Reconstructing the Mind's Eye: fMRI-to-Image with Contrastive Learning and Diffusion Priors

TL;DR

MindEye is a novel approach that uses contrastive learning and diffusion priors to accurately reconstruct and retrieve images from fMRI brain activity, achieving state-of-the-art results.

Contribution

It introduces a dual-module system for fMRI-to-image mapping, combining retrieval and reconstruction, with improved training techniques and larger models for enhanced performance.

Findings

Achieves state-of-the-art in image reconstruction and retrieval from fMRI data.

Can accurately retrieve images from large-scale databases like LAION-5B.

Better preserves low-level image features using img2img with autoencoders.

Abstract

We present MindEye, a novel fMRI-to-image approach to retrieve and reconstruct viewed images from brain activity. Our model comprises two parallel submodules that are specialized for retrieval (using contrastive learning) and reconstruction (using a diffusion prior). MindEye can map fMRI brain activity to any high dimensional multimodal latent space, like CLIP image space, enabling image reconstruction using generative models that accept embeddings from this latent space. We comprehensively compare our approach with other existing methods, using both qualitative side-by-side comparisons and quantitative evaluations, and show that MindEye achieves state-of-the-art performance in both reconstruction and retrieval tasks. In particular, MindEye can retrieve the exact original image even among highly similar candidates indicating that its brain embeddings retain fine-grained image-specific information. This allows us to accurately retrieve images even from large-scale databases like LAION-5B. We demonstrate through ablations that MindEye's performance improvements over previous methods result from specialized submodules for retrieval and reconstruction, improved training techniques, and training models with orders of magnitude more parameters. Furthermore, we show that MindEye can better preserve low-level image features in the reconstructions by using img2img, with outputs from a separate autoencoder. All code is available on GitHub.