PolypSense3D is a multi-source benchmark dataset designed specifically for depth-sensing polyp size measurement tasks, released in 2025 by Hangzhou Normal University in collaboration with the Technical University of Denmark, Hohai University, and other institutions. The related research paper is titled "...".PolypSense3D: A Multi-Source Benchmark Dataset for Depth-Aware Polyp Size Measurement in EndoscopyThe study, which has been selected for NeurIPS 2025, aims to provide high-quality training and evaluation resources for polyp detection, depth estimation, size measurement, and simulation-to-real transfer learning.

Data Scale and Composition

This dataset integrates three types of data: virtual simulation, physical phantoms, and real clinical scenarios.

• Virtual simulation data: Contains 32,000+ frames, providing synchronized RGB, dense depth, segmentation mask, and camera parameters, and covers 30 procedural polyp models (1.79–20.52 mm) for training and evaluation under high-precision, controlled conditions.

• Physical data: Contains 13 videos derived from a colonic phantom constructed and 3D printed based on real CT scans, with 13 solid polyps (4.00–14.89 mm) embedded, and with rigorously calibrated camera parameters for sim-to-real migration validation.
• Clinical data: derived from standard clinical colonoscopy examinations, including stable frozen frames, providing manual and model-assisted segmentation, dimension annotation based on calibrated biopsy forceps, and sparse depth annotation for model evaluation under real-world conditions.