Koa-Chang/TissueMNIST-224-1pct-gpt55-with-vlm-features
收藏资源简介:
该数据集名为TissueMNIST 224 1pct with GPT-5.5 VLM Features,是基于官方MedMNIST TissueMNIST 224px数据衍生的。它包含1%的TissueMNIST训练子集,其中带有由GPT-5.5生成的分类形态学特征(VLM特征)。验证集和测试集是完整的TissueMNIST传递分割,为了保持模式一致性,其vlm_model_name字段为空且vlm_feature字段为占位符值。VLM特征标签是用于CS231N课程中VLM-LUPI(视觉语言模型-学习使用特权信息)实验的分类特权信息注释,并非临床诊断。数据集包含以下列:image(原始的224x224 uint8灰度TissueMNIST图像)、label(TissueMNIST类别标签)、dataset_index(在源MedMNIST分割中的官方索引)、vlm_model_name(生成分类VLM特征的模型名称,传递分割中为空)和vlm_feature(整数编码的分类形态学特征,传递分割中包含零占位符)。VLM特征涵盖多个形态学维度,如背景杂乱、主导区域尺度、主导形状、边缘接触、伸长率、前景分数、内部信号纹理、局部排列、原始对比度、信号分布和可见对象计数等。
A 1% TissueMNIST train subset with categorical morphology features generated by GPT-5.5. Validation and test are full TissueMNIST passthrough splits with null `vlm_model_name` and placeholder `vlm_feature` values for schema consistency. This dataset is derived from the official MedMNIST TissueMNIST 224px data. The VLM feature labels are categorical privileged-information annotations for CS231N VLM-LUPI experiments, not clinical diagnoses.




