Abstract

A basic building block in many high-level Computer Vision tasks such as image classification, object detection, image retrieval, image segmentation, etc is the notion of a distance or similarity between images and/or parts thereof. This is conveniently formalized in the concept of distance functions and kernels, that can be used with many existing algorithms such as large margin classifiers or nearest neighbour algorithms. The importance of suitable distances and kernels is reflected in the large number of publications at major computer vision conferences. Researchers have mainly pursued two different routes, a) either encoding human prior knowledge in manual guided design, or b) learning based approaches that try to infer these functions automatically from training data. In particular, the latter have been applied quite successfully to the aforementioned tasks as evident from the ever increasing performance results on the standard benchmark datasets. However diverse problems persists and many approaches do employ only simple distance and kernel functions that are not tailored to the Computer Vision problem specifics.

Workshop Format

This workshop is designed to target both experts as well as novices in both the field of kernel and distance learning. To meet this purpose we include an overview presentation into the fields. This way the non-expert audience will get an overview of current state-of-the-art techniques and developments on both topics. The invited speakers will give a more detailed presentation of their own work and discuss potential next goals.

The workshop will include the following contributions:

  • Overview/Introduction to Kernel Learning
  • Overview/Introduction to Distance Learning
  • Presentation of selected publications by invited speakers
  • Contributions as posters and talks about ongoing and finished work

Call For Contribution

We invite authors to submit abstracts of relevant research for presentation at the workshop. Topics relevant to the workshop include (but are not limited to) ongoing research efforts related to distances and kernels in computer vision, such as novel algorithmic formulations, applications of distance/kernel learning in vision, or empirical evaluations of existing techniques.

Abstracts should be submitted as .pdf or .txt files, and should not exceed two pages. Each submission will be reviewed by members of the workshop committee, and successful abstracts will be selected for presentation at the workshop as either oral or poster presentations. Submission deadline is 28.August 2011 (see dates)

Submission site: https://cmt.research.microsoft.com/KDCV2011/

Please note that we do not plan workshop proceedings. The intention of the workshop is to provide an overview over recent advances which is why we also invite to present already published work. Since there are no proceedings you can also submit work in progress that has not been published so far without spoiling the chance of submitting in the future.