EndoSMAP: A Semi-automatic Multimodal Annotation Software for Endoscopic Point Tracking

The surge in minimally invasive surgery has led to an exponential growth of clinical endoscopic video data, offering rich information for postoperative analysis and the development of intelligent surgical systems. However, existing general-purpose annotation software is inefficient and inflexible for frame-by-frame point tracking, making manual labeling both time-consuming and prone to error. To address these challenges, we present EndoSMAP, a semi-automatic multimodal annotation software for endoscopic point tracking. Unlike conventional software, which is limited to single-modality inputs and basic interfaces, EndoSMAP integrates manual labeling, text annotation, software-guided correction, and model-driven prediction into a unified and intuitive workflow. Built-in support for tracking model enables real-time, full-frame prediction and flexible interaction. Through extensive evaluation of EndoSMAP on real surgical data, we demonstrate that EndoSMAP significantly improves annotation efficiency and accuracy, offering a scalable solution for high-quality surgical data curation and algorithm training. The source code of EndoSMAP is publicly available at https://github.com/polysulfone/EndoSMAP .