Object Detection on COCO (mAP, AP50, AP75, APs, APm, APl)
48.3mAPSAGA-S
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SAGA-SPara (M)=34, FLOPs (G)=248, Framework=RetinaNet, Training Schedule=1x2025.09 | 48.3 | 69.6 | 52.2 | 30.9 | 52.6 | 63.8 | |
| Mask RCNN-Swinrole=Teacher, schedule=3x+ms, backbone=Swin2022.07 | 48.2 | 69.8 | 52.8 | 32.1 | 51.8 | 62.7 | |
| SAGA-TPara (M)=24, FLOPs (G)=203, Framework=RetinaNet, Training Schedule=1x2025.09 | 46.2 | 67.8 | 49.5 | 29.6 | 50 | 61.6 | |
| MPViT-XSPara (M)=32, FLOPs (G)=248, Framework=RetinaNet, Training Schedule=1x2025.09 | 45.7 | 57.3 | 48.8 | 28.7 | 49.7 | 59.2 | |
| RMT-TPara (M)=23, FLOPs (G)=199, Framework=RetinaNet, Training Schedule=1x2025.09 | 45.1 | 66.2 | 48.1 | 28.8 | 48.9 | 61.1 | |
| GFL-Res101role=Teacher, schedule=2x+ms, backbone=ResNet-1012022.07 | 44.9 | 63.1 | 49 | 28 | 49.1 | 57.2 | |
| CrossFormer-SPara (M)=41, FLOPs (G)=272, Framework=RetinaNet, Training Schedule=1x2025.09 | 44.4 | 65.8 | 47.4 | 28.2 | 48.4 | 59.4 | |
| CMT-SPara (M)=44, FLOPs (G)=231, Framework=RetinaNet, Training Schedule=1x2025.09 | 44.3 | 65.5 | 47.5 | 27.1 | 48.3 | 59.1 | |
| Agent-PVT-SFLOPs (G)=274, Framework=RetinaNet, Training Schedule=1x2025.09 | 44.1 | 65.3 | 47.3 | 29.2 | 47.5 | 59.8 | |
| PKDschedule=2x+ms, student=FCOS-Res50, teacher=Mask RCNN-Swin2022.07 | 43.9 | 62.3 | 47.5 | 27.2 | 48 | 57.1 | |
| MPViT-XSPara (M)=20, FLOPs (G)=211, Framework=RetinaNet, Training Schedule=1x2025.09 | 43.8 | 65 | 47.1 | 28.1 | 47.6 | 56.5 | |
| PKDschedule=2x+ms, student=FCOS-Res50, teacher=GFL-Res1012022.07 | 43.5 | 61.9 | 47.1 | 26.5 | 47.7 | 55.7 | |
| Pola-PVT-TFramework=RetinaNet, Training Schedule=1x2025.09 | 43.2 | 64.1 | 46.4 | 28 | 46.4 | 57.9 | |
| SOFT++-TinyPara (M)=23, FLOPs (G)=200, Framework=RetinaNet, Training Schedule=1x2025.09 | 41.9 | 62.7 | 44.7 | 27.8 | 45.4 | 55.6 | |
| Swin-TPara (M)=38, FLOPs (G)=248, Framework=RetinaNet, Training Schedule=1x2025.09 | 41.7 | 63.1 | 44.3 | 27 | 45.3 | 54.7 | |
| PKDschedule=2x, student=Retina-Res50, teacher=Mask RCNN-Swin2022.07 | 41.5 | 60.6 | 44.1 | 22.9 | 45.2 | 56.4 | |
| PVTv2-B1Para (M)=23, FLOPs (G)=225, Framework=RetinaNet, Training Schedule=1x2025.09 | 41.2 | 61.9 | 43.9 | 25.4 | 44.5 | 54.3 | |
| POMPEvaluation Mode=Cross-dataset transfer, Source Dataset=Standard LVIS, RPN Pre-training=Standard LVIS2023.04 | 40.3 | 57.9 | 43.6 | 28.3 | 43.9 | 50.6 | |
| Agent-PVT-TFLOPs (G)=211, Framework=RetinaNet, Training Schedule=1x2025.09 | 40.3 | 61.2 | 42.9 | 25.5 | 43.4 | 54.3 | |
| Pola-PVT-TFramework=RetinaNet, Training Schedule=1x2025.09 | 40 | 60.7 | 42.7 | 25 | 43.6 | 52.9 | |
| FCOS-Res50role=Student, schedule=2x+ms, backbone=ResNet-502022.07 | 39.1 | 58.4 | 41.6 | 24 | 42.7 | 48.7 | |
| DeticEvaluation Mode=Cross-dataset transfer, Source Dataset=Standard LVIS, RPN Pre-training=Standard LVIS2023.04 | 38.8 | 56 | 41.9 | 25.6 | 42.2 | 50 | |
| Retina-Res50role=Student, schedule=2x, backbone=ResNet-502022.07 | 37.4 | 56.7 | 39.6 | 20 | 40.7 | 49.7 | |
| DetProEvaluation Mode=Cross-dataset transfer, Source Dataset=Standard LVIS, RPN Pre-training=Standard LVIS2023.04 | 34.9 | 53.8 | 37.4 | 22.5 | 39.6 | 46.3 | |
| ViLDEvaluation Mode=Cross-dataset transfer, Source Dataset=Standard LVIS, RPN Pre-training=Standard LVIS2023.04 | 34.1 | 52.3 | 36.5 | 21.6 | 38.9 | 46.1 |