LidongYang/EEG_Image_decode

该数据集包含THINGS-EEG和THINGS-MEG两个子数据集，用于视觉解码和图像重建研究。THINGS-EEG数据集记录了10名受试者在被动观看THINGS数据库图像时的脑电图（EEG）信号，采样率为250Hz。训练集包含1,654个概念，每个概念10张图像，总计16,540个图像条件，每个训练图像对每个受试者重复4次；测试集包含200个概念，每个概念1张图像，总计200个图像条件，每个测试图像对每个受试者重复80次。THINGS-MEG数据集包含磁脑图（MEG）记录，使用相同的THINGS数据库图像，训练集有1,654个概念，每个概念12张图像，总计19,848个图像条件，每个训练图像对每个受试者重复1次；测试集有200个概念，每个概念1张图像，总计200个图像条件，每个测试图像对每个受试者重复12次。数据集还包括预处理后的EEG数据、EEG特征嵌入、微调模型检查点、CLIP ViT-H-14图像特征、SDXL-VAE潜在代码、刺激图像集以及生成的重建图像，支持图像检索和图像重建实验。This dataset includes two sub-datasets, THINGS-EEG and THINGS-MEG, for visual decoding and image reconstruction research. The THINGS-EEG dataset contains EEG recordings from 10 subjects while they passively viewed images from the THINGS database, sampled at 250Hz. The training set comprises 1,654 concepts with 10 images per concept, totaling 16,540 image conditions, with each training image repeated 4 times per subject; the test set includes 200 concepts with 1 image per concept, totaling 200 image conditions, with each test image repeated 80 times per subject. The THINGS-MEG dataset contains MEG recordings using the same THINGS database images, with a training set of 1,654 concepts and 12 images per concept, totaling 19,848 image conditions, each training image repeated once per subject; the test set has 200 concepts with 1 image per concept, totaling 200 image conditions, each test image repeated 12 times per subject. The dataset also includes preprocessed EEG data, EEG feature embeddings, fine-tuned model checkpoints, CLIP ViT-H-14 image features, SDXL-VAE latent codes, stimulus image sets, and generated reconstructed images, supporting both image retrieval and image reconstruction experiments.