBackbone=ResNet101-FPN, Input Size=~ 1333 x 800, Time=417, Stage Category=Two-Stage2020.07 | 42.2 | 39.7 | 61.8 | 43.1 | 21 | 53.5 | — | — | — | — | — | — | |
| LSNetCategory=Contour-based, Backbone=R2-101-DCN, Epoch=30, multi-scale testing=true2021.04 | 42.2 | 40.2 | 66.2 | 42.1 | 25.8 | 51 | — | — | — | — | — | — |