fmri-to-image

Project for the NYU Computer Vision Course

This project investigates fMRI-based image reconstruction using the BOLD5000 dataset. We implemented and experimented with 5 different techniques for this task -

1. Multi-Channel Input Feed-Forward Network
2. Single Channel Input Feed-Forward Network
3. Autoencoder based approach
4. Generative Adversarial Network
5. CLIP and Diffusion

Link to all project files (model weights, losses, results) - https://drive.google.com/drive/folders/1F44NhFcpMAsduOUAGUPg84gz3Lh1JvTJ?usp=sharing

Link to presentation - https://docs.google.com/presentation/d/1UDfHlcm3WDm1KE0ps_PxyeImsY9C5RboOu7LR_19k_M/edit?usp=sharing