---
license: apache-2.0
tags:
  - coreml
  - yolo
  - pose-estimation
  - tongue
  - medical
  - ios
language: []
pipeline_tag: image-to-image
---

# OroSense — Lateralization Model (CoreML)

YOLO11n-pose model specialized for **tongue lateralization** detection (left and right). Trained on frontal oral-motor recordings from a clinical tongue ROM dataset.

## Model Details

| Property | Value |
|---|---|
| Architecture | YOLO11n-pose |
| Input size | 1280 × 1280 |
| Keypoints | 4 |
| Format | CoreML (.mlpackage) |
| Training tasks | `latR`, `latL` |
| Epochs | 50 |
| Augmentation | Heavy (rotation, scale, HSV, mosaic, mixup) |

## Keypoints

| Index | Name | Description |
|---|---|---|
| 0 | left_commissure | Left corner of the mouth (MediaPipe LM 61) |
| 1 | right_commissure | Right corner of the mouth (MediaPipe LM 291) |
| 2 | tongue_tip | Annotated tongue tip position |
| 3 | upper_lip | Upper lip center / vermilion border (MediaPipe LM 0) |

## Usage (Swift / iOS)

```swift
import CoreML
import Vision

let model = try best(configuration: MLModelConfiguration())
let request = VNCoreMLRequest(model: try VNCoreMLModel(for: model.model))
// input: 1280×1280 RGB image
```

## Training Metrics (best epoch)

| Metric | Value |
|---|---|
| Box mAP50 | 0.915 |
| Box mAP50-95 | 0.595 |
| Pose mAP50 | 0.960 |
| Pose mAP50-95 | 0.927 |

## Part of OroSense

This model is one of three task-specialized models:
- **[orosense_lat](https://huggingface.co/LucasTakanori/orosense_lat)** — lateralization (this model)
- **[orosense_elev](https://huggingface.co/LucasTakanori/orosense_elev)** — elevation
- **[orosense_side](https://huggingface.co/LucasTakanori/orosense_side)** — side-view protrusion