btrf

Backtracking regression forest

This is a modified implementation of paper

@inproceedings{meng2017backtracking, title={Backtracking Regression Forests for Accurate Camera Relocalization}, author={Meng, Lili and Chen, Jianhui and Tung, Frederick and Little J., James and Valentin, Julien and Silva, Clarence}, booktitle={IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2017)},
year={2017}
}

Dependences:

1. OpenCV 3.1 or later.
2. Eigen 3.2.6 or later.
3. flann 1.8.4 or later.

The code is tested on Xcode 6.4 on a Mac 10.10.5 system. But the code has minimum dependence on compile and system, so it should work well on linux and windows as well.

File structure: src/btrf_.hpp and src/btrf_.cpp: the main algorithm for backtracking regression forst.

src/cmd: three files for training, testing of world coordinates prediction from RGB-D images, and camera pose estitation

src/dt_common: common function for decition tree, for example, objective functions

src/opencv_util: wrap of opencv function for the project

src/pose_estimation: camera pose estimation using Kabsch and preemptive RANSAC

src/Walsh_Hadamard: Walsh hardamard transformation. Code modifed from : http://www.faculty.idc.ac.il/toky/Software/wh/code.htm

src/yael_io.*: code for binary matrix read/write. Code modifed from: https://gforge.inria.fr/projects/yael

parameters/4scenes_param.txt: dataset parameter, from http://graphics.stanford.edu/projects/reloc/ parameters/forest_param.txt: forest parameter example. parameters/apt1_kitchen/ : training/testing file sequence examples

How to build with cmake:

mkdir build
cd build
cmake ../src
make -j4