# PicoDet-L

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

- Family: PicoDet (Baidu PaddlePaddle)
- Parameters: 3.3M
- GFLOPs: 8.9
- Default input size: 640px
- Detection: anchor-free · nms
- Architecture: one-stage · ESNet backbone
- Weight license: Apache-2.0
- Paper-reported mAP@50-95: 40.9%
- Paper: https://arxiv.org/abs/2111.00902
- Original code: https://github.com/PaddlePaddle/PaddleDetection

## Benchmarks (COCO val2017)

| Hardware | Runtime | mAP@50-95 | FPS | Latency (ms) | VRAM (MB) |
|---|---|---|---|---|---|
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 43.6 | 4.3 | 232.1 | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 44.1 | 11.1 | 90.1 | 63 |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 44.0 | 19.1 | 52.3 | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 44.1 | 16.4 | 60.8 | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 44.1 | 34.3 | 29.2 | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 44.1 | 39.5 | 25.3 | 72 |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 44.0 | 38.0 | 26.3 | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 44.1 | 32.0 | 31.3 | - |

## Usage with LibreYOLO

```python
from libreyolo import LibreYOLO

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

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