Sketch Minimization using Crowds for Feedback

David Engel, Massachusetts Institute of Technology
Verena A. Kottler, Max Planck Institute for Developmental Biology
Christoph Malisi, Max Planck Institute for Developmental Biology
Marc Roettig, University Tuebingen
Eva-Maria Willing, Max Planck Institute for Plant Breeding Research
Sebastian J. Schultheiss, computomics.com

Design tasks are notoriously difficult, because success is defined by the perception of the target audience, whose feedback is usually not available during design stages. We present a design methodology for creating minimal sketches of objects that uses an iterative optimization scheme and brings the perception of the crowd into the loop.

We define a minimal sketch of an object as the minimal number of straight line segments required to allow 75% of naive viewers to correctly categorize the object in a free naming task (i.e. almost everyone can recognize the objects depicted in Figure 1 and it isn’t possible to draw the objects with less lines).

The standard approach to designing a minimal sketch would be to have designer to draw a sketch and then validate if it is recognizable and minimal. We improve on this scheme by replacing the designer with a crowd and including the perception of the audience into the loop. We arrive at an iterative scheme during which a naive crowd categorizes the current sketch and a set of designers then improves it based on the responses from the viewers.

We validated our method with “in-person” crowds of designers and viewers as well as on Mechanical Turk. The results show that the scheme converges quickly towards an optimum.

Over the course of the iterations the the sketches evolve towards the minimal sketch at which point they become stable (i.e. removing any further lines makes the sketch unrecognizable). We see and are investigating possible applications in the field of icon design and other design tasks.

For more, see our full paper, Sketch Minimization using Crowds for Feedback

About the author

dengel

David Engel is a Postdoctoral Associate at the Center for Collective Intelligence at MIT. His research interests include measuring collective intelligence and group coordination, human computer interfaces and combining human and machine intelligence.

View all posts