Restricted Boltzmann Machines (RBMs)
A Practical Guide to Training RBM.pdf
Background
RBMs have been used as generative models of many different types of data include labeled and unlabeled. In their conditional form they can be used to model high-dimensional temporal sequences such as video or motion capture data (Taylor et al., 2006) or speech (Mohamed and Hinton, 2010). Their most important use is as learning modules that are composed to form deep belief nets (Hinton et al., 2006a).
Coding Resources
or
RBMs in Computer Vision
In recent years Restricted Boltzmann Machines has attracted growing attention in the computer vision community. Restricted Boltzmann Machine and its variants have been used to many computer vision applications including object recognition, facial expression generation, human motion generation and activity recognition. The increased popularity of Restricted Boltzmann Machines for computer vision is due partly to their excellent ability in feature extraction.
Computer vision on Restricted Boltzmann Machines
Restricted Boltzmann Machines have been exploited in many computer vision applications. The following subsections summarize these applications including object recognition, human motion generation, facial expression generation.
3.1 Object Recognition
Object recognition is one of the fundamental challenges in computer vision. Its objective is efficiently detecting and classifying objects in an image or video sequence into generic categories such as “animals”, “vehicles”, “flowers”, etc. Object recognition in images and videos is very challenging due to high intra-class variety and viewpoint variants.
BRBM employs binary hidden and visible units, which is applicable to quasi-binary images (e.g., handwritten digits). The extension of RBMs such as the GBRBM, the mcRBM and the ssRBM are more suits to the continuous data. The characterization of object recognition using RBM has been a recent focus in the computer vision community as summarized in Table 3.1. and more please refer the following papers.
Advances in Restricted Boltzmann Machines for computer vision
