RAPTOR - Realtime adAPtive detecTOR
Implementation of the following paper Daniel Göhring, Judy Hoffman, Erik Rodner, Kate Saenko, and Trevor Darrell. Interactive Adaptation of Real-Time Object Detectors. International Conference on Robotics and Automation (ICRA). 2014
Details and paper: http://raptor.berkeleyvision.org/
Implemented by Daniel Göhring and Erik Rodner. Reorganized, cleaned, corrected by Eric McCann
Installation on Ubuntu
This demo consists of the object learner and detector. The detector requires less modules so we start the installation guide with the detector. The current description focuses on Ubuntu.
install ros fuerte (see guide under http://wiki.ros.org/fuerte/Installation/Ubuntu)
sudo apt-get install git
sudo apt-get install fftw3 sudo apt-get install libfftw3-dev
sudo apt-get install ros-fuerte-openni-* ros-fuerte-bosch-drivers
git clone https://github.com/BVLC/raptor.git
cd into that folder, then...
rosws init rosws merge /opt/ros/fuerte rosws merge workspace.rosinstall source setup.bash rosmake raptor
verify that this was done correctly by running the following line:
the computer should output the version number.
Go to /data/demo-office.config.template, data/runDemo.sh, and raptor/demo-office.config, and raptor/src/transfer_images_torecord.cpp. Change “goehring” with your login name, and adjust the paths if necessary.
To use a PrimeSense kinect instead of a webcam, follow the PrimeSense installation instructions on the following page; then edit the launch files by removing usb_cam..., and replacing it with "/camera/rgb/image_color"
Test the full detection setup, after execution of these 4 commands you should see an image with detection boxes:
roscore rosrun usb_cam usb_cam_node roslaunch raptor monitor.launch roslaunch raptor detection.launch
Go to http://mitchtech.net/ubuntu-kinect-openni-primesense/ and follow the instructions for the first and third download.
For the second, DO NOT use the git clone git://github.com/avin2/SensorKinect.git download, instead, use https://github.com/PrimeSense/Sensor, which is a newer version. However, the installation instructions are the same.