README.md

---
license: apache-2.0
base_model:
- Ultralytics/YOLO11
pipeline_tag: object-detection
tags:
- pytorch
---

## YOLOv11x-Face-Detection

A lightweight face detection model based on YOLO architecture ([YOLOv11 xlarge](https://huggingface.co/Ultralytics/YOLO11)), trained for 100 epochs on the WIDERFACE dataset. It's way more accurate than my [YOLOv11n](https://huggingface.co/AdamCodd/YOLOv11n-face-detection) model, but slower.

It achieves the following results on the evaluation set:

```
==================== Results ====================
Easy   Val AP: 0.9629194049702874
Medium Val AP: 0.9519172409689101
Hard   Val AP: 0.8800338681974709
=================================================
```

YOLO results:

![Yolov11x results](https://huggingface.co/AdamCodd/YOLOv11x-face-detection/resolve/main/results.png)

[Confusion matrix](https://huggingface.co/AdamCodd/YOLOv11x-face-detection/blob/main/confusion_matrix.png):

[[27338 3110]

[12337 0]]

### Usage
```python
from huggingface_hub import hf_hub_download
from ultralytics import YOLO

model_path = hf_hub_download(repo_id="AdamCodd/YOLOv11x-face-detection", filename="model.pt")
model = YOLO(model_path)

results = model.predict("/path/to/your/image", save=True) # saves the result in runs/detect/predict
```

### Limitations

- Performance may vary in extreme lighting conditions
- Best suited for frontal and slightly angled faces
- Optimal performance for faces occupying >20 pixels