Laplace Reading Group

of the ENS

Welcome to the “Laplace” reading group, a series of seminars and informal discussions organized by the CFM-ENS Chair “Modèles et Sciences des Données.

In these meetings, researchers can give presentations about their current research interests or discuss interesting papers. As with the Data Science Colloquium, the goal is to initiate discussions between researchers from different fields that all have a common interest in large scale or high-dimensional data. You can check the list of the next seminars below and the list of past reading groups.

All are welcome to attend, feel free to propose topics that you would like to see discussed, or work that you would like to present (contact at the bottom).

Next seminars

Nov. 29th, 2018, 10h30-11h30, room W, DMA, ENS (45 rue d'Ulm, 4th floor).
Gui-Song Xia (Wuhan University)
Title: Texture Synthesis with Conditional Generative CNNs
Abstract: Example-based texture synthesis (EBTS) is a well-studied yet challenging problem in computer vision and graphics, it aims to produce textures that are visually similar to given exemplars. Although some breakthroughs have been achieved by recently proposed methods using deep neural networks, they still have some difficulties in synthesizing textures with complex structures, e.g. non-local and near-/regular geometric patterns. Moreover, most of them rely on large-scale deep networks pre-trained for recognition tasks. Aiming at better generation of highly structured textures with lighter deep neural networks, we present a conditional generative convolutional neural net-work (cgCNN) model for texture synthesis in this paper. Given an exemplar, our model defines a conditional distribution of synthesized textures using a light-weight CNN. Our model is then trained by iteratively learning and sampling from the conditional distribution. Instead of using a pre-trained CNN, our model learns the weights of CNN in the training process. Furthermore, our model can be easily extended to synthesize dynamic textures by directly adding a temporal dimension to the convolution kernels. Experiments demonstrate that our model achieve state-of-the-art performances, especially on highly structured textures.


If you want to subscribe to (or unsubscribe from) the mailing list please send a mail to Nicolas Keriven (ENS).