The BOLD signal tracks the internal temporal structure of these 3-second events, meaning early and late parts of the signal correspond to early and late parts of the video.

Human-written sentence descriptions of the videos correlate more strongly with brain activity than simple labels like "object" or "action".

Video-evoked responses are reliably mapped across occipital, temporal, and parietal cortices.

To help you find more specific details, are you looking for the of the video clips (like frame rate or resolution) or the fMRI processing pipeline used in the paper?

Data and pre-trained models (like the TSM ResNet50 used in the study) are available on GitHub .

The "complete paper" associated with this dataset and its corresponding video clips is: published in Nature Communications (July 2024). Paper & Dataset Overview

The filename maddsmr_shortclip912.mp4 follows the naming schema used in the MAD (Movie Audio Descriptions) or related sub-collections (like Memento10k/MiT) that feed into the BOLD Moments research. Key Findings:

The study provides a benchmark for understanding the neural mechanisms of visual event understanding , bridging the gap between static image perception and long-form movie analysis.