First Workshop on Deep GEneraTive models for 3D understanding

June 15th | Long Beach, USA


Understanding and modelling the 3D structure and properties of the real world is a fundamental problem in computer vision with broad applications in engineering, simulation, and sensing modalities in general. Humans have an innate ability to rapidly build models of the world that they are observing. With only few exposures to a new concept, humans could relate and generalize this knowledge to other known concepts. And conversely, when visually taking in the environment, humans quickly identify concepts, geometry and the physical properties that are associated with them, even if parts of objects are obscured, for example objects being occluded in images.

This ability is crucial for building smarter AI systems. In these, the models have prior knowledge of objects that can occur in the world, learn to identify these objects and then make assumptions of how these objects would behave.

We would like to explore methods which can learn 3D representations properly to be effectively used for learning in end-to-end differentiable settings such as using neural networks. This becomes especially important when we want to use 3D representations with other modalities like text, sound, images etc to capture certain important spatial properties that might be ignored otherwise. The goal of this workshop is to study how neural networks can learn to build such intricate 3D models of the world.

Additionally, we are interested in research that explores and manipulates the 3D representation in the neural net. Therefore we encourage submissions in which the latent variables that are representing the 3D data are varied or interpolated. We would like to see new perspectives on objects, new shapes, or entirely new scenes that weren't part of the training data.

News and Updates


Paper Submission

We solicit paper submissions on novel methods and application scenarios of CV for 3D. We accept papers on a variety of topics, including 3D reconstruction, 3D generation, 3D unsupervised learning and 3D-aware modelling and manipulations. Papers will be peer reviewed under double-blind policy and the submission deadline is 20th March 2019. Accepted papers will be presented at the poster session, some as orals and one paper will be awarded as the best paper.

Call for Sculptures: Mini Sculpture Garden

We are inviting researchers to submit interesting 3D models they generated with their deep vision models. We will select the most interesting objects, work with the creators to 3D-print them and display them in a "mini sculpture garden" at the workshop. We will award a best in show prize to the most outstanding contribution. Sculpture documentation and model submission deadline is 1st May 2019.

Challenge: TBD

We will host a challenge on TBD

Invited Speakers

Shuran Song


Alec Jacobson

Univ. of Toronto

Invited Speaker 5