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 -
- Multi-Channel Input Feed-Forward Network
- Single Channel Input Feed-Forward Network
- Autoencoder based approach
- Generative Adversarial Network
- 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