JAKOB ENGEL

DSO is a direct and sparse visual odometry method I developed, that combines the benefits of direct methods with those of sparse, point-based methods - greatly exceeding LSD-SLAM in runtime, accuracy and robustness. I developed DSO partly during my internship with Prof. Vladlen Koltun at Intel, and partly during my PhD at TU Munich with Prof. Daniel Cremers. You can find the open-source code on GitHub. We also created the TUM monoVO dataset, 50 real-world, "out-of-the-lab" monocular sequences for benchmarking and evaluation which can be found here.