Skip to content

👾 Welcome to ViT-SSL

A PyTorch framework for Self-Supervised and Supervised Learning with Vision Transformers.

ViT-SSL is an educational project designed to demonstrate the core ideas of Self-Supervised Learning (SSL) through modular, readable, and reproducible code.
It’s currently under development and aims to make SSL concepts more accessible to developers and researchers.

This site contains detailed documentation of the project’s structure — explaining what each module does, how components interact, and why certain design decisions were made.

✨ Overview

ViT-SSL provides:
- Implementations of modern SSL methods: DINO, SimMIM, and Supervised baselines
- Customizable training pipelines built with PyTorch
- Clear config-driven design for experimentation
- Metrics and logging tailored for representation learning

⚙️ Installation

  1. Clone the repository: bash git clone https://github.com/kristi700/ViT-SSL.git cd ViT-SSL

  2. Create a virtual environment (recommended): bash python -m venv venv source venv/bin/activate

  3. Install dependencies: bash pip install -r requirements.txt

🚀 Quick Start

Training is unified under a single entry point: train.py.
To switch between methods (e.g., Supervised, DINO, SimMIM, Fine-tuning), simply set the training.type field in the config file.

🛠️ For more on available config options, see Configuration Guide.

🗂️ For a tour of the codebase, see Code Structure.