OroSense Elevation + Lower Lip โ€” 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 ร— 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 ยฑ12ยฐ, scale ยฑ40%, 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

Downloads last month
8
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support