SkinFix ViT Dermatology Classifier
A Vision Transformer (ViT-Base-16) fine-tuned for skin condition classification.
Model Details
| Property | Value |
|---|---|
| Architecture | ViT-Base-16 (torchvision) |
| Input | 224x224 RGB images |
| Output | 4 clinical categories |
| Validation Accuracy | 88.22% |
| Calibration ECE | 0.69% |
| Format | safetensors |
| Size | ~330 MB |
Classes
| ID | Class | Description |
|---|---|---|
| 0 | inflammatory |
Eczema, Psoriasis, Dermatitis, Rosacea |
| 1 | infectious |
Bacterial, Fungal, Viral infections |
| 2 | lesion_like |
BCC, Actinic Keratosis, Seborrheic Keratosis |
| 3 | pigmented |
Melanoma, Nevi, Lentigo |
Usage
from huggingface_hub import hf_hub_download
model_path = hf_hub_download(
repo_id="Megesh03/skinfix-vit-dermatology",
filename="best_model.safetensors"
)
Calibration
Temperature scaling applied (T=1.2584) to ensure predicted confidence matches real-world accuracy within 1%.
Disclaimer
⚠️ This model is for educational/research screening purposes only. It is NOT a replacement for professional medical diagnosis.