# YOLOv5su

> YOLOv5su: one-stage object detector from the yolov5u family. Planned for LibreYOLO (not yet packaged).

- Family: yolov5u
- Parameters: 9.1M
- GFLOPs: 24.0
- Default input size: 640px
- Detection: anchor-free · nms
- Architecture: one-stage · CSPDarknet backbone
- Weight license: non-permissive
- Paper-reported mAP@50-95: 43%

## Benchmarks (COCO val2017)

| Hardware | Runtime | mAP@50-95 | FPS | Latency (ms) | VRAM (MB) |
|---|---|---|---|---|---|
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 42.4 | 3.6 | 275.9 | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 42.4 | 22.8 | 44.0 | 66 |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 42.4 | 42.8 | 23.4 | 14 |
| Raspberry Pi 5 | ncnn FP32 | 42.4 | 6.2 | 162.5 | - |
| Raspberry Pi 5 | ONNX Runtime FP32 | 42.4 | 2.8 | 362.4 | - |
| Raspberry Pi 5 | PyTorch FP32 | 42.4 | 2.5 | 397.4 | - |


Source: https://www.visionanalysis.org/model/yolov5su. Benchmarks produced with LibreYOLO (https://github.com/Libre-YOLO/libreyolo).