Awesome

DSAC

Code associated with the paper:

E. Brachmann, A. Krull, S. Nowozin, J. Shotton, F. Michel, S. Gumhold, C. Rother,
"DSAC – Differentiable RANSAC for Camera Localization",
CVPR 2017

Please see the documentation.pdf for additional information. Also see the project page. You can download pre-trained models for 7Scenes here

Note: Beginning of August 2017, we updated the public version of the code to contain a fix in the pose evaluation metric, and to utilize a more stable variant of the PnP algorithm. These changes result in improved numbers compared to the original version of the paper. The improved numbers and a more detailed explanation can be found in the current version of the paper on arXiv. The current version of the code contains both fixes, and also the pre-trained models have been updated, accordingly.

Update: An improved version of this system is available here. Higher accuracy, stable end-to-end training, training without a 3D model or depth maps.