Overview of the data

Top-level directories

There are several top-level directories:
  • nsddata (~49 GB) - This is the main directory containing essential data files, including (but not limited to) anatomical data, results of the prf and floc experiments, behavioral data, FreeSurfer subject directories, and ROIs.
  • nsddata_betas (~8.3 TB) - This very large folder contains estimated fMRI single-trial responses ("betas") for the NSD experiment as well as associated results (e.g. noise ceiling estimates). There are multiple versions of the betas (e.g., betas_assumehrf (b1), betas_fithrf (b2), betas_fithrf_GLMdenoise_RR (b3)). Also, betas are prepared and available in different spaces (e.g., 1.8-mm volume (func1pt8mm), 1-mm volume (func1mm), subject-native surface (nativesurface), fsaverage, MNI).
  • nsddata_stimuli (~40 GB) - This contains the color natural scene images used in the NSD experiment.
  • nsddata_timeseries (~3.4 TB) - This very large folder contains the pre-processed fMRI time-series data from which the single-trial betas are estimated. Both 1.8-mm and 1-mm versions are available. In addition, this folder contains information associated with the time-series data, including physiological data (pulse and respiratory), experimental design information (i.e. which images were shown when), motion parameter estimates from the pre-processing of the fMRI data, and eyetracking data.
  • nsddata_other (~25 GB) - This contains miscellaneous items, including (but not limited to) materials used to run the experiments and original unedited FreeSurfer outputs.
  • nsddata_diffusion (~200 GB) - This contains derivatives from analyzing the diffusion data. NOTE: We are currently preparing the final versions of the diffusion derivative files, and they will be made available within a few weeks.
  • nsddata_rawdata (~946 GB) - This contains raw data in BIDS format.

The NSD dataset is very large in size. Depending on your needs, you may not need all of the files. For example, if you wish to work from the single-trial betas that we have provided, there is no need to download the raw data nor the pre-processed time-series data. As another example, if you want only the standard-resolution (1.8-mm) preparation of the data, you can exclude the high-resolution (1-mm) preparation, which will result in major space savings (requirement of ~6 times less space). As a third example, if you want only beta version b3, there is no need to also download beta versions b1 and b2.

Held-out data

Some data collected as part of the NSD effort are not yet publicly available. These include the following:
  • nsdimagery (1 scan session) - Data related to the nsdimagery 7T fMRI experiment are not yet available. These data will be described and released as part of a separate paper effort.
  • nsdsynthetic (1 scan session) - Data related to the nsdsynthetic 7T fMRI experiment are not yet available. These data will be described and released as part of a separate paper effort.
  • Last 3 NSD core sessions - Due to the involvement of the NSD data in the  Algonauts  prediction challenge, the last 3 NSD core scan sessions from each of the 8 NSD subjects are being temporarily held out from public release. The held-out data will be released at a future date. The data are now released (Aug 20 2023).
  • nsdmemory (behavioral experiment) - Data from the final memory test conducted after completion of the NSD fMRI experiment are now available (released May 27 2023).

For the scan sessions mentioned above, the raw and pre-processed data are held out. However, the behavioral data and experimental design information (including the actual stimuli shown) for the held-out scan sessions are still available. Note that the held-out scan sessions may include instances of images whose responses are available in some other scan session either from that subject or from other subjects.