maxbennett/cued-recall-imagenet

该数据集是用于cued-recall MemoryVLM实验的ImageNet训练/验证/测试分割版本。它包含235个类别的JPG图像捆绑文件（每个类别一个tar文件），解压后每个类别目录包含img_*.jpg图像。数据集还提供了权威的训练/验证/测试分配文件（splits.json），其中路径相对于data/目录。此外，包含Brady2008/2013刺激集和MST（用于记忆相似性和2-AFC任务）的memory_datasets.tar文件，以及用于配对关联试验的1000个名词词汇文件（word_vocab.txt）。用户可以通过提供的bash命令重建本地文件布局。This dataset is a train/validation/test split variant of ImageNet designed for the cued-recall MemoryVLM experiment. It comprises tar archives of JPG images across 235 categories, with one tar file per category. Upon extraction, each category directory holds images following the naming pattern img_*.jpg. The dataset also provides an authoritative train/validation/test split assignment file (splits.json), with paths relative to the data/ directory. Additionally, it includes the memory_datasets.tar archive containing the Brady2008/2013 stimulus set and MST (used for memory similarity and 2-AFC tasks), as well as the 1000-noun vocabulary file (word_vocab.txt) for paired-association trials. Users can reconstruct the local file layout using the provided bash commands.