metadata
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)
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 — lateralization (this model)
- orosense_elev — elevation
- orosense_side — side-view protrusion