Object Detection on LVIS v0.5 (val)
23.7APrUniDetector
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| UniDetectorTraining data=COCO + O365, Structure=P2023.03 | 23.7 | 22.2 | — | — | — | — | — | 22.5 | 21.2 | — | — | |
| UniDetectorTraining data=COCO + O365 + OImg, Structure=P2023.03 | 23.6 | 23.5 | — | — | — | — | — | 24.3 | 22.4 | — | — | |
| UniDetectorTraining data=COCO + O365 (+pseudo [62]), Structure=U2023.03 | 22.5 | 20.8 | — | — | — | — | — | 22.7 | 19.7 | — | — | |
| UniDetectorTraining data=COCO + O365 (+mosaic), Structure=U2023.03 | 22.3 | 21.4 | — | — | — | — | — | 21.5 | 21 | — | — | |
| UniDetectorTraining data=COCO + O365, Structure=S2023.03 | 22.2 | 21 | — | — | — | — | — | 21.8 | 19.4 | — | — | |
| BACLStrategy=Decoupled, Backbone=ResNet-101-FPN2023.08 | 22.1 | 29.4 | — | — | — | — | — | 30.1 | 31.3 | — | — | |
| UniDetectorTraining data=COCO + OImg, Structure=P2023.03 | 22.1 | 19.9 | — | — | — | — | — | 20.7 | 17.9 | — | — | |
| LOCEStrategy=Decoupled, Backbone=ResNet-101-FPN2023.08 | 21.9 | 27.9 | — | — | — | — | — | 27.7 | 30.5 | — | — | |
| UniDetectorTraining data=OImg, Structure=-2023.03 | 21.8 | 16.8 | — | — | — | — | — | 17.6 | 13.8 | — | — | |
| UniDetectorTraining data=O365, Structure=-2023.03 | 21.3 | 20.2 | — | — | — | — | — | 20.2 | 19.8 | — | — | |
| BACLStrategy=Decoupled, Backbone=ResNet-50-FPN2023.08 | 20.1 | 27.8 | — | — | — | — | — | 28.2 | 30.3 | — | — | |
| UniDetectorTraining data=COCO + O365, Structure=U2023.03 | 19.6 | 20.9 | — | — | — | — | — | 21 | 21.3 | — | — | |
| UniDetectorTraining data=COCO, Structure=-2023.03 | 18.7 | 16.4 | — | — | — | — | — | 17.1 | 14.5 | — | — | |
| EQLv2Strategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 18.3 | 27.4 | — | — | — | — | — | 26.7 | 31.8 | — | — | |
| LOCEStrategy=Decoupled, Backbone=ResNet-50-FPN2023.08 | 18.3 | 26.7 | — | — | — | — | — | 27.5 | 28.9 | — | — | |
| BALMSStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 17.6 | 25.5 | — | — | — | — | — | 25 | 29.3 | — | — | |
| BALMSStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 17.3 | 27.2 | — | — | — | — | — | 27.3 | 31 | — | — | |
| EQLv2Strategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 17.2 | 26.5 | — | — | — | — | — | 26.2 | 30.7 | — | — | |
| SeesawStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 16.8 | 27.3 | — | — | — | — | — | 26.8 | 32 | — | — | |
| BAGSStrategy=Decoupled, Backbone=ResNet-50-FPN2023.08 | 16.8 | 25.5 | — | — | — | — | — | 25.6 | 28.8 | — | — | |
| Forest R-CNNStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 16.6 | 26 | — | — | — | — | — | 26.3 | 29.4 | — | — | |
| ACSLStrategy=Decoupled, Backbone=ResNet-101-FPN2023.08 | 16.5 | 25.7 | — | — | — | — | — | 25.8 | 29.1 | — | — | |
| RIOStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 16.4 | 26.4 | — | — | — | — | — | 26.4 | 30.5 | — | — | |
| SeesawStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 16.3 | 26.3 | — | — | — | — | — | 26.1 | 30.6 | — | — | |
| BAGSStrategy=Decoupled, Backbone=ResNet-101-FPN2023.08 | 16.2 | 26.6 | — | — | — | — | — | 26.7 | 30.7 | — | — | |
| DisAlignStrategy=Decoupled, Backbone=ResNet-101-FPN2023.08 | 15.9 | 27.4 | — | — | — | — | — | 27.8 | 31.5 | — | — | |
| RIOStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 15.7 | 24.4 | — | — | — | — | — | 24 | 28.4 | — | — | |
| Forest R-CNNStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 15.2 | 26.9 | — | — | — | — | — | 27.6 | 30 | — | — | |
| GroupSoftmaxPre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | 15 | — | — | — | — | — | — | 25.5 | 30.4 | 25.8 | — | |
| DisAlignPre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Cosine2021.03 | 14.8 | — | — | — | — | — | — | 27.9 | 32.4 | 27.6 | — | |
| RFSStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 14.8 | 25.9 | — | — | — | — | — | 25.5 | 30.8 | — | — | |
| ACSLStrategy=Decoupled, Backbone=ResNet-50-FPN2023.08 | 14.8 | 23.7 | — | — | — | — | — | 23.5 | 27.5 | — | — | |
| De-confound-TDEStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 14.3 | 27.3 | — | — | — | — | — | 28 | 31.5 | — | — | |
| RFSStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 14.1 | 25 | — | — | — | — | — | 24.8 | 29.6 | — | — | |
| DisAlignStrategy=Decoupled, Backbone=ResNet-50-FPN2023.08 | 14.1 | 25.2 | — | — | — | — | — | 25.2 | 29.5 | — | — | |
| SimCalPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | 13.7 | — | — | — | — | — | — | 20.6 | 28.7 | 22.6 | — | |
| DisAlignPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Cosine2021.03 | 13.7 | — | — | — | — | — | — | 25.6 | 30.5 | 25.6 | — | |
| SimCalStrategy=Decoupled, Backbone=ResNet-50-FPN2023.08 | 13.6 | 22.5 | — | — | — | — | — | 20.3 | 29 | — | — | |
| De-confound-TDEStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 13.2 | 25.3 | — | — | — | — | — | 25.4 | 30 | — | — | |
| RFSPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | 12.8 | — | — | — | — | — | — | 22.3 | 29.4 | 23.6 | — | |
| PCBStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 11.4 | 26.5 | — | — | — | — | — | 26.2 | 32.9 | — | — | |
| FASAStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 11.3 | 24.3 | — | — | — | — | — | 23.2 | 30.9 | — | — | |
| DropLossStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 11.2 | 26.1 | — | — | — | — | — | 28.5 | 29 | — | — | |
| FASAStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 10.7 | 23.6 | — | — | — | — | — | 22.8 | 29.6 | — | — | |
| BaselinePre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Cosine2021.03 | 10.2 | — | — | — | — | — | — | 21.1 | 30.1 | 22.8 | — | |
| BaselinePre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Cosine2021.03 | 10.2 | — | — | — | — | — | — | 23.9 | 32.3 | 25 | — | |
| EQLStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 9.9 | 25.5 | — | — | — | — | — | 26.1 | 31.1 | — | — | |
| DropLossStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 9.7 | 23.3 | — | — | — | — | — | 24.7 | 27.1 | — | — | |
| EQLStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 9.4 | 24 | — | — | — | — | — | 24.4 | 29.2 | — | — | |
| PCBStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 9.1 | 23.9 | — | — | — | — | — | 22.9 | 31.2 | — | — | |
| DisAlignPre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Linear2021.03 | 8.2 | — | — | — | — | — | — | 26.3 | 32.4 | 25.5 | — | |
| BCE*Strategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 7.8 | 25.2 | — | — | — | — | — | 25.2 | 32 | — | — | |
| DisAlignPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Linear2021.03 | 7.5 | — | — | — | — | — | — | 25 | 29.1 | 23.9 | — | |
| BCE*Strategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 6.7 | 23.9 | — | — | — | — | — | 23.8 | 31 | — | — | |
| BCEStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 5.7 | 23.7 | — | — | — | — | — | 23.3 | 31.3 | — | — | |
| BCEStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 5.3 | 21.8 | — | — | — | — | — | 21 | 29.5 | — | — | |
| SCEStrategy=End-to-end, Backbone=ResNet-50-FPN2023.08 | 4 | 22 | — | — | — | — | — | 21.2 | 30.1 | — | — | |
| BaselinePre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Linear2021.03 | 3.3 | — | — | — | — | — | — | 19.5 | 29.4 | 20.8 | — | |
| SCEStrategy=End-to-end, Backbone=ResNet-101-FPN2023.08 | 2.8 | 23.3 | — | — | — | — | — | 23.2 | 31.5 | — | — | |
| BaselinePre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Linear2021.03 | 2.6 | — | — | — | — | — | — | 21.8 | 32 | 22.8 | — | |
| Faster RCNN (closed world)Training data=LVIS v0.52023.03 | 1.9 | 17.7 | — | — | — | — | — | 16.5 | 25.4 | — | — | |
| BAGSBackbone=R-50-FPN2022.01 | — | — | — | — | — | — | — | — | — | — | 25.8 | |
| BAGSFramework=Mask-RCNN R50-FPN2022.07 | — | — | — | — | — | — | — | — | — | — | 25.8 | |
| BAGSFramework=Mask R-CNN, Backbone=R-50-FPN2021.08 | — | — | — | — | — | — | — | — | — | — | 25.8 | |
| Balanced Softmaxvariant=multiple binary logistic regression2020.07 | — | — | — | — | — | — | — | — | — | — | 27 | |
| BALMS2020.07 | — | — | — | — | — | — | — | — | — | — | 27.6 | |
| BALMSFramework=Mask-RCNN R50-FPN2022.07 | — | — | — | — | — | — | — | — | — | — | 27.6 | |
| BALMSFramework=Mask R-CNN, Backbone=R-50-FPN2021.08 | — | — | — | — | — | — | — | — | — | — | 27.6 | |
| BaselineFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | — | — | — | — | — | — | — | — | — | 24.3 | — | |
| CapsuleFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | — | — | — | — | — | — | — | — | — | 27.1 | — | |
| Capsule-TDEFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | — | — | — | — | — | — | — | — | — | 30.4 | — | |
| Class Balanced Loss2020.07 | — | — | — | — | — | — | — | — | — | — | 23.9 | |
| Class-aware SamplingBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | — | — | — | — | — | — | — | — | — | 18.4 | — | |
| Class-aware SamplingBackbone=ResNet-50-FPN2020.08 | — | — | — | — | — | — | — | — | — | — | 18.4 | |
| Class-balanced LossBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | — | — | — | — | — | — | — | — | — | 21 | — | |
| Class-balanced LossBackbone=ResNet-50-FPN2020.08 | — | — | — | — | — | — | — | — | — | — | 21 | |
| CosineFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | — | — | — | — | — | — | — | — | — | 27.1 | — | |
| Cosine-TDEFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | — | — | — | — | — | — | — | — | — | 30.6 | — | |
| De-confoundFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | — | — | — | — | — | — | — | — | — | 27.7 | — | |
| De-confound-TDEFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | — | — | — | — | — | — | — | — | — | 31 | — | |
| DisAlignFramework=Mask-RCNN R50-FPN2022.07 | — | — | — | — | — | — | — | — | — | — | 27.6 | |
| DropLossFramework=Mask-RCNN R50-FPN2022.07 | — | — | — | — | — | — | — | — | — | — | 25.8 | |
| EQLBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | — | — | — | — | — | — | — | — | — | 23.3 | — | |
| EQLFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | — | — | — | — | — | — | — | — | — | 27.9 | — | |
| EQLBackbone=ResNet-50-FPN2020.08 | — | — | — | — | — | — | — | — | — | — | 23.3 | |
| EQLBackbone=ResNet-101-FPN2020.08 | — | — | — | — | — | — | — | — | — | — | 25.2 | |
| EQLPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | — | — | — | — | — | — | — | — | — | 23.3 | — | |
| EQLFramework=Mask R-CNN, Backbone=R-50-FPN2021.08 | — | — | — | — | — | — | — | — | — | — | 24.1 | |
| EQL v2Backbone=R-50-FPN2022.01 | — | — | — | — | — | — | — | — | — | — | 26.5 | |
| EQL v2Framework=Mask R-CNN, Backbone=R-50-FPN2021.08 | — | — | — | — | — | — | — | — | — | — | 27 | |
| EQLv2Framework=Mask-RCNN R50-FPN2022.07 | — | — | — | — | — | — | — | — | — | — | 27 | |
| Equalization Loss2020.07 | — | — | — | — | — | — | — | — | — | — | 25.7 | |
| Focal LossBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | — | — | — | — | — | — | — | — | — | 21.9 | — | |
| Focal Loss2020.07 | — | — | — | — | — | — | — | — | — | — | 23.8 | |
| Focal LossFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | — | — | — | — | — | — | — | — | — | 22.6 | — | |
| Focal LossBackbone=ResNet-50-FPN2020.08 | — | — | — | — | — | — | — | — | — | — | 21.9 | |
| Focal LossPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | — | — | — | — | — | — | — | — | — | 21.9 | — | |
| Forest R-CNNBackbone=ResNet-50-FPN2020.08 | — | — | — | — | — | — | — | — | — | — | 25.9 | |
| Forest R-CNNBackbone=ResNet-101-FPN2020.08 | — | — | — | — | — | — | — | — | — | — | 27.5 | |
| Forest R-CNNBackbone=ResNet-50-FPN, multi-scale training=true2020.08 | — | — | — | — | — | — | — | — | — | — | 27 |