CSD: Caption Scene Dataset

Abstract

Experimental task Text image semantic matching task: In each trial, participants first read a scene text description for 3 seconds, followed by a 3-second rest period. They then viewed the corresponding scene image for 3 seconds, followed by another 3-second rest. During image presentation, participants judged whether the text description matched the scene and indicated a mismatch by pressing a button.Basic information of the datasetThis study recruited 8 healthy participants and acquired structural images (T1, T2, dMRI) as well as functional images, including visual category localizer scans (fLoc), retinotopic mapping scans (Retinotopic), resting-state scans (Resting), and the main CSD experiment (200 runs in total). In the CSD main experiment, each participant viewed 4,400 text–image pairs, comprising 4,000 matched pairs (1,000 shared stimuli + 1,000 unique stimuli, each repeated twice) and 400 unmatched pairs.Scanning parametersThe CSD main experiment was conducted using a United Imaging uMR890 3T MRI system. Functional images were acquired with the following parameters: isotropic voxel size of 2.5 mm, repetition time (TR) 2,000 ms, echo time (TE) 30 ms, field of view (FOV) 200 × 200 mm, 60 axial slices, flip angle 81°, echo spacing 0.52 ms, bandwidth 2,320 Hz, and multiband acceleration factor of 2.Raw data The raw data are stored in the /sub-xx/ directory and organized into three subfolders:anat: structural images (T1, T2); dwi: diffusion images (dMRI); func: functional images. Each folder contains the raw image files along with corresponding JSON metadata.Preprocessed time-series data The preprocessed time-series data is stored in /derivatives/ppdata/sub-xx/ and include the following components:anat: high signal-to-noise ratio T1 structural image obtained by averaging multiple scans; design: experimental design file for each CSD run, with elements represented as N or 0. Here, N indicates the stimulus index (1-indexed) and marks the onset of image presentation, while 0 denotes other time points; func: preprocessed functional time-series data aligned to the individual’s native volume spaceThe corresponding stimulus index file can be found at /stimuli/COCO_CN/Stimulus_index.tsv.Preprocessed beta data Preprocessed Beta data are stored in /derivatives/pp_betas/sub-xx/. These data are derived from the preprocessed time-series data, projected onto each participant’s native surface space, and calculated using GLMSingle. The data are organized separately for each run and hemisphere, with lh representing the left hemisphere and rh representing the right hemisphere. Each H5 file contains two datasets: beta (Beta weights computed by GLMSingle); beta_zscore (z-score standardized beta values within the same session)Freesurfer reconstruction results Freesurfer reconstruction results are stored in /derivatives/freesurfer_results/sub-xx/. Pycortex database Pycortex database results are stored in /derivatives/pycortex_db/ and can be used for flattened visualizations of individual brain surfaces.StimuliExperimental stimuli are stored in the /stimuli/ directory, including:COCO_CN: stimulus images used in the experiment, corresponding text descriptions, English translations, and stimulus sequence information; CSD: scripts recording stimulus presentation, encoded in GBK