# YOLOX-S

> YOLOX-S: one-stage object detector from the YOLOX family. Available in LibreYOLO, the MIT-licensed open-source library (free for commercial use).

- Family: YOLOX (Megvii)
- Parameters: 9.0M
- GFLOPs: 13.5
- Default input size: 640px
- Detection: anchor-free · nms
- Architecture: one-stage · CSPDarknet backbone
- Weight license: Apache-2.0
- Paper-reported mAP@50-95: 40.5%
- Paper: https://arxiv.org/abs/2107.08430
- Original code: https://github.com/Megvii-BaseDetection/YOLOX

## Benchmarks (COCO val2017)

| Hardware | Runtime | mAP@50-95 | FPS | Latency (ms) | VRAM (MB) |
|---|---|---|---|---|---|
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 44.3 | 17.1 | 58.6 | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 44.3 | 16.9 | 59.2 | 87 |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 44.2 | 34.9 | 28.6 | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 44.3 | 26.1 | 38.3 | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 44.3 | 60.6 | 16.5 | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 44.3 | 50.0 | 20.0 | 87 |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 43.4 | 70.9 | 14.1 | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 44.3 | 74.5 | 13.4 | - |

## Usage with LibreYOLO

```python
from libreyolo import LibreYOLO

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

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