SemVideo: Reconstructs What You Watch from Brain Activity via Hierarchical Semantic Guidance
The episode discusses the paper "SemVideo: Reconstructs What You Watch from Brain Activity via Hierarchical Semantic Guidance," which uses fMRI scans to reconstruct videos from brain activity. The hosts detail how SemVideo translates video into three semantic levels—anchor, motion narrative, and holistic summary—to guide a text-to-video model. The research shows high accuracy in video retrieval and improved temporal coherence.