Code and data (changelog)


This is the project page of our work "Learning Hybrid image Templates (HiT) by Information Projection". This paper presents a novel framework for learning a generative image representation ---- the hybrid image template (HiT) from a small number (e.g., 3 ~ 20) of image examples. Each learned template is composed of a small number image patches whose geometric attributes (location, scale, orientation) may adapt in a local neighborhood for deformation, and whose appearances are characterized respectively by four types of descriptors: local sketch (edge or bar), texture gradients with orientations, flatness regions, and colors. These heterogeneous patches are automatically ranked and selected from a large pool according to their information gains using an information projection framework.

examples structure texture flatness color

In the above two figures, the LEFT figure illustrated the image space and its composition. A hedgehog image may be seen as a collection of local image patches which are from different subspaces (primitive, texture, color, etc.) of varying dimensions and complexities. The RIGHT figure shows a few automatically learned hybrid image templates learned by composing the four types of patch prototypes. For each object/scene category, four example images are shown, followed by four bands of the hybrid templates.


Zhangzhang Si and Song-Chun Zhu
Learning Hybrid image Templates (HIT) by Information Projection
IEEE Transactions on Pattern ananlysis and machine intelligence. (under review). PDF
Zhangzhang Si, Haifeng Gong, Ying Nian Wu, Song-Chun Zhu
Learning mixed image templates for object recognition
IEEE Conference on Vision and Pattern Recognition, June 2009. PDF | Latex (zip) | poster (pptx)
Ying Nian Wu, Zhangzhang Si, Chuck Fleming, Song-Chun Zhu
Deformable template as active basis.
International Conference on Computer Vision, 2007. PDF | Code