Pyyan / Compare / Faster R-CNN vs RF-DETR
Faster R-CNN vs RF-DETR
2 of 5Object Detection · verified 13 Aug 2026
Faster R-CNNMicrosoftcurrentRF-DETRRoboflowcurrent
| Specification | Faster R-CNN | RF-DETR |
|---|---|---|
| Summary | The two-stage baseline, still a reference point. | First real-time detector past 60 mAP on COCO. |
| COCO mAP | ~42 | 60.5 |
| Speed | Low | Real time |
| Licence | MIT | Apache 2.0 |
| Family | Two stage | Transformer, set prediction |
| Open vocabulary | No | No |
| Released | 2015 | 2025 |
| Category | Object Detection | Object Detection |
| Official | Microsoft ↗ | Roboflow ↗ |
Highlighted rows are where these differ.
Faster R-CNN
- Region proposals then refinement
- Included as the historical baseline
Best for accuracy where latency does not matter.
RF-DETR
- Leads the RF100-VL domain-transfer benchmark
- DINOv2 backbone, markedly better on occluded objects
- Apache 2.0, which YOLO is not
Best for most custom detection tasks.