On an RTX 5070 Ti in PyTorch, D-FINE averages 2.5 mAP points higher than YOLOv9 at matched parameter counts, winning all 3 nearest-params pairings. YOLOv9 is 19.9% faster on average. D-FINE leads on accuracy across the whole range; YOLOv9 wins on raw throughput.
D-FINE fields 5 measured variants and YOLOv9 fields 4, all on the same COCO val2017 protocol on an RTX 5070 Ti. Three variants line up closely by parameter count, giving 3 matched pairs.
| Model | Family | Params (M) | mAP@50-95 | FPS |
|---|---|---|---|---|
| YOLOv9-T | YOLOv9 | 2.0 | 4178.0 | 31.9 |
| D-FINE-N | D-FINE | 3.8 | 4575.0 | 32.5 |
| YOLOv9-S | YOLOv9 | 7.2 | 5045.0 | 30.8 |
| D-FINE-S | D-FINE | 10.3 | 5339.0 | 34.2 |
| D-FINE-M | D-FINE | 19.6 | 5782.0 | 28.8 |
| YOLOv9-M | YOLOv9 | 20.1 | 5610.0 | 37.1 |
| YOLOv9-C | YOLOv9 | 25.5 | 5709.0 | 42.3 |
| D-FINE-L | D-FINE | 31.2 | 5996.0 | 21.9 |
| D-FINE-X | D-FINE | 62.6 | 6143.0 | 18.1 |
Accuracy at matched compute
mAP is shown in percent. Each D-FINE variant is paired with the nearest YOLOv9 variant by parameter count. Across the 3 pairings D-FINE averages 2.5 mAP points higher and wins all three. D-FINE-X tops the range at 61.4 mAP; YOLOv9-C, the largest YOLOv9 variant measured, reaches 57.1. At the small end D-FINE-N (3.78M, 45.8 mAP) faces YOLOv9-T (2.04M, 41.8 mAP).
Speed
Averaged across the matched pairs, YOLOv9 is 19.9% faster than D-FINE in PyTorch, and its lead is largest at the top of the range.
- D-FINE license
- Apache-2.0
- YOLOv9 license
- MIT
- Both families
- permissive, cleared for commercial use
Which family to pick
Pick D-FINE for accuracy per parameter: it wins every matched pairing and reaches 61.4 mAP. Pick YOLOv9 for throughput, at 19.9% faster on average here. D-FINE ships under Apache-2.0 and YOLOv9 under MIT; both are permissive, so licensing does not force the call.
Every number on this page comes from the verified dataset: same 500-image COCO val2017 slice, conf 0.001, IoU 0.6, max 300 detections, pycocotools mAP, identical protocol across all hardware and runtimes. The full protocol is on the methodology page. To rerun this comparison with your own filters, open compare. Accuracy is measured on LibreYOLO retrained checkpoints; other weight sources can yield different values.