SIG-09 will include presentations of the peer-reviewed papers. Original, high-quality submissions are solicited on the topics of interest that include, but are not limited to the following:

  • Learning vocabularies of visual primitives and parts
  • Stochastic hierarchical models of events, activities, scenes and 2D/3D objects
  • Hierarchy of classifiers for categorizing events, activities, scenes, and 2D/3D objects
  • Taxonomies of a large number of events, and objects
  • Stochastic context-sensitive image grammars
  • Statistical and graph-theoretic learning/inference algorithms for image grammars
  • Image grammars augmented with semantic relationships (e.g., hyponymy, antonymy, meronymy, etc.) available from lexical databases (e.g., WordNet)
  • The use of grammars for image parsing, and generating sentences of image interpretation at different levels of abstraction
  • Image databases and benchmarks for learning and evaluating stochastic image grammars
Important Dates:
Submission deadline 5pm PST, March 28, 2009
Notification of acceptance April 13, 2009
Camera-ready copies due April 15, 2009
Workshop date June 21, 2009

The formatting and originality requirements for SIG-09 submissions are somewhat more flexible than usual. In particular, SIG-09 submissions are allowed to have a maximum of 12 pages, and an overlap of at most 50% with any other conference/workshop publication by the same author(s). This is to ensure a large pool of submissions (in spite of the close proximity of CVPR/ICCV 2009 deadlines), and consequently a high scientific quality of accepted papers. SIG-09 will especially favor submissions with fundamental contributions to a unifying theory of image grammars over those papers proposing incremental advances in a particular vision or grammar-design problem. Since it usually takes much effort and time for meeting these high expectations, it seems unrealistic to enforce the same norms of originality on SIG-09 submissions as those set by major vision conferences. However, a SIG-09 submission on already published work must efficiently use the additional four pages allowed over the standard eight to present new theoretical aspects and corresponding new experimental evaluation of the work.

Submissions should adhere to the same formatting specifications as papers submitted to the main CVPR 2009 conference, except for the maximum paper length. Also, submissions may not be double blind.

The program chairs of SIG-09 are planning to organize a special issue of International Journal of Computer Vision on the aforementioned topics of interest. While the prospective special issue will have an independent call for papers, naturally, extended versions of the workshop's best papers are likely to have an advantage. All accepted papers will be included in the CVPR 2009 CD-ROM Proceedings.