EEG-ImageNet: An Electroencephalogram Dataset and Benchmarks with Image Visual Stimuli of Multi-Granularity Labels

The text introduces EEG-ImageNet, a large-scale EEG dataset with 4,000 images and 16 subjects, five times larger than existing benchmarks. It offers multi-granularity labels for studying visual perception and brain-computer interfaces. The dataset establishes object classification (60% accuracy) and image reconstruction benchmarks, highlighting its potential to advance EEG research and improve machine visual models. EEG-ImageNet addresses the lack of comprehensive datasets by providing a more extensive and diverse resource for tasks like object classification, image reconstruction, and understanding brain activity in response to visual stimuli. The study follows ethical guidelines and explores various models and tasks, with future work focusing on enhancing dataset utility and addressing privacy concerns.