Semantic Photo Synthesis

Composite images are synthesized from existing photographs by artists who make concept art, e.g., storyboards for movies or architectural planning. Current techniques allow an artist to fabricate such an image by digitally splicing parts of stock photographs. While these images serve mainly to quickly convey how a scene should look, their production is laborious. This project has been set forward to explore the possibilities of fusing advances in computer vision, graphics, and user interfaces into innovative tools for image manipulation. It incorporates cutting edge research in:

Segmentation
Image Stitching
Object Recognition
Image Retrieval

combining them into image editing tools which understand image meaning. In addition, this project provides a forum in which to explore new ways of presenting data to the user and allowing the user to specify constraints in this new generation of intelligent tools.


Query		First Result	Second Result


Input 1	Input 2		Result

Query: A car driving on a sunny day with a blue sky

Semantic Photo Synthesis. M. Johnson, G. Brostow, J. Shotton, O. Arandjelovic, V. Kwatra, R. Cipolla. Computer Graphics Forum, 25(3):407-413, September 2006. [PDF]

Eurographics 2006. PPT

Semantic Photo Synthesis. Eurographics Video