{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/310f239612f74d39b92072ef537075f9\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/310f239612f74d39b92072ef537075f9-35a146e373c7e0ad.gif","duration":242.999,"title":"DeepSight Part 1","description":"This Loom demonstrates DeepSight, an emotional response simulation engine that converts video into time-specific predicted emotional intensity. It uses Meta’s Tribv2 model to output a fMRI activity heat map, then pre-processes video by segmenting into visually interesting 3 to 7 second timestamps and extracting audio with local Whisper to avoid failures mid-clip. Tribv2 is run on an H100 due to its heavy compute load. On top of Tribv2 outputs, the presenter describes a custom neural network trained on the NeuroEmo dataset, using PCA to reduce features from about 5,000 for classifying each timestamp into five emotions."}