Important dates:
| Deadline for paper submission |
March 29, 2009 (5PM PST) |
| Notification of acceptance |
April 9, 2009 |
| Camera-ready copies due |
April 14, 2009 |
| Workshop date |
June 25, 2009 (Jointly held with VCL’09 Workshop) |
Overview:
One of the holy grails of computer vision research is achieving the total understanding of a visual scene, in a similar way that humans can do effortlessly. Great progress has been made in tackling one of the most critical components of visual scene understanding: object recognition and categorization. Recently, a few pioneering works have begun to look into the interactions among objects and their environments towards the goal of a more holistic representation and explanation of the visual scene. This workshop offers an opportunity to bring together experts working on different aspects of scene understanding and image interpretation and provides a common playground for a stimulating debate.
Scope:
The Representation and Recognition Aspect
- What is total scene understanding? What defines objects, things, stuff, context, scenes and other high-level visual concepts (e.g., activities and events)?
- How is this problem related to the ‘image annotation’ problem from the CBIR/TRECVID community?
- Can we break down the task into individual component recognition? If yes, how? If no, why?
The learning aspect
- What are the challenges faced in learning for these problems? Are they the same or different from isolated object recognition?
- What can we do to exploit the huge amount of data on the web?
New methods for evaluation and benchmarking
- Are the object recognition datasets still suitable for this problem? If not, what are the new datasets? What are the appropriate benchmark tasks and evaluation metrics?
Paper topics may include total scene understanding, visual recognition and annotation in complex real-world images, high-level understanding (events, activities) in single images, and all their related learning, representational, and dataset issues.
Submission and Reviews:
a) General instruction
Submitted papers must have a maximum length of 8 pages and must adhere to the same CVPR 2009 layout specifications as papers submitted to the main conference. All reviewing will be carried out double-blind by the Program Committee. Please refer to CVPR 2009 website (http://www.cvpr2009.org/) for instructions on the submission format. In submitting a manuscript, authors acknowledge that no paper substantially similar in content has been or will be submitted to another conference or workshop during the review period.
b) Submitting to VCL vs. ViSU
In the submission form, the authors will be asked to indicate which workshop (VCL or ViSU) the authors prefer to submit to. If no preference is indicated, the VCL-ViSU program chairs will make the decision. Double submission to ViSU’09 workshop and the main CVPR’09 conference are allowed.
|
