---
license: agpl-3.0
library_name: mlx
tags:
- mlx
- yolo
- yolo11
- omniparser
- object-detection
- apple-silicon
pipeline_tag: object-detection
---

# OmniParser icon_detect — MLX (YOLO11)

**MLX weights of [microsoft/OmniParser-v2.0](https://huggingface.co/microsoft/OmniParser-v2.0)'s `icon_detect`** — a YOLO11m fine-tuned on 67K screenshots to detect interactive UI elements. Runs on Apple Silicon with **no PyTorch / no ultralytics at inference**.

## ⚠️ License: AGPL-3.0
This model derives from **Ultralytics YOLO11**, which is **AGPL-3.0** (strong copyleft). This repo is therefore licensed **AGPL-3.0** — see `LICENSE`. If you build a network service on top of it, AGPL §13 requires you to offer users the corresponding source. For a commercial/proprietary product, consider an [Ultralytics Enterprise License](https://www.ultralytics.com/license) instead.

## Provenance (corresponding source)
- Original detector: microsoft/OmniParser-v2.0 `icon_detect/model.pt` (AGPL-3.0, Ultralytics YOLO11m).
- MLX conversion format + inference code: [yolo11-mlx](https://github.com/walter-grace/yolo11-mlx).
- `omniparser_mlx.json`: `nc=1`, class `icon`.

## Usage
```python
from yolo11_mlx import YOLO11   # pip install from github.com/walter-grace/yolo11-mlx
model = YOLO11("omniparser_mlx.npz")
boxes = model.predict("screenshot.png", conf=0.05, iou=0.5)[0].boxes.xyxy
```