DAFT

Implemenation of DAFT (Depth-Adaptive Feature Transform) algorithm, plus tools. For more information, see https://ias.in.tum.de/people/gossow/rgbd.

Directory structure:

• eval: Matlab evaluation framework (modified version of [1])
• libdaft: DAFT implementation
• opencv_ext: OpenCV addons (contains Keypoint3D class)
• test_images: printable images for debugging purposes
• tools: command-line tools for feature extraction

Checkout & compile:

The following instructions have been tested with Ubuntu 11.10 and OpenCV 2.3.

You can install OpenCV like this:

sudo apt-get install libopencv2.3

For an out-of-source build, do:

git clone https://github.com/dgossow/daft.git
mkdir daft_build
cd daft_build
cmake ../daft
make

This will create the static library libdaft/libdaft.a. You will need to link against it and also have your include paths set up.

Extract DAFT features from an image like so:

#include <daft/daft.h>

// ...

cv::Mat gray_img;
cv::Mat mask_img;
cv::Mat depth_img;
cv::Matx33f K;

// load data ..

std::vector<cv::KeyPoint3D> keypoints;
cv::Mat descriptors;

cv::daft::DAFT daft;
daft( gray_img, mask_img, depth_img, K, keypoints, descriptors );

[1] http://www.robots.ox.ac.uk/~vgg/research/affine/evaluation.html