Shulou(Shulou.com)11/24 Report--

CTOnews.com July 5 news, Tsinghua University Shenzhen International Graduate School, Tencent Artificial Intelligence Laboratory (AI Lab) and Pengcheng Laboratory researchers recently published a paper said that the research team developed an image generation model called DreamDiffusion, which can directly generate high-quality images from EEG signals. Related papers were published on arXiv, an online academic paper platform owned by Cornell University.

Many teams have studied ways to generate images from the human brain using text-to-image diffusion models, but most use functional magnetic resonance imaging (fMRI) technology to capture brain activity to generate images. This technique lacks practicality because it requires expert operation and requires expensive and difficult-to-carry fMRI equipment.

EEG, by contrast, is a non-invasive, low-cost way to record electrical activity in the brain, and there are portable commercial products that can easily collect EEG signals. Therefore, the research team proposed a "stable diffusion" image generation method, which can reduce the noise interference of EEG signals and make the pre-training of diffusion model more stable and effective.

The research team showed six subjects 2000 images belonging to 40 different subject categories, and then generated high-quality images by collecting EEG signals from the subjects. In the figure below, the left part of each group labeled GT is the original image, and the right part of the Sample image is the EEG generated image.

To assess the accuracy of the method, the research team qualitatively compared it to another recent, similar study, Brain2Image, which confirmed that its accuracy was significantly higher than the images generated by Brain2Image, thus demonstrating the effectiveness of the method.

CTOnews.com Attaches Paper Address: Click here to go

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.