# D-FINE-L

> D-FINE-L: transformer object detector from the D-FINE family. Available in LibreYOLO, the MIT-licensed open-source library (free for commercial use).

- Family: D-FINE (USTC)
- Parameters: 31.0M
- GFLOPs: 91.0
- Default input size: 640px
- Detection: detr · nms-free
- Architecture: transformer · HGNetv2 backbone
- Weight license: Apache-2.0
- Paper-reported mAP@50-95: 54%
- Paper: https://arxiv.org/abs/2410.13842
- Original code: https://github.com/Peterande/D-FINE

## Benchmarks (COCO val2017)

| Hardware | Runtime | mAP@50-95 | FPS | Latency (ms) | VRAM (MB) |
|---|---|---|---|---|---|
| NVIDIA A100 | PyTorch FP32 | 57.3 | 17.1 | 58.3 | 245 |
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 60.0 | 5.0 | 199.6 | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 60.0 | 4.7 | 211.7 | 246 |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 60.0 | 18.6 | 53.6 | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 60.0 | 12.5 | 80.2 | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 60.0 | 50.9 | 19.6 | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 60.0 | 21.9 | 45.8 | 245 |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 60.0 | 69.3 | 14.4 | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 60.0 | 72.3 | 13.8 | - |

## Usage with LibreYOLO

```python
from libreyolo import LibreYOLO

model = LibreYOLO("LibreDFINEl.pt")
result = model("image.jpg", conf=0.25, iou=0.45)
```

Source: https://www.visionanalysis.org/model/dfine-l. Benchmarks produced with LibreYOLO (https://github.com/Libre-YOLO/libreyolo).