Official repository for the paper "Brain-Diffuser: Natural scene reconstruction from fMRI signals using generative latent diffusion" by Furkan Ozcelik and Rufin VanRullen.
The following are a few reconstructions obtained :
In this repo is essentially a wrapper for the original paper's architecture and results. Members can use the notebook to run the diffusion model and create a image reconstructed from CLIP image, text weights produced by fMRI data and a initial guess image produced by fMRI data. To run the notebook:
-
Make sure you are logged into the NTAB Software account
-
Open the notebook on colab and select the T4 GPU as your runtime
-
Run all the cells in the order they are descibed, you will have to re-run the first cell 3 times