Determinantal Patch Processes for Texture Synthesis
Claire Launay, Arthur Leclaire
Because of their repulsive property, determinantal point processes (DPPs) provide an efficient tool to subsample discrete distributions.
In this paper, we investigate the use of DPPs to subsample the distribution of patches of a texture image. We show that in a texture model
based on optimal transportation, DPP subsampling helps to reach a good compromise between visual quality of the synthesis and execution time.
Paper
A communication paper for the 2019 GRETSI Conference can be found
here. The presentation is also available at
this adress.
Source codes
Source codes and test material can be found on
Arthur Leclaire's webpage.
If you have any question or if you find some bugs in these codes, don't hesitate to send us an email, we will be happy to answer them.