Instance Segmentation on LVIS v1.0 (val)
48.3AP (Rare)EVA
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| EVAtest-scale=single-scale2022.11 | 48.3 | — | — | — | — | — | 55 | 48.3 | — | — | — | — | — | — | — | |
| competition winner 2021 fullBackbone=CBNetV2 (2x Swin-L), pre-train=21K, sup2022.03 | 45.4 | 49.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| 2021 competition 1st2022.11 | 45.4 | — | — | — | — | — | 49.2 | 44.1 | — | — | — | — | — | — | — | |
| Distribution Balanced AlgorithmTTA=false, max_per_img=10002021.11 | 43.5 | 45.3 | 46.3 | 44.9 | — | — | — | — | — | — | — | — | — | — | — | |
| DeticBackbone=Swin-B, pre-train=21K, sup; CLIP2022.03 | 41.7 | 41.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeticBackbone=Swin-B, Additional data=ImageNet-L, Input size=896x8962022.01 | 41.7 | 41.7 | 40.8 | 42.6 | — | — | — | — | — | — | — | — | — | — | — | |
| 2020 competition 1st2022.11 | 41.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Distribution Balanced AlgorithmTTA=true, max_per_img=3002021.11 | 41.1 | 45.4 | 46.5 | 45.9 | — | — | — | — | — | — | — | — | — | — | — | |
| ViTDetBackbone=ViT-H, pre-train=1K, MAE2022.03 | 36.9 | 48.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViTDet-H2022.11 | 36.9 | — | — | — | — | — | 48.1 | — | — | — | — | — | — | — | — | |
| Baseline (Box-Supervised)Backbone=Swin-B, Input size=896x8962022.01 | 35.9 | 40.7 | 40.5 | 43.1 | — | — | — | — | — | — | — | — | — | — | — | |
| competition winner 2021 baselineBackbone=CBNetV2 (2x Swin-L), pre-train=21K, sup2022.03 | 34.3 | 43.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViTDetBackbone=ViT-L, pre-train=1K, MAE2022.03 | 34.3 | 46 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7 NAS-FPNInput image size=1280, Augmentation=Copy-Paste2020.12 | 32.1 | 38.1 | 37.1 | 41.9 | — | — | — | — | — | — | — | — | — | — | — | |
| Copy-pasteBackbone=EfficientNet-B7, Testing scale=single-scale2022.01 | 32.1 | 38.1 | 37.1 | 41.9 | — | — | — | — | — | — | — | — | — | — | — | |
| SiameseIMEpochs=16002022.06 | 30.1 | 38.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| lvisTraveler2020.09 | 30 | 41.5 | 41.9 | 46 | — | — | — | — | — | — | — | — | — | — | — | |
| 2020 Challenge winnerBackbone=ResNeSt269+HTC, Protocol=fully-supervised learning with additional tricks2021.04 | 30 | 41.5 | 41.9 | 46 | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7 FPNInput image size=1024, Augmentation=Copy-Paste2020.12 | 29.7 | 36 | 35.8 | 38.9 | — | — | — | — | — | — | — | — | — | — | — | |
| Copy-PasteBackbone=Eff-B7 FPN, pre-train=none (random init)2022.03 | 29.7 | 36 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAEEpochs=16002022.06 | 29.1 | 38.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LVIS Challenge 2020 WinnerTest-time augmentation=None2020.12 | 28.5 | 38.8 | 39.5 | 42.7 | — | — | — | — | — | — | — | — | — | — | — | |
| Tan et al.Backbone=ResNeSt-269, Testing scale=single-scale2022.01 | 28.5 | 38.8 | 39.5 | 42.7 | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncSLLBackbone=ResNeSt-269, Additional data=true, Testing scale=single-scale2022.01 | 27.8 | 36 | 36.7 | 39.6 | — | — | — | — | — | — | — | — | — | — | — | |
| SiameseIMEpochs=4002022.06 | 27.7 | 36.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BARONEnsemble=true, Learned Prompt=true2023.02 | 27.6 | 22.6 | 29.8 | 27.6 | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet-101 FPNInput image size=1024, Augmentation=Copy-Paste2020.12 | 27.4 | 34 | 33.9 | 37.2 | — | — | — | — | — | — | — | — | — | — | — | |
| BARONEnsemble=true, Learned Prompt=false2023.02 | 26.8 | 19.2 | 29.4 | 26.5 | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet-50 FPNInput image size=1024, Augmentation=Copy-Paste2020.12 | 26.5 | 32.3 | 31.8 | 35.3 | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7 FPNInput image size=1024, Augmentation=Standard2020.12 | 26.4 | 33.7 | 33.1 | 37.6 | — | — | — | — | — | — | — | — | — | — | — | |
| Seesaw LossBackbone=ResNeSt-200, Testing scale=single-scale2022.01 | 26.4 | 37.3 | 36.3 | 43.1 | — | — | — | — | — | — | — | — | — | — | — | |
| SupervisedEpochs=3002022.06 | 26.4 | 34.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViLD-ensembleBackbone=EfficientNet-b7, Teacher=ALIGN, weight w=12021.04 | 26.3 | 29.3 | 27.2 | 32.9 | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7 NAS-FPNInput image size=1280, Augmentation=Standard2020.12 | 26 | 34.7 | 33.4 | 39.8 | — | — | — | — | — | — | — | — | — | — | — | |
| MoCo-v3Epochs=600+2022.06 | 25.8 | 35.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAEEpochs=4002022.06 | 25.7 | 36.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViLD*Ensemble=true, Learned Prompt=false2023.02 | 25.6 | 16.8 | 28.5 | 25.2 | — | — | — | — | — | — | — | — | — | — | — | |
| DetProEnsemble=true, Learned Prompt=true2023.02 | 25.6 | 19.8 | 28.9 | 25.9 | — | — | — | — | — | — | — | — | — | — | — | |
| OV-DETREnsemble=false, Learned Prompt=false2023.02 | 25 | 17.4 | 32.5 | 26.6 | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet-101 FPNInput image size=1024, Augmentation=Standard2020.12 | 24.7 | 31.9 | 30.5 | 36.3 | — | — | — | — | — | — | — | — | — | — | — | |
| CenterNet2Backbone=ResNeXt-101, Testing scale=single-scale2022.01 | 24.6 | 34.9 | 34.7 | 42.5 | — | — | — | — | — | — | — | — | — | — | — | |
| ViLDEnsemble=true, Learned Prompt=false2023.02 | 24.6 | 16.6 | 30.3 | 25.5 | — | — | — | — | — | — | — | — | — | — | — | |
| BARONEnsemble=false, Learned Prompt=false2023.02 | 24.4 | 18 | 28.9 | 25.1 | — | — | — | — | — | — | — | — | — | — | — | |
| GOL + AGLU (ours)Backbone=APA*-ResNet101, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 23.6 | — | 31.3 | 33.1 | — | — | — | — | — | — | — | — | — | 30.7 | — | |
| GOL (baseline)Backbone=SE-ResNet101, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 23 | — | 29.9 | 32.5 | — | — | — | — | — | — | — | — | — | 29.7 | — | |
| GOLFramework=Mask-RCNN R101-FPN2022.07 | 22.8 | 29 | 29 | 31.7 | — | — | — | — | — | — | — | — | — | — | — | |
| GOLSampler=RFS, Backbone=MaskRCNN ResNet101, Schedule=2x2022.07 | 22.8 | 29 | 29 | 31.7 | — | — | — | — | — | — | — | — | — | — | — | |
| GOLBackbone=ResNet-101, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 22.8 | — | 29 | 31.7 | — | — | — | — | — | — | — | — | — | 29 | — | |
| Seesaw + PCBBackbone=R-101-FPN2022.01 | 22.6 | 28.8 | 28.3 | 32 | 43.3 | 30.9 | — | — | — | — | — | — | — | — | — | |
| ViLD-text+CLIPBackbone=ResNet-50+ViT-B/32, Style=R-CNN style, Runtime=630x Mask R-CNN style2021.04 | 22.6 | 26.1 | 24.8 | 29.2 | — | — | — | — | — | — | — | — | — | — | — | |
| PCB + SeesawBackbone=ResNet-101, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 22.6 | — | 28.3 | 32 | — | — | — | — | — | — | — | — | — | 28.8 | — | |
| EQL v2framework=Cascade-R101, schedule=2x2020.12 | 22.3 | 28.8 | 27.8 | 32.8 | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet-50 FPNInput image size=1024, Augmentation=Standard2020.12 | 22.2 | 30.3 | 29.5 | 34.7 | — | — | — | — | — | — | — | — | — | — | — | |
| ECMBackbone=ResNet-101, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 21.9 | — | 27.9 | 32.3 | — | — | — | — | — | — | — | — | — | 28.7 | — | |
| ECMFramework=Mask R-CNN, Backbone=R101-FPN, Training Schedule=2x2024.11 | 21.9 | 28.7 | 28.4 | 32.2 | — | — | — | — | — | — | — | — | — | — | 29.4 | |
| ViLD-ensembleBackbone=EfficientNet-b7, Teacher=ViT-L/14, weight w=12021.04 | 21.7 | 29.6 | 29.1 | 33.6 | — | — | — | — | — | — | — | — | — | — | — | |
| MosaicOSBackbone=ResNeXt-101, Additional data=true, Testing scale=single-scale2022.01 | 21.7 | 28.3 | 27.3 | 32.4 | — | — | — | — | — | — | — | — | — | — | — | |
| GOL + AGLU (ours)Backbone=APA*-ResNet50, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 21.6 | — | 29.6 | 31.7 | — | — | — | — | — | — | — | — | — | 29.1 | — | |
| GOLFramework=Mask-RCNN R50-FPN2022.07 | 21.4 | 27.7 | 27.7 | 30.4 | — | — | — | — | — | — | — | — | — | — | — | |
| GOLSampler=RFS, Backbone=MaskRCNN ResNet50, Schedule=2x2022.07 | 21.4 | 27.7 | 27.7 | 30.4 | — | — | — | — | — | — | — | — | — | — | — | |
| GOLBackbone=ResNet-50, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 21.4 | — | 27.7 | 30.4 | — | — | — | — | — | — | — | — | — | 27.7 | — | |
| ROGBackbone=ResNet-101, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 21.1 | — | 29.1 | 31.8 | — | — | — | — | — | — | — | — | — | 28.8 | — | |
| 2DRCLFramework=Mask R-CNN, Backbone=R101-FPN, Training Schedule=2x2024.11 | 21.1 | 28.8 | 28.7 | 32.3 | — | — | — | — | — | — | — | — | — | — | 29.6 | |
| SeesawBackbone=R-101-FPN2022.01 | 21 | 28.2 | 27.8 | 31.8 | 42.7 | 30.2 | — | — | — | — | — | — | — | — | — | |
| NorCal with RFSFramework=Mask-RCNN R101-FPN2022.07 | 20.8 | 27.3 | 26.5 | 31 | — | — | — | — | — | — | — | — | — | — | — | |
| NorCalSampler=RFS, Backbone=MaskRCNN ResNet101, Schedule=2x2022.07 | 20.8 | 27.3 | 26.5 | 31 | — | — | — | — | — | — | — | — | — | — | — | |
| NorCalBackbone=ResNet-101, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 20.8 | — | 26.5 | 31 | — | — | — | — | — | — | — | — | — | 27.3 | — | |
| DeticClassifier=CLIP classifier, Supervision=Image-level2022.01 | 20.7 | 24.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EQL v2framework=Mask-R101, schedule=2x2020.12 | 20.6 | 27.2 | 25.9 | 31.4 | — | — | — | — | — | — | — | — | — | — | — | |
| EQL v2Backbone=R-101-FPN2022.01 | 20.6 | 27.2 | 25.9 | 31.4 | — | — | — | — | — | — | — | — | — | — | — | |
| EQLv2Framework=Mask-RCNN R101-FPN2022.07 | 20.6 | 27.2 | 25.9 | 31.4 | — | — | — | — | — | — | — | — | — | — | — | |
| EQLv2Sampler=random, Backbone=MaskRCNN ResNet101, Schedule=2x2022.07 | 20.6 | 27.2 | 25.9 | 31.4 | — | — | — | — | — | — | — | — | — | — | — | |
| GOL (baseline)Backbone=SE-ResNet50, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 20.6 | — | 28.9 | 30.8 | — | — | — | — | — | — | — | — | — | 28.2 | — | |
| EQL v2Framework=Mask R-CNN, Backbone=R-101-FPN2021.08 | 20.6 | 27.2 | 25.9 | 31.4 | — | — | — | — | — | — | — | — | — | — | — | |
| EQLv2Framework=Mask R-CNN, Backbone=R101-FPN, Training Schedule=2x2024.11 | 20.6 | 27.2 | 25.9 | 31.4 | — | — | — | — | — | — | — | — | — | — | 27.9 | |
| 2DRCLFramework=Mask R-CNN, Backbone=R50-FPN, Training Schedule=2x2024.11 | 20.4 | 27.7 | 27.1 | 31.4 | — | — | — | — | — | — | — | — | — | — | 28.3 | |
| SeeSawFramework=Mask R-CNN, Backbone=R101-FPN, Training Schedule=2x2024.11 | 20.3 | 28.2 | 28.1 | 31.8 | — | — | — | — | — | — | — | — | — | — | 29 | |
| ECM + GAPBackbone=ResNet-50, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 20.1 | — | 26.8 | 30 | — | — | — | — | — | — | — | — | — | 26.9 | — | |
| SeesawFramework=Mask-RCNN R101-FPN2022.07 | 20 | 28.1 | 28 | 31.8 | — | — | — | — | — | — | — | — | — | — | — | |
| SeesawSampler=RFS, Backbone=MaskRCNN ResNet101, Schedule=2x2022.07 | 20 | 28.1 | 28 | 31.8 | — | — | — | — | — | — | — | — | — | — | — | |
| ViLDEnsemble=false, Learned Prompt=false2023.02 | 20 | 16.1 | 28.3 | 22.5 | — | — | — | — | — | — | — | — | — | — | — | |
| SeesawBackbone=ResNet-101, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 20 | — | 28 | 31.8 | — | — | — | — | — | — | — | — | — | 28.1 | — | |
| SeesawBackbone=R-50-FPN2022.01 | 19.8 | 26.8 | 26.3 | 30.5 | 41.3 | 28.4 | — | — | — | — | — | — | — | — | — | |
| ECMBackbone=ResNet-50, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 19.7 | — | 27 | 31.1 | — | — | — | — | — | — | — | — | — | 27.4 | — | |
| ECMFramework=Mask R-CNN, Backbone=R50-FPN, Training Schedule=2x2024.11 | 19.7 | 27.4 | 27 | 31.1 | — | — | — | — | — | — | — | — | — | — | 27.9 | |
| cRTBackbone=ResNeXt-101-32x8d2020.12 | 19.6 | 27.2 | 26 | 31.9 | — | — | — | — | — | — | — | — | — | — | — | |
| SeesawBackbone=ResNet-50, Framework=Mask R-CNN FPN, Schedule=2x2024.07 | 19.6 | — | 26.1 | 29.8 | — | — | — | — | — | — | — | — | — | 26.4 | — | |
| SeeSawFramework=Mask R-CNN, Backbone=R50-FPN, Training Schedule=2x2024.11 | 19.6 | 26.9 | 26.8 | 30.5 | — | — | — | — | — | — | — | — | — | — | 27.3 | |
| LOCEBackbone=R-101-FPN2022.01 | 19.5 | 28 | 27.8 | 32 | — | — | — | — | — | — | — | — | — | — | — | |
| SeesawFramework=Mask-RCNN R50-FPN2022.07 | 19.5 | 26.4 | 26.1 | 29.7 | — | — | — | — | — | — | — | — | — | — | — | |
| LOCEFramework=Mask-RCNN R101-FPN2022.07 | 19.5 | 28 | 27.8 | 32 | — | — | — | — | — | — | — | — | — | — | — | |
| SeesawSampler=RFS, Backbone=MaskRCNN ResNet50, Schedule=2x2022.07 | 19.5 | 26.4 | 26.1 | 29.7 | — | — | — | — | — | — | — | — | — | — | — | |
| LOCESampler=MFS, Backbone=MaskRCNN ResNet101, Schedule=2x2022.07 | 19.5 | 28 | 27.8 | 32 | — | — | — | — | — | — | — | — | — | — | — | |
| LOCEFramework=Mask R-CNN, Backbone=R-101-FPN2021.08 | 19.5 | 28 | 27.8 | 32 | — | — | — | — | — | — | — | — | — | — | — | |
| baseline by host2020.09 | 19.47 | 27.26 | 26.13 | 31.95 | — | — | — | — | — | — | — | — | — | — | — | |
| NorCal with RFSFramework=Mask-RCNN R50-FPN2022.07 | 19.3 | 25.2 | 24.2 | 29 | — | — | — | — | — | — | — | — | — | — | — | |
| NorCalSampler=RFS, Backbone=MaskRCNN ResNet50, Schedule=2x2022.07 | 19.3 | 25.2 | 24.2 | 28.6 | — | — | — | — | — | — | — | — | — | — | — | |
| GOL*Sampler=RFS, Backbone=MaskRCNN ResNet101, Schedule=2x2022.07 | 19.3 | 28 | 27.5 | 32.4 | — | — | — | — | — | — | — | — | — | — | — | |
| BACLBackbone=R-101-FPN2023.08 | 19.3 | — | 27 | 30.9 | — | — | 27.2 | 28.4 | — | — | — | — | — | — | — | |
| Seesaw + PCBBackbone=R-50-FPN2022.01 | 19 | 27.2 | 27.1 | 30.9 | 41.7 | 29.4 | — | — | — | — | — | — | — | — | — |