maxs-m87/inspector_md_full_v1

Inspector MD Full Dataset是一个多配置的Hugging Face导出数据集，主要用于目标检测和视觉问答任务。数据集包含三种配置：detect、point和query_issues，分别用于不同的任务需求。detect配置包含训练集11394个样本，验证集2240个样本，测试集3225个样本；point配置与detect配置相同；query_issues配置包含训练集13795个样本，验证集2382个样本，测试集3375个样本。数据集中的图像已嵌入到发布的数据集文件中，本地训练路径未保留。查询配置保留了规范的序列化issues[]负载，存储在final_answer_json和target_text中。默认情况下，该数据集是私有的，因为源许可证再分发元数据在本地合并数据集输出中未标准化。

Inspector MD Full Dataset is a multi-config Hugging Face export of training datasets, primarily used for object detection and visual question answering tasks. The dataset includes three configurations: detect, point, and query_issues, each serving different task requirements. The detect configuration contains 11,394 training samples, 2,240 validation samples, and 3,225 test samples; the point configuration is the same as detect; the query_issues configuration contains 13,795 training samples, 2,382 validation samples, and 3,375 test samples. Images are embedded into the published dataset artifacts, and local training paths are not preserved. Query configurations keep the canonical serialized issues[] payloads in both final_answer_json and target_text. By default, this dataset is private because source-license redistribution metadata is not normalized in the local merged-dataset outputs.