Dream2Image : An Open Multimodal EEG Dataset for Decoding and Visualizing Dreams with Artificial Intelligence
arXiv (Cornell University) October 3, 2025 Yann Bellec
A new open-access dataset called Dream2Image combines EEG brain recordings, written dream reports, and AI-generated images from 38 participants and over 31 hours of dream EEG. It includes 129 samples with brain activity from the final seconds before awakening (at 15, 30, 60, and 120 seconds prior), raw dream descriptions, and approximate visual reconstructions of dreams. The dataset aims to help researchers study the neural correlates of dreaming, develop models to decode dreams from brain activity, and explore intersections of neuroscience, psychology, and artificial intelligence. Limitations include a relatively small sample size and variability in dream recall, which may affect generalizability.