SPGN_Generated_Dataset

Recent advancements in multimodal learning have revolutionized text, video, and audio processing, yet EEG research lags due to data scarcity from specialized e- quipment and privacy risks in personal signal sharing. These limitations, coupled with the shortcomings of prior generative models that produce signals lacking spa- tiotemporal coherence, biological plausibility, and stimulus-response alignment, hinder the development of EEG-based applications, such as emotion analysis and brain-computer interfaces, by restricting access to diverse, high-quality data. The absence of a dedicated task for modeling the mapping from naturalistic video s- timuli to personalized EEG responses has impeded progress in privacy-preserving EEG synthesis. To advance the field, we propose the task of stimulus-/subject- conditional EEG generation under naturalistic stimulation, which is crucial for enabling low-cost, scalable data generation while addressing ethical concerns. To support this task, we introduce a novel multimodal, alignment-based dataset built on SEED-DV, featuring over 1000 aligned video-EEG samples that synchro- nize video features with EEG dynamics. We further propose the Video2EEG- SPGN-Diffusion framework, a graph-enhanced diffusion model tailored for video- conditioned EEG generation, providing tools for emotion analysis, data augmen- tation, and brain-computer interfaces. We further establish a dedicated evaluation system to assess EEG generation quality in dynamic visual perception tasks. In the SEED-DV visual stimulus task, SPGN achieved a signal stability index of 0.9363 and a comprehensive performance index of 0.9373.