diamandislabii/VIT-tumor-16

Vision Transformer (ViT) Tumor Classification

This repository fine-tune a Vision Transformer (ViT) model using PyTorch for histopathological tumor classification. The model is adapted from the Hugging Face Hub (MahmoodLab/UNI2-h) and configured to classify 16 tumor types.

---

Overview

• Goal: Classify histopathological tumor images into 16 classes.
• Approach: Vision Transformer (ViT) with the following hyperparameters:
  • Image size: 224
  • Patch size: 14
  • Embed dimension: 1536
  • Depth: 24 blocks
  • Number of attention heads: 24
  • MLP ratio: 5.33334 (×2.66667 × 2)
  • Number of classes: 16
• Libraries used:
  • PyTorch
  • timm
  • Hugging Face Hub (for model loading)

---