An interactive AI-powered web application for exploring the robustness of Vision Transformers through attention map visualization and adversarial attack simulations.
ViT Robustness Lab helps users understand how Vision Transformer models respond to clean and perturbed images. The app provides a visual interface to study attention behavior, compare prediction confidence, and simulate robustness failures in a more interpretable way.
- Vision Transformer robustness analysis
- Real-time attention map visualization
- Adversarial attack simulation
- Clean vs perturbed image comparison
- AI-assisted explanations using Google AI Studio / Gemini
- Interactive frontend built with React, Vite, and TypeScript
- React
- TypeScript
- Vite
- Node.js / Express
- Google Gemini API
- Google AI Studio
- Render
- Google AI Studio (Gemini) for AI-assisted development
ViT-Robustness-Lab/
├── src/
├── server.ts
├── index.html
├── package.json
├── vite.config.ts
├── tsconfig.json
├── .env.example
└── README.md