skin-lesion-classifier

Skin Lesion Classifier — EfficientNet-B2

Trained on HAM10000 balanced dataset. 7 skin lesion classes.

Performance

Classes

API Usage

import requests, base64

with open("skin.jpg", "rb") as f:
    encoded = base64.b64encode(f.read()).decode("utf-8")

response = requests.post(
    "https://api-inference.huggingface.co/models/YOUR_USERNAME/skin-lesion-classifier",
    headers={"Authorization": "Bearer hf_YOUR_TOKEN"},
    json={"inputs": encoded}
)
print(response.json())

Research only — not for clinical diagnosis.

---

### Step 4: Download `model.pth` from Google Drive

1. Open **Google Drive** in browser
2. Navigate to `Y2S2/HAM10000_balanced/model_outputs/`
3. Right-click `best_model_final.pth` → **Download**
4. Rename the downloaded file to `model.pth`
5. Move it into your `skin-lesion-hf` folder

Your folder should now look like this:

skin-lesion-hf/ ├── model.pth ← downloaded from Drive ├── config.json ← created above ├── labels.json ← created above ├── handler.py ← created above └── README.md ← created above

---

### Step 5: Upload Files to Hugging Face

1. Go to your repo page: `huggingface.co/YOUR_USERNAME/skin-lesion-classifier`
2. Click **"Files"** tab
3. Click **"Add file"** → **"Upload files"**
4. Drag and drop all 5 files at once
5. Scroll down → commit message: `Add model files`
6. Click **"Commit changes to main"**

> `model.pth` is large (~37MB) — wait for the upload bar to complete fully before committing.

---

### Step 6: Get Your API URL

Once uploaded your API URL is simply:

https://api-inference.huggingface.co/models/YOUR_USERNAME/skin-lesion-classifier

That's it. No extra setup needed.

---

### Step 7: Add to Your Render Backend

In Render dashboard → your backend service → **Environment** tab, add:

HF_TOKEN = hf_YOUR_TOKEN_HERE HF_MODEL_URL = https://api-inference.huggingface.co/models/YOUR_USERNAME/skin-lesion-classifier