Learning Layouts for Single-Page Graphic Designs

Peter O'Donovan1     Aseem Agarwala2     Aaron Hertzmann1,2    
1University of Toronto    2Adobe Systems, Inc.

Design Analysis

alignment importance segmentation
Alignment Detection Importance Prediction Segmentation

Design Synthesis

var1 var2 var3
Synthesized Example 1 Synthesized Example 2 Synthesized Example 3
retargetOrig retargetOpt
Original Retarget
impOrig impOpt
Original Improved


Peter O'Donovan, Aseem Agarwala, Aaron Hertzmann. Learning Layouts for Single-Page Graphic Designs. IEEE Transactions on Visualization and Computer Graphics(TVCG), August 2014. Vol. 20, No. 8, pg. 1200-1213 BibTex


This paper presents an approach for automatically creating graphic design layouts using a new energy-based model derived from design principles. The model includes several new algorithms for analyzing graphic designs, including the prediction of perceived importance, alignment detection, and hierarchical segmentation. Given the model, we use optimization to synthesize new layouts for a variety of single-page graphic designs. Model parameters are learned with Nonlinear Inverse Optimization (NIO) from a small number of example layouts. To demonstrate our approach, we show results for applications including generating design layouts in various styles, retargeting designs to new sizes, and improving existing designs. We also compare our automatic results with designs created using crowdsourcing and show that our approach performs as well as, or better than, novice designers.



Graphic Design Importance Model

Design Manual Importance Image Saliency Our Predicted Importance
More examples of our MTurk labeling and importance model: Importance Model Examples

Datasets and code for training/testing the model: Coming soon.

Other Supplementary Material

Design Styles Results
Design styles for landscape and portrait aspect ratio designs:
Landscape Styles
Portrait Styles

Design Retargeting Results
The training data used for retargeting, as well as the full set of 98 automatic and MTurk retargets:
Retargeting Training Data
Designer Retargeting Results and MTurk Evaluation
Crowdsourcing Retargeting Results and MTurk Evaluation

Design Improvement Results
The following links show the training data used for the improvement application, as well as the MTurk evaluation of improving the worst/best/all sets of designs.
Improvement Training Data
Worst Designs Set
Best Designs Set
All Designs Set

Initial System Parameters
The following are the initial system parameters, as well as the regularization prior: CSV File


This research is supported in part by Adobe, NSERC, and CIFAR