README.md

metadata

title: Skin Disease Detection Demo
emoji: 🩺
colorFrom: blue
colorTo: pink
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: apache-2.0

🩺 Skin Disease Detection — Demo

A demo Space where users can:

• Upload a skin image (or click a bundled test image) and get an AI classifier's prediction across common skin-lesion categories, plus a plain-language explanation from an LLM.
• Chat with an assistant about skin-health topics in general terms.

⚠️ This is an educational demo, not a medical device. It does not diagnose anyone. Always see a licensed dermatologist or doctor for real concerns.

How it works

Component	What it does	Default model
Image classifier	Predicts a skin-lesion category from a photo	Anwarkh1/Skin_Cancer-Image_Classification (ViT, trained on a HAM10000-derived dataset)
Chat / explanation	Turns predictions into plain language, answers general questions	HuggingFaceH4/zephyr-7b-beta via the Hugging Face Inference API

Both model IDs are configurable via Space variables — see below.

🚀 Deploy in 3 steps

1. Create a new Space Go to huggingface.co/new-space → choose Gradio as the SDK → CPU basic hardware is fine for the classifier.

2. Upload these files Upload app.py, requirements.txt, this README.md, and the examples/ folder (if you generated test images — see below) to the Space repo, either via the web UI ("Add file") or:

   git clone https://huggingface.co/spaces/<your-username>/<your-space-name>
   cp -r app.py requirements.txt README.md examples <your-space-name>/
   cd <your-space-name>
   git add . && git commit -m "Initial commit" && git push
   

3. Add an HF_TOKEN secret (to enable the chatbot) In your Space → Settings → Variables and secrets → New secret:

   • Name: HF_TOKEN
   • Value: a Hugging Face access token (create one at huggingface.co/settings/tokens, "Read" scope is enough)

   Without this secret, image classification still works — only the LLM-generated explanations and the chat tab need the token.

That's it — the Space will build and the app will be live.

🖼️ Adding test images (optional but recommended)

So users have something to click without needing their own photo, generate a few sample images once, locally, before you push:

pip install datasets pillow
python scripts/download_examples.py

This saves a handful of JPEGs into examples/. Include that folder when you push to your Space. You can also just drop in your own sample .jpg/.png files instead.

⚙️ Configuration

Set these as Space variables (not secret, unless noted) to customize:

• IMAGE_MODEL_ID — any Hugging Face image-classification model compatible with transformers.pipeline("image-classification", ...).
• CHAT_MODEL_ID — any chat-completion-capable model available via the HF Inference API.
• HF_TOKEN (secret) — required for the chat tab and the LLM explanations.

🩹 Limitations & responsible use

• The bundled classifier was trained on a research dataset (HAM10000-derived dermatoscopic images) and has real, published accuracy limits — treat its output as a talking point, not a result.
• Performance depends heavily on image quality, lighting, and skin tone representation in the training data.
• This app must not be used as a substitute for professional medical evaluation, and the chatbot is instructed to avoid giving diagnoses, treatments, or dosages.

Local development

pip install -r requirements.txt
export HF_TOKEN=your_token_here   # optional, for chat
python app.py