Probabilistic Image Generation from Layout

Despite significant recent progress on generative models, controlled generation of images depicting multiple and complex object
layouts is still a difficult problem. Among the core challenges are the diversity of appearance a given object may possess and, as
a result, exponential set of images consistent with a specified layout. In this project, we propose to address these challenges by a
probabilistic generative model, which can generate a set of realistic images by giving the coarse spatial layout (bounding boxes +
object categories). To further increase the controllability of the image generation, attributes or textual description of each object in
the layout may also be provided. A probabilistic generative model will be designed to better model the uncertainty the target
image. It will compose distributions based on given information and decodes samples from these distributions into realistic
images. The expected outcomes include the source code of the proposed model, one research paper to be published at a top-tier
international conference or journal, and one patent of the proposed algorithm.

Faculty Supervisor:

Leonid Sigal

Student:

Partner:

The University of Tokyo

Discipline:

Computer science

Sector:

Education

University:

The University of British Columbia

Program: