This folder contains self-sufficient Matlab code to demonstrate detection of people using poselets.
The starting file is "demo.m". It reads an example image, finds the people and displays bounding boxes around them.

The algorithm is described in the following paper:
Lubomir Bourdev and Jitendra Malik, Poselets: Body-Part Detectors Trained using 3D Pose Annotations, ICCV 2009

This code is distributed with a non-commercial license. Please see the associated license file.
For questions or comments please email Lubomir Bourdev at lbourdev@eecs.berkeley.edu

___________________
Instructions for interactive visualization

Detection results can be previewed with a call to "visualize_detected_objects"

'esc': Exit visualization
left/right arrow:  Move to prev/next image, if more than one image
'p','b','t':  Toggle between 'poselets' mode (default), visible bounds mode, and torsos mode.

Poselets mode:
  Figure 1: Displays a poselet activation window (green dashed bounds) and visible object bounds suggested by the poselet
  Figure 3: Displays three examples from the current poselet 
  '-' and '=' move to the previous/next poselet. Poselets are ordered in highest-to-lowest probability within the image

Visible bounds / torsos mode:
  Figure 1: Displays the predicted visible bounds (or torsos) of the top N objects (default: N=1)
  up/down arrow: Increase/decrease N
  'space' while the mouse is over a single bounding box: Goes to Single Object View (described below)

Single Object View:
  Figure 11: Displays a zoomed version of the currently viewed object and its predicted visible bounds
  Figure 10: Displays figure/ground thumbs of the poselets activated on that object, sorted by probability
  Figure 3: (If a single poselet is selected): Displays three examples from the current poselet
  Mouse click on a figure/ground thumb: Activates or deactivates the thumbnail. Activated thumbnails are painted in color and their figure/ground predictions are shown on Figure 11
  'esc': Exit Single Object View mode
