Human Brain, with a Video Playing Inside It

Field notes

The page turns a brain-encoding pipeline into a manipulable surface: viewers can switch between folded and inflated cortex views, separate the hemispheres, and compare predictions where ground truth exists. Its estimates are generated one second at a time from fused visual, audio, and language features, then shifted five seconds to reflect the sluggish blood-oxygen response. The project identifies TRIBE v2 and the FreeSurfer fsaverage template as its model and mesh sources, and says the computation runs on a local GPU.