# YOLOv8-S

> YOLOv8-S: one-stage object detector from the YOLOv8 family. Planned for LibreYOLO (not yet packaged).

- Family: YOLOv8 (External baseline)
- Parameters: 11.2M
- GFLOPs: 28.6
- Default input size: 640px
- Detection: anchor-free · nms
- Architecture: one-stage · CSPDarknet backbone
- Weight license: non-permissive
- Paper-reported mAP@50-95: 44.9%

## Benchmarks (COCO val2017)

| Hardware | Runtime | mAP@50-95 | FPS | Latency (ms) | VRAM (MB) |
|---|---|---|---|---|---|
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 44.2 | 3.2 | 313.3 | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 44.2 | 22.1 | 45.1 | 79 |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 44.2 | 39.7 | 25.2 | 14 |
| Raspberry Pi 5 | HailoRT INT8 | 36.9 | 68.0 | 14.7 | - |
| Raspberry Pi 5 | ncnn FP32 | 44.2 | 5.7 | 175.5 | - |
| Raspberry Pi 5 | ONNX Runtime FP32 | 44.2 | 2.4 | 414.2 | - |
| Raspberry Pi 5 | PyTorch FP32 | 44.2 | 2.4 | 421.1 | - |


Source: https://www.visionanalysis.org/model/yolov8s. Benchmarks produced with LibreYOLO (https://github.com/Libre-YOLO/libreyolo).