Measure Anything
Measure Anything is an open-source framework designed to provide real-time dimensional measurements of objects with circular cross-sections from stereo camera input. Built upon the powerful Segment Anything Model (SAM), the framework estimates key geometric features such as diameter, length, and volume for rod-like structures with varying curvature, as well as general objects with constant skeleton slopes. It consists of five submodules that integrate segmentation, mask processing, skeleton construction, line segment and depth identification, and 2D-3D transformations into a cohesive, user-friendly interface. Its versatility has been demonstrated through applications such as estimating the diameters of thin, non-uniform Canola stems and identifying optimal grasp points for robotic manipulation. The framework’s modular design ensures adaptability across diverse domains, from agricultural research to robotics, with the codebase and dataset publicly available to support further research and development.
Publication: Lee, Y., Panda, S. K., Wang, W., & Jawed, M. K. (2024). Measure Anything: Real-time, Multi-stage Vision-based Dimensional Measurement using Segment Anything. arXiv preprint arXiv:2412.03472.
Funding Source: We are grateful for financial support from the United States Department of Agriculture (Award numbers: 2024-67021-42528 and 2022-67022-37021).