dzimmerdkfz/chexpert-sam3-anatomical-masks

CheXpert解剖分割掩码（SAM3）数据集是基于CheXpert胸部X光数据集生成的解剖分割掩码数据集，使用SAM3（Segment Anything Model 3）模型和文本提示的Promptable Concept Segmentation (PCS)方法生成。数据集包含多个解剖结构的标签映射，如右肺、左肺、心脏/纵隔、气管、锁骨、肋骨、膈肌和胃泡等。每行数据包括原始索引、图像路径、图像尺寸、RLE编码的二进制掩码和每类的检测统计信息。数据集生成使用了SAM3模型（860M参数），置信度阈值为0.1，掩码阈值为0.5，源数据集为danjacobellis/chexpert（仅正面视图），共处理了1000张图像，耗时0.2小时。

The CheXpert Anatomical Segmentation Masks (SAM3) dataset is a collection of anatomical segmentation masks generated from the CheXpert chest X-ray dataset using the SAM3 (Segment Anything Model 3) model with text-prompted Promptable Concept Segmentation (PCS). The dataset includes label mappings for various anatomical structures such as the right lung, left lung, heart/mediastinum, trachea, clavicles, ribs, diaphragm, and gastric bubble. Each row contains the original index, image path, image dimensions, RLE-encoded binary masks, and per-class detection statistics. The dataset was generated using the SAM3 model (860M parameters) with a confidence threshold of 0.1 and a mask threshold of 0.5. The source dataset is danjacobellis/chexpert (frontal views only), with a total of 1000 images processed in 0.2 hours.