SkinVision AI — EfficientNet-B0 (HAM10000)
EfficientNet-B0 (via timm), fine-tuned for 15 epochs on the HAM10000 dataset to classify skin lesion images into 7 diagnostic categories.
Educational project — not a medical diagnostic tool.
Results (held-out test set, 1,431 images)
| Metric | Score |
|---|---|
| Test accuracy | 82.4% |
| Macro F1 | 0.663 |
| Macro ROC-AUC (OVR) | 0.953 |
Classes
akiec (Actinic Keratosis), bcc (Basal Cell Carcinoma), bkl (Benign Keratosis), df (Dermatofibroma), mel (Melanoma), nv (Melanocytic Nevus), vasc (Vascular Lesion)
Usage
Full inference pipeline, Grad-CAM explainability, and app code: github.com/Pabodha123/Skin-Vision-AI
import timm
import torch
model = timm.create_model("efficientnet_b0", pretrained=False, num_classes=7)
model.load_state_dict(torch.load("best_model.pth", map_location="cpu"))
model.eval()
- Downloads last month
- -