# D-FINE-S

> D-FINE-S: 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: 10.0M
- GFLOPs: 25.0
- Default input size: 640px
- Detection: detr · nms-free
- Architecture: transformer · HGNetv2 backbone
- Weight license: Apache-2.0
- Paper-reported mAP@50-95: 48.7%
- 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 | 50.7 | 24.0 | 41.7 | 140 |
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 53.4 | 10.4 | 96.1 | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 53.4 | 10.3 | 96.8 | 139 |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 53.5 | 30.2 | 33.1 | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 53.4 | 22.4 | 44.7 | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 53.4 | 70.9 | 14.1 | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 53.4 | 34.2 | 29.3 | 142 |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 53.2 | 86.5 | 11.6 | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 53.4 | 98.7 | 10.1 | - |

## Usage with LibreYOLO

```python
from libreyolo import LibreYOLO

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

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