GPU Bundle Adjustment Against Ceres Solver
Large Photogrammetry Problems Reach Ceres's Final Cost Several Times Faster.
24.5 s vs 168 sdubrovnik-356 to Ceres cost
2.2 s vs 4.7 sfinal-93
32Ceres threads
smallproblems where Ceres is faster
- GPU Levenberg-Marquardt with implicit Schur complement PCG on BAL; Ceres 2.2 with 32 threads
- dubrovnik-356: Ceres's final cost in 24.5 s vs 168 s, and a lower final cost (491,384 vs 493,850)
- final-93: 2.2 s vs 4.7 s
- Ceres is faster on small problems (under ~2 s), trafalgar-257 and venice-52
More garmo solvers benchmark results
- Proof-Carrying Max-Cut on a GH200: 67 Biq Mac Optima Proved vs HiGHS's 49
- Weighted MaxSAT on a GH200: From 14/33/50 to 33/38/26 Against NuWLS-c-IBR
- Large-Instance TSP on 16 Arm Cores: 0.692 % → 0.165 % Mean Gap, Still Behind LKH-3
- A Native QCDCL Solver vs DepQBF: From 0 vs 4 to a 15–15 Tie, Then 15 vs 18 on Fresh Instances
- Vehicle Routing vs PyVRP: a C++ Solver Wins 37/26/21 at Equal Budget
- ISPD98 Bisection: Our Partitioner Beats KaHyPar and mt-KaHyPar at Equal Budget
- An Evidence-Driven Solver Router: Regret 0.046 vs 0.190 for the Best Single Backend
- A Verified MIP Race: HiGHS, SCIP and GPU Primal Heuristics Against cuOpt on MIPLIB 2017
- Constraint Programming in C++: 39 of 40 Schedules Proved Optimal vs CP-SAT's 20
- Certified GPU Numerics: Every Answer Carries a Checkable Certificate: and Where It Costs
- LABS on a GH200: Exponent 1.24 to the Proven Optimum, Beside the QAOA Exponents
- UNISOLVE 0.16.1: One Solver Platform, 13 Problem Domains, Every Answer Checked
- Four Certified Builds: Toroidal Proofs, Unroutability Certificates, CertBQP and a 235× Batch Service
- Unit Commitment on PGLib-UC Against HiGHS
- AC Optimal Power Flow and State Estimation
- Exact Min-Cost Flow Against OR-Tools
- GPU Semidefinite Programming on SDPLIB