relocforests

Scene Coordinate Regression Forests for Camera Relocalization in RGB-D Images.

Usage

Dependencies

• Cmake
• Opencv
• Eigen

Building

Set an environment variable for Eigen: EIGEN_DIR=/path/to/eigen/include.

Build like any other cmake project.

cd relocforests
mkdir build
cd build

# UNIX Makefile
cmake ..

# Mac OSX
cmake -G "Xcode" ..

# Microsoft Windows
cmake -G "Visual Studio 14" ..
cmake -G "Visual Studio 14 Win64" ..
...

Usage of Example Application

The example application takes input frames and poses from the TUM RGBD datasets, and requires that your create an association file to associate the RGB, Depth, and pose information. Instructions for this process can be found here.

# Training
# depeding on your setup this could take upwards of 10 minutes to train
$ ./relocforests_example ~/data/rgbd_dataset_freiburg1_desk/ train desk1.rf 

# Testing
# test the forest we created with a different dataset
$ ./relocforests_example ~/data/rgbd_dataset_freiburg1_desk2/ test desk1.rf 

Citations

@Inprceedings {export:184826,
author       = {Jamie Shotton and Ben Glocker and Christopher Zach and Shahram Izadi and Antonio
                Criminisi and Andrew Fitzgibbon},
booktitle    = {Proc. Computer Vision and Pattern Recognition (CVPR)},
month        = {June},
publisher    = {IEEE},
title        = {Scene Coordinate Regression Forests for Camera Relocalization in RGB-D Images},
url          = {http://research.microsoft.com/apps/pubs/default.aspx?id=184826},
year         = {2013},
} 

@software{brooks16,
author        = {Conner Brooks},
title         = {C++ Implementation of SCORE Forests for Camera Relocalization},
howpublished  = {\url{https://github.com/isue/relocforests}},
year          = {2016}
}

This implementation uses a modified version of MeanShift++, as well as an implementation of the Kabsch algorithm.