OmniPath
收藏资源简介:
OmniPath是由浙江大学开发的一个专门用于病理分析的大型视觉-语言模型数据集,涵盖了20个器官的病理任务,包含约490,000个训练样本。该数据集整合了多源数据,包括PubMed数据库、病理教科书、TCGA数据集等,旨在通过人机交互提供全面的病理辅助诊断服务。数据集的创建过程结合了多种病理任务的视觉指令数据,并通过混合任务引导的特征增强和提示引导的细节特征补全策略,提升了模型在多尺度特征提取上的效率和准确性。OmniPath数据集主要应用于临床病理诊断,旨在提高诊断的准确性和效率,解决传统纯视觉模型在特征提取上的冗余问题。
OmniPath is a large-scale vision-language model dataset specialized for pathological analysis, developed by Zhejiang University. It covers pathological tasks across 20 organs and contains approximately 490,000 training samples. This dataset integrates multi-source data including the PubMed database, pathology textbooks, TCGA dataset and other resources, aiming to provide comprehensive auxiliary pathological diagnosis services via human-computer interaction. The development of OmniPath incorporates visual instruction data from diverse pathological tasks, and adopts mixed-task-guided feature enhancement and prompt-guided detailed feature completion strategies, which enhance the efficiency and accuracy of the model's multi-scale feature extraction. Primarily applied in clinical pathological diagnosis, the OmniPath dataset targets improving the accuracy and efficiency of diagnosis while addressing the feature extraction redundancy issue of traditional pure vision models.




