Assignement Homework 3

Setup the Lua environment with the kinect and the face detector

  1. Installing Torch5 from the source with Qt: instructions from sources
  2. (The dace detector does not work in windows, so you must have a linux or a mac)
  3. Installing Xlearn download sources and run sudo make install at the root of the uncrompressed folder
  4. (libfreenect must be installed beforehand in you machine with sudo make install)
    (opencv must be installed beforehand in you machine)
  5. Run the demo face-detect-multiscale.lua in the demos folder
  6. Run this standalone script on any image with a face for detection

Face detection on the kinect with segmentation

The goal of this homework is to combine all the blocks we have done so far.
Ultimately we want to construct a segmented 3D face detector.
The goal is simple we have to use the face detector to apply the mincut segmenation on rectified RGB+depth image.


  1. Show that you got the face detector working by providing an image with a face and its detection (you can use any image)
  2. Now, instead of using an image as input to the detector you must use the kinect and apply the detection on the RGB image/video coming from the kinect (demo will be shown during class)
  3. Use the qualibration from homework2 to align the depth with the rgb of the kinect
  4. Use the bounding box of the detection to apply the segmentation on the aligned rgb+depth images
Demo should be provided during class