The updated schedule and slides will be posted here:


Time Speaker Title
4:15-4:25pm Craig Yu, Sai-Kit Yeung Introduction
4:25-4:55pm Daniel Aliaga Modeling 3D Worlds: Outdoor
4:55-5:25pm Sai-Kit Yeung Modeling 3D Worlds: Indoor
5:25-5:55pm Hao (Richard) Zhang Remodeling 3D Indoor Scenes
5:55-6:00pm Craig Yu, Sai-Kit Yeung Summary


We will first introduce the speakers of this course, who are active computer graphics researchers with strong and diverse expertise in 3D modeling. We will then motivate the topic of this course: modeling and remodelling 3D worlds, by discussing the different practical uses of virtual worlds in computer graphics and AR/VR applications. Next we will give the audience some background about the approaches commonly used by digital artists to create virtual worlds nowadays, with a focus on the application scenarios and limitations of these approaches, which will serve as motivations for different techniques we will discuss.

Modeling 3D Worlds: Outdoor

We will focus on discussing the techniques for modeling outdoor scenes, such as terrains, cities, street networks and heritages. Such scenes are commonly used in movie production for creating stunning scenery pictures and for computer games. We will focus on the discussion of procedural modeling techniques, such as shape grammars, to generate city layouts, street networks, terrains, buildings and vegetation. In addition to discussing the basic concepts of procedural modeling, we will also discuss how procedural modeling techniques can be used to devise interactive 3D modeling tools, and the recent progress and challenges in inverse procedural modeling.

Modeling 3D Worlds: Indoor

We will focus on the techniques for modeling indoor scenes, which are commonly used for games and AR/VR applications (e.g., room-scale VR). We will discuss how optimization-based procedural can be used for generating layout designs, which can be realized into 3D virtual environments. In particular, the automatic synthesis of human-centric layout designs will be discussed. We will also discuss computer vision techniques for reconstructing indoor scenes from the real world. In particular, we will put emphasis on scene capturing techniques based on consumer-grade depth cameras, such as the Microsoft Kinect, where general users can easily access and apply for reconstructing their own scenes. Finally, we will discuss the recent interesting approaches on semantic 3D scene reconstruction (i.e., using 3D CAD models to construct a 3D scene that matches with a real-world scene semantically) and discuss their application potential such as for interior design suggestions in augmented reality.

Remodeling 3D Indoor Scenes

In this part of the talk, we will focus on discussing the approaches for analyzing the semantics and functionality of 3D scenes that are modeled by human designers or reconstructed from the real world, and how the knowledge obtained from such analysis can be applied for devising interactive and automatic scene modeling tools for scene synthesis and re-designing purposes.

We will separate our talk into 3 main components: scene analysis, scene datasets and scene creation tools:

1) Scene Analysis. First, we will discuss the approaches for analyzing scenes captured from the real world and modeled by human artists, in terms of semantics, functionality, affordance, aesthetics, interaction context and so forth. The knowledge obtained from such analysis will enable smart AR/VR applications to be developed based on the scenes. We will also discuss advanced scene modeling tools enabled by the latest scene understanding work, such as suggestive scene modeling interfaces and context-based search for 3D scene models.

2) Scene Datasets. As data-driven approaches are important and promising for indoor scene understanding and modeling research, we will also discuss the various indoor scene datasets available (e.g., SUN3D Database, SceneNN) and their potential applications.

3) Scene Creation Tools. We will also discuss how procedural modeling techniques can be applied to automatically re-design virtual worlds. Recent works that focus on generating activity-centric indoor scenes and on text-driven scene synthesis will be discussed, with an emphasis on their potential to bring new virtual world and design applications. In addition to discussing the algorithms driving the techniques, we will put emphasis on how such techniques can be used for creating automatic and interactive 3D scene modeling tools.


We will summarize the techniques covered in our talks. We will also discuss the application potential of these techniques, especially for 3D content creation for VR/AR applications, and the promising directions for future work. We will then hold a discussion session to provide an opportunity for the audience to discuss with the speakers on interesting topics related to virtual world modeling. SUMMARY We will set up a course website, which will contain the schedule, slides and links to the video recording of the talks, similar to the websites we set up for our previous workshops. We will promote our course through the internet, emails and social media (e.g., Facebook), to reach out to a large body of audience from different backgrounds. Based on our rich experience organizing previous events in graphics and vision conferences, we are confident that the proposed course will achieve great success.