OroSense Elevation + Lower Lip — CoreML (1280px)
YOLO11n-pose model tracking tongue tip, commissures, upper lip, and lower lip during tongue elevation tasks.
Architecture
| Field | Value |
|---|
| Base | YOLO11n-pose |
| Input | 1280 × 1280 |
| Keypoints | 5 |
| Format | CoreML (.mlpackage) |
| Task | elev (tongue elevation) |
Keypoint Schema
| Index | Name | Source |
|---|
| kp0 | left_commissure | MediaPipe LM 61 |
| kp1 | right_commissure | MediaPipe LM 291 |
| kp2 | tongue_tip | Manual annotation |
| kp3 | upper_lip | MediaPipe LM 0 |
| kp4 | lower_lip | MediaPipe LM 17 |
Training
- Dataset: 3833 frames, frontal elevation clips
- Epochs: 50, imgsz: 1280, batch: auto (~21)
- Heavy augmentation: rotation ±12°, scale ±40%, HSV, mosaic 0.8, mixup 0.1
- Preprocessing: CLAHE + unsharp mask
Metrics (train=val, optimistic)
| Metric | Value |
|---|
| Box mAP50 | 0.978 |
| Box mAP50-95 | 0.782 |
| Pose mAP50 | 0.980 |
| Pose mAP50-95 | 0.948 |
Usage (Swift)
Download
Or clone the full package:
Related Models
- orosense_elev — 4-kp elevation model (no lower lip)
- orosense_lat — lateralization
- orosense_side — side-view protrusion
