Generic Object Decoding

This repository contains the data and demo codes for replicating results in our paper: Horikawa and Kamitani (2017) Generic decoding of seen and imagined objects using hierarchical visual features. Nature Communications 8:15037. The generic object decoding approach enabled decoding of arbitrary object categories including those not used in model training.

Dataset

• Raw fMRI data: OpenNeuro
• Preprocessed fMRI data and image features: figshare
• Stimulus images: upon request via https://forms.gle/ujvA34948Xg49jdn9

Code

Demo programs for Matlab and Python are available in code/matlab and code/python, respectively. See README.md in each directory for the details.

Note

Visual images

For copyright reasons, we do not make the visual images used in our experiments publicly available. You can request us to share the stimulus images at https://forms.gle/ujvA34948Xg49jdn9.

Stimulus images used for higher visual area locazlier experiments in this study are available via https://forms.gle/c6HGatLrt7JtTGQk7.

Some of the test images were taken from ILSVRC 2012 training images. See data/stimulus_info_ImageNetTest.csv for the list of images included in ILSVRC 2012 training images.