Generic Object Decoding

Here we provide preprocessed fMRI data and image features from Horikawa &amp; Kamitani (2017) Generic decoding of seen and imagined objects using hierarchical visual features. Nat Commun<br>Raw (unpreprocessed) fMRI data are available at OpenNeuro<br>Analysis demo code is available at GitHub<br><br><b>fMRI data — SPM5-preprocessed (original data from the paper)</b><br>`Subject[1-5].h5`<br>All experiments (image presentation training/test and imagery) combined per subject.<br>Preprocessed with SPM5 into individual T1w native space: head motion correction, coregistration to the within-session T2-weighted anatomical image and then to the whole-head T1-weighted anatomical image, reinterpolation to 3×3×3 mm³. No spatial normalization was applied.<br>Post-preprocessing: 1-volume (3 s) hemodynamic shift, outlier reduction (3 SD threshold, 10 iterations), within-run linear detrending, voxel-amplitude scaling relative to the run mean, and trial-wise averaging (3 volumes for image presentation; 5 volumes for imagery) (corrected based on the actual preprocessing procedure).`Subject[1-5]_SpaceTemplate.nii`<br>space-defining template image for the above data.`Subject[1-5]_T1wAligned.nii`<br>T1-weighted anatomical image.<b>fMRI data — fMRIPrep-preprocessed</b>`Subject[1-5]_ImageNetTraining.h5`, `Subject[1-5]_ImageNetTest.h5`, `Subject[1-5]_Imagery.h5`<br>Separated by experiment per subject: training image session (`_ImageNetTraining`), test image session (`_ImageNetTest`), and imagery experiment (`_Imagery`).<br>Preprocessed with fMRIPrep 1.2.1 into individual T1w native space: slice timing correction, head motion correction, coregistration to the T1-weighted anatomical image. No spatial normalization was applied.<br>Post-preprocessing: motion parameter regression (with DC removal and within-run linear detrending), outlier reduction (3 SD threshold, 10 iterations), 1-volume (3 s) hemodynamic shift, and trial-wise averaging (3 volumes for image presentation; 5 volumes for imagery).`Subject[1-5]_fMRIPrepSpaceTemplate.nii`<br>space-defining template image for the above data.<b>Image features</b>`ImageFeatures.h5`, `ImageFeatures.mat`<br>Visual features from stimulus images computed with CNN (AlexNet, CNN1–8), HMAX (1–3), GIST, and SIFT+BoF. Identical content in HDF5 and MATLAB formats.<br>History:2026-04-23 Preprocessing descriptions added.2026-04-21 Space template files corresponding to the fmriprep-preprocessed data were uploaded (Subject*_fMRIPrepSpaceTemplate.nii).2020-10-16 fMRI data preprocessed with fmriprep were updated (Subject*_ImageNetTraining.h5, Subject*_ImageNetTest.h5, and Subject*_Imagery.h5).2020-07-27 'category_index' and 'image_index' in fMRI data files (Subject*.h5) were fixed.2019-05-08 fMRI data mat files were replaced with h5 files.<br>