AutoDive+: An Adaptive Model Enhanced Multimodal Online Annotation Tool

The demand for data by machine learning models, especially pre-trained large-scale models, has surged dramatically in recent times, resulting in an urgent need for efficient data annotation. Despite the prevalence of annotation tools, they grapple with challenges such as high data conversion costs, limited scalability, and inefficiencies in annotating multimodal data. To tackle these challenges, this paper introduces AutoDive+, an advanced data annotation tool empowered by active learning mechanisms and integrated automatic extraction models. Expanding upon the capabilities of its predecessor, AutoDive, which excels in direct text annotation within PDF documents, AutoDive+ undergoes extensive architectural enhancements. It not only exploits high-value resource ranking modules to boost annotation efficiency across entity and content annotation but also boasts scalable interfaces for seamless integration of new annotation modes and modalities. Furthermore, our novel model community concept facilitates adaptable integration of appropriate automatic extraction and annotation models. We substantiate the effectiveness of our proposed modules through user studies spanning diverse domains. Our vision entails the establishment of an annotation task-driven open community, bolstered by the support of AutoDive+. A live demo of Autodive+ is available at http://autodive.sciwiki.cn/, and a video demo http://autodive.sciwiki.cn/introVideo/introduce-v2.0.mp4. The source code is available at https://github.com/Autodive.