BOLD5000

Brain, Object, Landscape Dataset

What is BOLD5000?

BOLD5000 is a large-scale, slow event-related fMRI dataset collected on 4 subjects, each observing 5,254 images over 15 scanning sessions. Our images are selected from three computer vision datasets.

  1. 1,000 images from Scene Images (with scene categories based on SUN categories)
  2. 2,000 images from the COCO dataset
  3. 1,916 images from the ImageNet dataset

News

New Processed and raw data in BIDS format, ROI masks, MRIQC reports on OpenNeuro!

New Paper is now available on arXiv!

New Vectors of brain responses to images are now available on Kilthub, Figshare!

New t-SNE visualizations on processed data is now available for viewing and download!

New Code for collecting our data is now available on GitHub!


Raw fMRI data for BOLD5000 is available on Kilthub, Figshare!

Collaborators

  • Nadine Chang Robotics, CMU
  • John Pyles Psychology, CMU
  • Abhinav Gupta Robotics, CMU
  • Michael Tarr Psychology, CMU
  • Elissa Aminoff Psychology, Fordham University

Diverse Images

We incorporate the two core values of neuroscience and computer vision into one dataset. From neuroscience, we have taken a careful setup of slow event-related to ensure a stimulus to reponse mapping. From computer vision, we have taken some of the most iconic and standard benchmark images. We have indoor/outdoor scenes and a lot of naturalistic images. They're complicated and contain object to object interaction. Explore our images and check out our rich data.

Largest slow event-related fMRI dataset

BOLD5000 is a novel, large-scale, slow event-related fMRI dataset. There's over 5,000 images presented to each subject. Each subject was scanned for 16 sessions. That is a total of 96 hours of scanning done for this dataset!

Easy to Download

All data are readily available for download. We have raw BOLD data, behavioral data, physiological data, images used as stimuli, and even labels. Download, explore, and share your results.