agianbig/medvlm-calib-outputs

该数据集是medvlm-calib项目的输出结果，专门用于医学视觉语言模型的校准和符合预测任务。它包含了从§11不平衡扫描（协议A和B）以及§12视网膜单元生成的单元级预测CSV文件和视网膜逻辑CSV文件。这些文件是重度伪影，主要用于模型校准评估。数据集作为git仓库的补充，仅存储大型CSV文件，而代码和其他小型伪影则位于git仓库中。目录结构按协议、模型、发现和种子等参数组织，支持多种医学影像编码器，如BiomedCLIP、CheXzero、MedCLIP、MedSigLIP、OpenCLIP、FLAIR和RET-CLIP。数据集涵盖CheXpert中的五种胸部X光发现（如肺不张、心脏肥大等），并基于CC-BY-4.0许可证分发，但不包含原始医学图像。This dataset contains outputs from the medvlm-calib project, designed for calibration and conformal prediction in medical vision-language models. It includes per-cell prediction CSVs and retinal logits CSVs generated from the §11 imbalance sweep (Protocol A and B) and the §12 retinal cells. These are heavy artifacts used for calibration evaluation. The dataset serves as a companion to the git repository, storing only large CSV files, while code and smaller artifacts reside in the git repo. The layout is organized by protocol, model, finding, seed, and other parameters, supporting multiple medical imaging encoders such as BiomedCLIP, CheXzero, MedCLIP, MedSigLIP, OpenCLIP, FLAIR, and RET-CLIP. It covers five CheXpert chest X-ray findings (e.g., atelectasis, cardiomegaly) and is distributed under the CC-BY-4.0 license, but does not include raw medical images.