How to get data - Triple-N Dataset Documentation

Download Data

Triple-N Dataset at ScienceDB
Our data are hosted on Science Data Bank (ScienceDB) and can be accessed via the following link.

Triple-N Data Package

This package contains four datasets: one processed dataset and three raw datasets 1, 2, and 3.

You can download individual files directly through your web browser, or use FTP clients (e.g., FileZilla) for batch downloading. The hostname, port, and login credentials are provided in the Data File Download panel on the platform.
For raw data, please download the .tar.gz files and extract them into your “SessionFolder” directory. Downloading all raw data (~2.7 TB in total) may take 1–2 days depending on network conditions.
Download notes: Based on our experience, download speed may be limited on a single machine. Using multiple PCs in parallel can significantly accelerate the downloading process. A single account can be used simultaneously on multiple PCs for this purpose.

Raw Session Folder

Due to the platform's file count limitations, each session folder has been compressed into a .tar.gz archive. For example:

    Example file:  
    Raw/SesFolder/240629.tar.gz

Stimuli

The shared1000 can be extracted with NSD Stim Info and Shared1000 Index. For more details see NSD documentation. For convenience, we also provide stimuli resized to 227 * 227 pixels and 72 localizer stimuli in two formats:

    others/img_pool.mat: convenient for direct use within MATLAB during plotting.
    others/StimuliNNN.zip: allows easy access to the raw image files after extraction.

Raw GoodUnit

In our analysis pipeline, spiking data are stored in the GoodUnit files. You can download the sessions and directly load these files in MATLAB to inspect processed spike waveforms, raster data, and PSTHs.
A detailed description is provided in the ProcessedFiles section.

    Example file:  
    Raw/GoodStructV2/GoodUnit_YYMMDD_Subject_NSD1000_LOC_gx.mat

Generating Raw H5 and info.mat

For figure generation in the NNN paper, the GoodUnit files were converted into .h5 format for faster loading, with additional metadata saved in corresponding *_info.mat files. You need to run S0_ConvertMatTOh5.m from here to generate these files. Instructions for working with these files can be found in the code section.

    Example file:  
    Raw/H5FILES/ses01_240629_M1_2.h5 
    Raw/H5FILES/ses01_240629_M1_2_info.mat