Object Detection on MS COCO 20 novel categories 2017 (val)
19Novel AP @ 10 ShotsMeta-DETR
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Meta-DETRDetector=DDETR*, Backbone=ResNet101, Note=customized single scale deformable DETR model2021.06 | 19 | 22.2 | 19.7 | 22.8 | |
| DeFRCNDetector=FRCN, Backbone=ResNet1012021.06 | 18.5 | 22.6 | — | — | |
| DETReg-ft-fullDetector=DDETR, Backbone=ResNet50, Fine-tuning regime=ft-full2021.06 | 13.7 | 22.6 | 15.1 | 24.3 | |
| DDETR-ft-fullDetector=DDETR, Backbone=ResNet50, Fine-tuning regime=ft-full2021.06 | 12.4 | 20.4 | 13.3 | 21.8 | |
| DDETR-ft-fullDetector=DDETR*, Backbone=ResNet101, Fine-tuning regime=ft-full, Note=customized single scale deformable DETR model2021.06 | 11.7 | 16.3 | 12.1 | 16.7 | |
| TFADetector=FRCN, Backbone=ResNet1012021.06 | 10 | 13.7 | 9.3 | 13.4 | |
| FRCN-ft-fullDetector=FRCN, Backbone=ResNet101, Fine-tuning regime=ft-full2021.06 | 9.2 | 12.5 | 9.2 | 12 | |
| Meta R-CNNDetector=FRCN, Backbone=ResNet502021.06 | 8.7 | 12.4 | 6.6 | 10.8 | |
| MetaDetDetector=FRCN, Backbone=VGG162021.06 | 7.1 | 11.3 | 6.1 | 8.1 | |
| FRCN-ft-fullDetector=FRCN, Backbone=ResNet50, Fine-tuning regime=ft-full2021.06 | 6.5 | 11.1 | 5.9 | 10.3 | |
| FSRWDetector=YOLOv2, Backbone=DarkNet-192021.06 | 5.6 | 9.1 | 4.6 | 7.6 | |
| FRCN-ft-fullDetector=FRCN, Backbone=VGG16, Fine-tuning regime=ft-full2021.06 | 3.3 | 7.8 | 1.9 | 6 | |
| YOLO-ft-fullDetector=YOLOv2, Backbone=DarkNet-19, Fine-tuning regime=ft-full2021.06 | 3.1 | 1.7 | 7.7 | 6.4 |