|
Workshop on Generic Object Recognition and Categorization @ CVPR 2004
|
||||||||||||||||||||||||||||||||||||||||||||||||||
|
Home
Part of CVPR 2004
Organizers
Sven Dickinson
Ales Leonardis
Bernt Schiele
Support
|
27 June 2004 Washington, DC, USA
The capacity to categorize objects plays a crucial role for a cognitive and autonomous visual system in order to compartmentalize the huge numbers of objects it has to handle into manageable categories. Quite interestingly, for humans it was shown that entry-level categorization (i.e. Is this a dog/cat?) is much faster in human vision than recognition or identification (Is this my dog/cat?). These findings suggest that humans do a sort of coarse to fine categorization and recognition of objects. Even though generic object recognition and classification have been one of the goals of computer vision since its beginnings, we are still far from achieving this goal. On the other hand, the identification of known objects in different poses and under novel viewing conditions has made significant progress recently. At the same time, impressive results have been achieved for the detection of canonical views of individual categories, such as faces, cars, pedestrians, and horses. While the more general task of multi-class object categorization is still unsolved, we have seen at recent conferences such as CVPR 2003 and ICCV 2003 that research in the area regains momentum and new approaches emerge. Generic object recognition endeavors to recognize objects based on their coarse, prototypical shape. Although a popular topic in the 1970's, generic object recognition has given up the recognition spotlight over the years to such schemes as alignment, geometric invariant-based indexing, and more recently, appearance-based and local feature-based recognition. While all of these approaches have their advantages and disadvantages it is not clear what the role of different visual cues (such as contour, shape, color, texture, etc.) is, and what the role of object models are for generic object recognition. Traditionally, contour-, shape-, and part-based methods are considered most adequate for handling the generalization requirements needed for categorization tasks, even though most current object recognition and detection systems are appearance-based. So the workshop aims to bring together the leading researchers in the field of generic object recognition and appearance-based object categorization in order to discuss and consolidate the state of the art in the field. We will also encourage participants to test and report results on recently emerging object categorization databases, such as the one put together by ETH Zurich (this databases contains 80 objects of 8 different categories, taken from 41 different viewpoints). Organization and Workshop Format In order to achieve the most stimulating discussions around the theme of generic object recognition and visual object categorization we will invite presentations by well-known researchers in the field with a record in the area of generic object recognition and visual object categorization. The workshop day will be concluded by a general discussion by all workshop participants about current and future trends in the field.
The workshop is part of
CVPR 2004 and is held prior to the main conference.
The programme of the workshop can be downloaded as pdf: cvpr-gorc04.pdf Confirmed List of Invited Speakers The programme of the workshop can be downloaded as pdf: cvpr-gorc04.pdf. The following speakers have accepted to give an invited talk the workshop (in alphabetical order): Disclaimer: Please note that the following material and in particular the slides may contain copyright material. So anyone downloading that material is responsible to respect this.
The workshop is open to all CVPR-participants.
|
