Artificial Intelligence and Human Brain Activity Used to Reconstruct High-Resolution Images - TechEBlog

Unlike a brain-computer interface, artificial intelligence can be combined with human brain activity to reconstruct high-resolution images. Researchers from the Graduate School of Frontier Biosciences at Osaka University used a diffusion model (DM) to decode fMRI data from human brains and reconstruct visual experiences, relying on a latent diffusion model (LDM) that many know as Stable Diffusion. Training or fine-tuning of complex deep generative models is not required, as this simple framework reconstructs high-resolution images from functional Magnetic Resonance Imaging (fMRI) signals using Stable Diffusion. The text-to-image conversion process implemented by Stable Diffusion incorporates the semantic information expressed by the conditional text, while simultaneously retaining the appearance of the original image. We show that our proposed method can reconstruct high-resolution images with high fidelity in straightforward fashion, without the need for any additional training and fine-tuning of complex deep-learning models.