Qingnan Zhou, research engineer at Adobe Research and Assistant Professor Alec Jacobson were recognized with the Dataset Award at the Eurographics Symposium on Geometry Processing (SGP 2017) in London, U.K., for Thingi10k.
Thingi10k is a collection of 10,000 virtual 3D models which can be used to test new methods on a large set of shapes.
The co-collaborators created Thingi10k while developing a new method to model complex shapes for 3D printing. While they knew that theoretically their algorithm was correct, they wanted to provide evidence to the community that their software implementation was also bulletproof.
Thingi10k is the largest benchmark to date and serves as a test to compare methods. For example, consider two methods A and B that conduct the same task. If method A succeeds on 90% or 9,000 out of 10,000 models, and method B only succeeds on 50% of the models, then method A is declared to be more "robust" empirically.
Before Thingi10k, most methods only demonstrated success on a handful of rather "clean" academic models. Thingi10k’s models come from real users of 3D printers and represent data "in the wild."
To learn more, visit Thingi10k.