Computer Vision with Dirichlet Processes
Yesterday (3rd December, 2012) I gave a talk to both the Oxford and Oxford Brookes computer vision groups. The slides may be downloaded from here:
Computer Vision with Dirichlet Processes Slides
The abstract I gave for the talk is:
Dirichlet processes have been of keen interest to the machine learning community for a short while now, but it is only relatively recently that they have made it into computer vision. I will give a quick tutorial on Dirichlet processes, before giving three applications within the computer vision domain, specifically: 1. Background subtraction, where you discover the areas of interest in a video stream for future processing. 2. Topic models for abnormal behaviour detection, where you model behaviour using bags of words (video features) and attempt to detect unusual bags. 3. Active learning, which allows you to build classifiers with the computer selecting the exemplars to be labelled, to minimise the overall effort.
Its probably not of much general interest - the slides assume I am there to talk over them, so they probably don't help too much if you're trying to learn something.
Computer Vision with Dirichlet Processes Slides
The abstract I gave for the talk is:
Dirichlet processes have been of keen interest to the machine learning community for a short while now, but it is only relatively recently that they have made it into computer vision. I will give a quick tutorial on Dirichlet processes, before giving three applications within the computer vision domain, specifically: 1. Background subtraction, where you discover the areas of interest in a video stream for future processing. 2. Topic models for abnormal behaviour detection, where you model behaviour using bags of words (video features) and attempt to detect unusual bags. 3. Active learning, which allows you to build classifiers with the computer selecting the exemplars to be labelled, to minimise the overall effort.
Its probably not of much general interest - the slides assume I am there to talk over them, so they probably don't help too much if you're trying to learn something.