Esser et al 2019.pol - guillaumedescoteauxisabelle/ma-biblio GitHub Wiki

---

page: 1 type: text-highlight created: 2020-12-03T20:29:55.214Z color: yellow Unsupervised Robust Disentangling of Latent Characteristics for Image Synthesis

page: 1 type: text-highlight created: 2020-12-03T20:30:03.065Z color: yellow Patrick Esser, Johannes Haux, Bj ̈ orn Ommer Heidelberg Collaboratory for Image Processing IWR, Heidelberg University, Germany [email protected]

page: 1 type: text-highlight created: 2020-12-03T20:30:06.355Z color: yellow Abstract

page: 1 type: text-highlight created: 2020-12-03T20:32:56.069Z color: #FF6900 We propose an additional classifier that estimates the minimal amount of regularization required to enforce disentanglement. Thus both representations to- gether can completely explain an image while being inde- pendent of each other. Previous methods based on adver- sarial approaches fail to enforce this independence, while methods based on variational approaches lead to uninfor- mative representations.