Visual Recognition Challenge
Note: This workshop will start after the
International Workshop on Object Categorization -- they will
(Caltech 256 and PASCAL VOC2007)
ICCV'07 Workshop, Monday 15th October 2007
There are two datasets that are becoming standard for measuring visual recognition performance in vision papers: the Caltech dataset, and the PASCAL Visual Object Classes Challenge datasets. For 2007 both have released new versions that are more challenging, for example with more classes. The objective of this workshop is to compare the best recognition methods on both datasets.
The workshop will be divided into two parts, one devoted to the PASCAL VOC2007 challenge and the other to the new Caltech 256 classes dataset. In each part there will be overview talks summarizing the competition and results, and announcing the winners. Participants who have performed well will give talks on their methods.
The PASCAL VOC2007 challenge workshop
Organizers: Mark Everingham (Leeds), Luc van Gool (Zurich), Chris Williams (Edinburgh), John Winn (Microsoft, Cambridge), Andrew Zisserman (Oxford).
A new database has been prepared consisting of 20 classes with about 25000 annotated instances in total. The images are obtained from flickr. The classes include people, cats, dogs, cars, motorbikes, bottles and sofas. The annotation includes a rectangular bounding box and flags to indicate pose and level of difficulty.
As in previous challenges there are two main competitions, one testing image classification ("does the image contain an instance of this class?"), and one testing object detection ("provide a bounding box for each instance of the class, if any"). In addition this year two 'taster' competitions have been introduced: the first evaluates the object layout in more detail ("detect the hands, feet etc for a person"), the second evaluates object segmentation at the pixel level.
Full details of the database and how to enter the competition.
The Caltech 256 workshop
Organizers: Pietro Perona, Gregory Griffin and Merrielle Spain (Caltech).