mufasabrownie/glowlytics-skin-models

Glowlytics Skin Analysis Models

ML models for the Glowlytics (RadianceIQ) skin analysis platform.

Models

1. Structure Model (MobileNetV3-Large)

• File: structure_best.pt (PyTorch), structure_model.onnx (ONNX)
• Dataset: FFHQ-256 (10K face images)
• Task: Skin structure/texture analysis
• Training: 30 epochs on A10G GPU

2. Hydration Model (EfficientNet-B0 + Handcrafted Features)

• File: hydration_best.pt (PyTorch), hydration_model.onnx (ONNX)
• Dataset: FFHQ-256 with Gabor/LBP features
• Task: Skin hydration level estimation
• Training: 30 epochs on A10G GPU

3. Elasticity Model (EfficientNet-B0 + Frangi Filter)

• File: elasticity_best.pt (PyTorch), elasticity_model.onnx (ONNX)
• Dataset: UTKFace (23K images, age-to-elasticity mapping)
• Task: Skin elasticity scoring
• Training: ~25 epochs on A10G GPU

4. Lesion Detection (YOLOv8s)

• File: lesion_yolov8_best.pt
• Dataset: 929 acne images, 7K bounding boxes
• Task: Acne/lesion detection and localization
• Training: 50 epochs, mAP50: 0.148
• Results: See lesion_results/ for training curves and confusion matrix

Usage

import torch

# Load a PyTorch model
model = torch.load("structure_best.pt", map_location="cpu")

# Or use ONNX for inference
import onnxruntime as ort
session = ort.InferenceSession("structure_model.onnx")

License

Proprietary - Glowlytics / RadianceIQ