Search papers, labs, and topics across Lattice.
This survey systematically reviews global solvers for 3D vision, categorizing them into Branch-and-Bound (BnB), Convex Relaxation (CR), and Graduated Non-Convexity (GNC) paradigms. It analyzes their theoretical foundations, algorithmic designs, and practical enhancements across ten core vision tasks, highlighting the optimality-robustness-scalability trade-offs. The survey identifies future research directions including scaling algorithms, integrating data-driven priors, establishing benchmarks, and addressing societal implications.
Certifiably optimal solutions to 3D vision problems are now within reach, but choosing the right global solver (BnB, CR, or GNC) requires navigating a complex trade-off between optimality, robustness, and scalability.
Global solvers have emerged as a powerful paradigm for 3D vision, offering certifiable solutions to nonconvex geometric optimization problems traditionally addressed by local or heuristic methods. This survey presents the first systematic review of global solvers in geometric vision, unifying the field through a comprehensive taxonomy of three core paradigms: Branch-and-Bound (BnB), Convex Relaxation (CR), and Graduated Non-Convexity (GNC). We present their theoretical foundations, algorithmic designs, and practical enhancements for robustness and scalability, examining how each addresses the fundamental nonconvexity of geometric estimation problems. Our analysis spans ten core vision tasks, from Wahba problem to bundle adjustment, revealing the optimality-robustness-scalability trade-offs that govern solver selection. We identify critical future directions: scaling algorithms while maintaining guarantees, integrating data-driven priors with certifiable optimization, establishing standardized benchmarks, and addressing societal implications for safety-critical deployment. By consolidating theoretical foundations, practical advances, and broader impacts, this survey provides a unified perspective and roadmap toward certifiable, trustworthy perception for real-world applications. A continuously-updated literature summary and companion code tutorials are available at https://github.com/ericzzj1989/Awesome-Global-Solvers-for-3D-Vision.
