Generic Object Decoding: Preprocessed fMRI and Visual Feature Data

Kamitani Lab provides preprocessed fMRI data and image features from the Horikawa & Kamitani (2017) study on generic decoding of seen and imagined objects. The dataset includes data from five subjects, processed with both SPM5 and fMRIPrep pipelines, and is stored in the BData format for use with the Brain Decoder Toolbox for Python. It is a key resource for research at the intersection of neuroscience, computer vision, and machine learning.

Use Cases

• Training brain decoding models based on hierarchical visual features from deep neural networks.
• Testing generalization of visual object classifiers across seen and imagined stimuli.
• Benchmarking fMRI preprocessing pipelines (SPM5 vs. fMRIPrep) for decoding performance.
• Analyzing neural representations of objects in individual subject native anatomical space.

Strengths

• Includes data from five human subjects, providing a basis for individual-differences analysis.
• Offers two distinct preprocessing pipelines (SPM5 and fMRIPrep), allowing for methodological comparison.
• Dataset is large, with a total size of approximately 23.9 GB, indicating substantial data volume.
• Well-documented provenance, directly linked to a published Nature Communications paper and associated analysis code.

Limitations

• The dataset has a relatively small sample size of five subjects, which may limit statistical power for some analyses.
• Metadata completeness is medium; specific column names and row counts are not provided in the available descriptions.
• The last update date (2026-06-03) appears to be a future date, which is a conflict in the factual metadata.

Provenance

Source

Kamitani Lab, based on experiments from Horikawa & Kamitani (2017).

Collection Method

Functional MRI recorded during image presentation and mental imagery tasks, subsequently preprocessed.

Freshness

2026-06-03 (Note: This is a future date, indicating a potential metadata error.)