jdh-algo/MedXray-CoT

CoT-Generation Pipeline是一个数据处理流程，它首先通过CT投影合成X射线图像及其边界框，然后使用这些边界框直接生成CT-CoT报告，并通过训练有素的检测模型生成X-ray-CoT报告。该流程支持使用MIMIC和Rate数据集，但由于数据集的访问控制，原始记录无法上传，而是提供了处理脚本和数据字典以重现数据集。图像预处理遵循Qwen的智能缩放策略，并使用qwen_vl_utils工具进行，处理后的图像存储在特定的目录结构中。

The CoT-Generation Pipeline is a data processing flow that first synthesizes X-ray images and their bounding boxes through CT projection, then directly generates CT-CoT reports using these bounding boxes, and produces X-ray-CoT reports through a trained detection model. The pipeline supports the use of the MIMIC and Rate datasets, but due to the access controls on the datasets, the original records cannot be uploaded, instead, processing scripts and a data dictionary are provided to reproduce the dataset. Image preprocessing follows the smart resize strategy in Qwen and is performed using the qwen_vl_utils tool, with the processed images stored in a specific directory structure.