Deep image reconstruction from human brain activity

Shen, Horikawa, Majima, and Kamitani (2019) Deep image reconstruction from human brain activity. PLos Comput Biol. (bioRxiv preprint) Here we provide DNN features decoded from human brain activity (decodedfeatures-*.mat files) as well as preprocessed fMRI data used for the decoding (sub-*.h5 files). Raw fMRI data are available at OpenNeuro. Demo code is available at GitHub. History 2022-10-14: Add fMRI data (sub-03) preprocessed by fmriprep (*_fmriprep_volume_native.h5) 2022-10-13: Add fMRI data (sub-01, sub-02) preprocessed by fmriprep (*_fmriprep_volume_native.h5) 2022-01-27: Minor update on fMRI data (v2; "stimulus_name" was added). 2020-09-09: True DNN features (unit activations) of VGG19 for natural image training and test images were added. 2020-09-08: True DNN features (unit activations) of VGG19 (except conv1) for natural image training images were added. 2020-08-05: fMRI data version 2, which applied slightly different preprocessng and ROIs from the original study, were added (for the natural image experiment only; *_perceptionNaturalImageTraining/Test_VC_v2.h5). 2020-05-18: The following fMRI data files of artificial images, letters, and imagery experiments contained incorrect samples and were fixed. sub-01_perceptionLetterImage_original_VC.h5 sub-02_perceptionArtificialImage_original_VC.h5 sub-02_perceptionLetterImage_original_VC.h5 sub-02_imagery_original_VC.h5 sub-03_perceptionArtificialImage_original_VC.h5 sub-03_perceptionLetterImage_original_VC.h5 sub-03_imagery_original_VC.h5Shen、Horikawa、Majima与Kamitani（2019）于《公共科学图书馆·计算生物学（PLoS Comput Biol）》发表论文《基于人类脑活动的深度图像重建》，该研究曾发布于bioRxiv预印本平台。 本数据集提供了从人类脑活动中解码得到的深度神经网络（DNN）特征（文件命名格式为decodedfeatures-*.mat），以及用于解码的预处理功能磁共振成像（fMRI）数据（文件命名格式为sub-*.h5）。 原始功能磁共振成像数据可于OpenNeuro平台获取。 演示代码可于GitHub平台获取。 更新日志 2022-10-14：新增经fmriprep预处理的功能磁共振成像数据（sub-03，文件命名格式为*_fmriprep_volume_native.h5） 2022-10-13：新增经fmriprep预处理的功能磁共振成像数据（sub-01、sub-02，文件命名格式为*_fmriprep_volume_native.h5） 2022-01-27：对功能磁共振成像数据进行小幅更新（版本v2；新增"stimulus_name"字段） 2020-09-09：新增用于自然图像训练集与测试集的VGG19模型真实DNN特征（单元激活值） 2020-09-08：新增用于自然图像训练集的VGG19模型（不含conv1层）真实DNN特征（单元激活值） 2020-08-05：新增fMRI数据版本v2，该版本采用与原始研究略有差异的预处理流程与感兴趣区（ROI），仅针对自然图像实验（文件路径为*_perceptionNaturalImageTraining/Test_VC_v2.h5） 2020-05-18：修复了以下人工图像、字母与意象实验对应的fMRI数据文件中的错误样本： sub-01_perceptionLetterImage_original_VC.h5 sub-02_perceptionArtificialImage_original_VC.h5 sub-02_perceptionLetterImage_original_VC.h5 sub-02_imagery_original_VC.h5 sub-03_perceptionArtificialImage_original_VC.h5 sub-03_perceptionLetterImage_original_VC.h5 sub-03_imagery_original_VC.h5