Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

VIBE: Video-Input Brain Encoder for fMRI Response Modeling

About

We present VIBE, a two-stage Transformer that fuses multi-modal video, audio, and text features to predict fMRI activity. Representations from open-source models (Qwen2.5, BEATs, Whisper, SlowFast, V-JEPA) are merged by a modality-fusion transformer and temporally decoded by a prediction transformer with rotary embeddings. Trained on 65 hours of movie data from the CNeuroMod dataset and ensembled across 20 seeds, VIBE attains mean parcel-wise Pearson correlations of 0.3225 on in-distribution Friends S07 and 0.2125 on six out-of-distribution films. An earlier iteration of the same architecture obtained 0.3198 and 0.2096, respectively, winning Phase-1 and placing second overall in the Algonauts 2025 Challenge.

Daniel Carlstr\"om Schad, Shrey Dixit, Janis Keck, Viktor Studenyak, Aleksandr Shpilevoi, Andrej Bicanski• 2025

Related benchmarks

TaskDatasetResultRank
Brain response predictionFriends In-distribution s07 (test)
Mean Pearson r0.32
13
Brain response predictionOOD benchmark
Mean Pearson r0.21
13
Showing 2 of 2 rows

Other info

Follow for update