software

Energy Minimization with Learned State Filters

This is a MATLAB implementation of Learned State Filters for fast discrete pairwise energy minimization as presented in the CVPR2013 publication. For convenient comparison to other energy minimization work, we have also included code for TRW-S, alpha-expansion, alpha-expand beta-shrink and loopy belief propagation.

Fast Video Segmentation v1.2 (October 2014)

This software release covers the segmentation method presented in "Fast object segmentation in unconstraint video", A. Papazoglou and V. Ferrari, ICCV13

Segmentation Transfer 1.2 (August 2012)

This is the software release covering the segmentation transfer in "Figure-ground segmentation by transferring window masks" D. Kuettel, V. Ferrari, CVPR12 and partly "Segmentation Propagation in ImageNet" D. Kuettel, M.Guillaumin, V.Ferrai, ECCV12

Efficient appearance distances between windows v1.0 (February 2012)

This release covers the paper "Exploiting spatial overlap to efficiently compute appearance distances for image windows" (B. Alexe, V. Petrescu and V. Ferrari, NIPS 2011)

Objectness measure v2.2 (December 2013)

This release covers the paper "What is an Object?" (B. Alexe, T. Deselaers, V. Ferrari, CVPR 2010) and the journal paper "Measuring the objectness of image windows" (B. Alexe, T. Deselaers, V. Ferrari, PAMI 2012)

ClassCut for Unsupervised Class Segmentation v1.0 (April 2011)

This release covers the paper "ClassCut for Unsupervised Class Segmentation" (B. Alexe, T. Deselaers and V. Ferrari, ECCV 2010)

Global and Efficient Self-Similarity v1.0 (August 2010)

This release covers the paper "Global and Efficient Self-Similarity for Object Classification and Detection" (T. Deselaers, V. Ferrari, CVPR 2010)

Calvin upper-body detector v1.04 (October 2011)

Software for upper-body detection in still images.

2D articulated human pose estimation v1.22 (October 2014)

Software for 2D articulated human pose estimation in still images.

Learning and detecting shape models v1.3 (September 2009)

Matlab source code for learning shape models of object categories from unsegmented training images, and for localizing outlines of new category members in cluttered test images. Although the training stage requires the objects to be annotated only by bounding-boxes, at test time the method can localize objects up to their boundaries.

This release covers the CVPR 2007 and IJCV 2010 papers.

New in V 1.3

Upper-body detector and tracker v1.0 (June 2008)

This release covers the paper "Progressive Search Space Reduction for Human Pose Estimation", (V. Ferrari, M. Marin, and A. Zisserman, CVPR 2008).

This detector is obsolete - please use the Calvin upper-body detector instead

k Adjacent Segments detector v1.04 (July 2010)

Matlab source code for the kAS local features detector. kAS are small groups of connected contour segments, forming paths through the Contour Segment Network of the image. Since each kAS has a well defined location, scale, and a descriptor invariant to translation and scale transformations, these features can be easily used for recognition and image matching.

New in v1.04
+ corrected order of components of kAS descriptor in README
+ corrected Quickstart instructions in README

New in v1.03
+ includes routines to draw kAS overlaid on edgemaps
+ all pure matlab code

New in v1.02
+ documentation for all data-structures saved out in the Matlab file format. This enables access to detailed information about edgels, edgel-chains, contour segments, and kAS. For example, it is now possible to determine exactly which edgels belong to a kAS, which is useful for visualization, or for designing alternative descriptors.

This release covers the paper "Groups of Adjacent Contour Segments for Object Detection", (V. Ferrari, L. Fevrier, F. Jurie, and C. Schmid, PAMI 2008).

Image Exploration v1.1 (July 2009)

Source code for the Image Exploration algorithm. This is a technique for automatically finding dense correspondences between multiple views of the same object or scene. The software is well suited for finding two-view correspondences as well.

This release covers the ECCV 2004 and CVPR 2004 papers. The technique was used in the CVPR 2006 and IJRR 2009 papers.