theodore-ioann/Skin-Lesion-Segmentation

🧴 Skin Lesion Segmentation Demo

This interactive web application allows you to upload a skin image, select a segmentation model, and visualize the predicted lesion mask with optional post-processing.

🧠 Available Models

• —UNet
• —SegFormer
• —Inception-based CNN
• —KMeans (Unsupervised)
• —Gaussian Mixture Model (Unsupervised)

🔧 Postprocessing Options

• —None
• —Morphological Open
• —Morphological Close
• —Erosion
• —Dilation

💡 How it Works

The model takes in a skin lesion image (resized to 128×128), performs segmentation, and visualizes both:

• —A binary predicted mask
• —A color overlay on the original image

🛠️ Built With

• —PyTorch & torchvision
• —scikit-learn
• —Hugging Face Transformers
• —Gradio
• —OpenCV

Demo created by Theodore Ioannidis (https://github.com/TheodoreIoannidis).