Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Evaluation of optimization algorithms for Large scale Bundle adjustment
See description
Bundle adjustment can be seen as a special instance of Structured Matrix Factorization, which is one of the classical problems for non-convex optimization. Most methods, however, do not scale well with the size of the problem, such that most promising approaches are variants of first-order (gradient based) type.
The state-of-the-art method is a first-order method [1], but surprisingly does not apply one of the well-known non-convex algorithms that exist in the literature [2,3,4], for which convergence is proven and that are known to be very suitable to the problem class.
Hence, the goal of this thesis is to implement different state-of-the-art methods and evaluate them on large bundle adjustment problems.
[1] Zhang, “Distributed Very Large Scale Bundle Adjustment by Global Camera Consensus”, ICCV 2017
[2] Bolte et al. “Proximal Alternating Linearized Minimization for Nonconvex and Nonsmooth Problems”, arxiv 2014
[3] Malitsky et al. “Model Function Based Conditional Gradient Method with Armijo-like Line Search”, arxiv 2019
[4] Komodakis, “MRF energy minimization and beyond via dual decomposition”, PAMI 2011
Bundle adjustment can be seen as a special instance of Structured Matrix Factorization, which is one of the classical problems for non-convex optimization. Most methods, however, do not scale well with the size of the problem, such that most promising approaches are variants of first-order (gradient based) type. The state-of-the-art method is a first-order method [1], but surprisingly does not apply one of the well-known non-convex algorithms that exist in the literature [2,3,4], for which convergence is proven and that are known to be very suitable to the problem class. Hence, the goal of this thesis is to implement different state-of-the-art methods and evaluate them on large bundle adjustment problems.
[1] Zhang, “Distributed Very Large Scale Bundle Adjustment by Global Camera Consensus”, ICCV 2017 [2] Bolte et al. “Proximal Alternating Linearized Minimization for Nonconvex and Nonsmooth Problems”, arxiv 2014 [3] Malitsky et al. “Model Function Based Conditional Gradient Method with Armijo-like Line Search”, arxiv 2019 [4] Komodakis, “MRF energy minimization and beyond via dual decomposition”, PAMI 2011
Implement and evaluate different algorithms for large scale bundle adjustment problems
Implement and evaluate different algorithms for large scale bundle adjustment problems
Christoph Vogel (Christoph.vogel@microsoft.com)
Ondrej Miksik (ondrej.miksik@microsoft.com)
Christoph Vogel (Christoph.vogel@microsoft.com) Ondrej Miksik (ondrej.miksik@microsoft.com)