SimulaMet/MedMultiPoints

MedMultiPoints是一个为医学领域多任务学习设计的多模态医学成像数据集，包含用于对象检测、定位和计数任务的数据。该数据集整合了来自内窥镜和显微镜模态的数据，反映了临床多样性。数据集包含10600张来自不同模态的图像，以及丰富的多类型注释，包括边界框、点注释和计数标签。数据集兼容视觉语言模型和指令调整管道，并以JSON格式设计注释，以便与多模态训练无缝集成。

MedMultiPoints is a multimodal medical imaging dataset curated for multi-task learning in the medical domain, encompassing object detection, localization, and counting tasks. The dataset integrates data from endoscopic and microscopic modalities, reflecting real-world clinical diversity. It contains 10,600 images from diverse modalities with rich multi-type annotations including bounding boxes, point annotations, and count labels. The dataset is compatible with Vision-Language Models (VLMs) and instruction-tuned pipelines, with JSON-formatted annotations designed for seamless integration with multimodal training.