voc-release5
---
The main highlights in this release:

* Weak-label structural SVM (wl-ssvm)
	The Learning code has been generalized to support both lsvm and
	wl-ssvm. By default lsvm is used for training mixture models
	and wl-ssvm is used for training the person grammar model from [1].

* Person grammar model [1]
	The person grammar model from [1] is implemented in the person_grammar
	directory.

* Optimization improvements
	The stochastic subgradient descent solver (learn.cc) has been replaced
	by L-BFGS (using Mark Schmidt's minConf package). The learning and 
	feature vector cache code has been completely rewritten from scratch.
	Notably, the feature vector cache is now stored in memory and not as
	a large file on disk. The memory requirements are slightly higher than
	before, but your disk will thank you.

* Code cleanup, reorganization, and speed improvements
	The grammar API has been simplified and streamlined. The code has been
	organized into subsystems. Training models is faster. For reference,
	a PASCAL VOC 2007 bicycle model can be trained in 2 hours on a 4-year-old
	quad-core machine (using 'matlabpool open 8').

* Scale prior
	Grammar productions now have learnable "scale prior" parameters that
	help calibrate detection scores at different scales. In conjunction with 
	this, the feature pyramid code supports adding an extra octave of HOG 
	features computed on 2x2 pixel cells (off by default).
	
* Bug fixes
	Several bug from voc-release4.01 (mostly involving memory usage) have
	been fixed.

* New configuration system
	The globals.m configuration file has been replaced by voc_config.m.
	This permits overriding configurations by setting a global variable.
	Such configuration overriding makes it easy to test different config
	options on different machines from the same code directory.


voc-release4.01 (11/23/11)
---
Fixed a runtime performance bug when using fconvblas.cc as the convolution
routine with MATLAB >= 2010b on Intel CPUs.

The default convolution routine now uses SSE optimized code (fconvsse.cc).


[1] Object Detection with Grammar Models.
		R. Girshick, P. Felzenszwalb, D. McAllester.
		In Neural Information Processing Systems (NIPS) 2011.
		http://people.cs.uchicago.edu/~rbg/papers/grammar-nips11.pdf