Blocks is an open source modular MATLAB framework which allows the user to avoid needlessly repeating computation.
Blocks may be easily used for your own experiments, and comes with some useful modules for computer vision experiments: an implementation of a simple image level bag-of-features (BoF) classifier, a fast agglomerative information bottleneck (AIB) sliding window localization scheme, and a class segmentation scheme using superpixel neighborhoods and a conditional random field (CRF). It also provides support for running experiments on clusters of computers.