Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Automatic Layout Generation of Visual Ads on Social Media Platforms
A method for automatic creation of visual ads based on raw input provided by a user within the constraints defined by a platform and while considering fundamental principles of visual attention, saliency, attention, rules of composition and keywording.
Social media platforms have become predominant channels to place targeted and personalized advertisements. Besides text, almost all platforms support the addition of visual content including logos, background images, and text, however, they impose a number of constraints to the media types and layout options including text length and size, image format and resolution, aspect ratio etc. An effective ad placement will have to present its message within the constraints defined by the platform and while considering fundamental principles of visual attention, saliency, attention, rules of composition and keywording.
The scope of the thesis is to develop methods for automatic creation of visual ads based on raw input provided by a user. Such input includes raw textual descriptions of the business or product, example images, logos, and other information. We envision a hybrid method that learns layouts and composition from a large dataset, but also considers hard constraints, geometric optimization, general rules of visual attention, and compositional principles of aesthetics. We will focus onto Google and Facebook visual ads. The thesis will include an extensive literature review on automatic layout and visual composition rules for advertisement, experiments on eye tracking etc.
Required Skills:
Knowledge in image processing and computer vision
Knowledge in graphics and geometry
Knowledge in machine learning
Programming proficiency in C++ /Python
Recommended Skills:
Experience in natural language processing
Interest in photography and cinematography
Interest in visual arts
Social media platforms have become predominant channels to place targeted and personalized advertisements. Besides text, almost all platforms support the addition of visual content including logos, background images, and text, however, they impose a number of constraints to the media types and layout options including text length and size, image format and resolution, aspect ratio etc. An effective ad placement will have to present its message within the constraints defined by the platform and while considering fundamental principles of visual attention, saliency, attention, rules of composition and keywording.
The scope of the thesis is to develop methods for automatic creation of visual ads based on raw input provided by a user. Such input includes raw textual descriptions of the business or product, example images, logos, and other information. We envision a hybrid method that learns layouts and composition from a large dataset, but also considers hard constraints, geometric optimization, general rules of visual attention, and compositional principles of aesthetics. We will focus onto Google and Facebook visual ads. The thesis will include an extensive literature review on automatic layout and visual composition rules for advertisement, experiments on eye tracking etc.
Required Skills: Knowledge in image processing and computer vision Knowledge in graphics and geometry Knowledge in machine learning Programming proficiency in C++ /Python
Recommended Skills: Experience in natural language processing Interest in photography and cinematography Interest in visual arts
A written report and an oral presentation conclude the thesis. The thesis will be overseen by Prof. Dr. Markus Gross (CGL/DRZ), Mohamed Djadoun (CGL)
A written report and an oral presentation conclude the thesis. The thesis will be overseen by Prof. Dr. Markus Gross (CGL/DRZ), Mohamed Djadoun (CGL)