SMPL made Simple

CVPR 2021 Tutorial (June 20, 10:00-18:00 EDT)

Note: even if you missed the tutorial, you will find recordings of all presentations on this site.

Naureen Mahmood, Timo Bolkart, Ahmed A. A. Osman, Joachim Tesch, Dimitrios Tzionas, Michael J. Black

Since the release of the SMPL body model in 2015, it has been widely adopted in academia and industry. SMPL has enabled a new sub-field of computer vision that focuses on “human mesh recovery” as well as research on human contact with scenes and objects. In industry it has enabled applications in clothing sizing, clothing design, fitness, and animation. It is further widely used as ground truth for training neural networks to understand human pose and shape.

This tutorial focuses on SMPL, what it is, and how to use it. The instructors and presenters include both creators of the technology as well as researchers at the current forefront of applying it. Since SMPL’s introduction, there have been many variants introduced and the model gets used in many ways. The team has collected frequently asked questions that form the core of the tutorial and will encourage attendees to submit more questions in advance.

Full-Day Tutorial

The tutorial will be virtual and is split into two parts covering a full day. The morning presentations introduce SMPL, discuss issues, offer solutions, and give insights. These presentations are given by the creators of SMPL and its derivatives. We will present the inside scoop with little-known tricks and tips.

After building the fundamental understanding, the afternoon session focuses on using SMPL with presentations by leading researchers both from within the team and outside. This exposes the attendees to state-of-the-art use cases.

There will be ample time to discuss issues with using SMPL and how researchers address these. Finally, we will end with a discussion about future directions, unresolved issues, and how to build a lasting community of users.

Who is it for?

The course is designed for computer vision students, researchers, and industry users interested in 3D body modeling, human pose estimation, behavior analysis, animation, and applications. The course assumes basic knowledge of the current tools in computer vision (e.g. deep neural networks) and familiarity with 3D representations like meshes. The course is ideal for anyone who needs a 3D body model for animation, pose estimation, training, etc.

Attendees will walk away with a state-of-the-art 3D body model that they can readily use for their own research purposes. Because it is virtual, all materials from the tutorial will be provided on-line as a form of ongoing documentation and support for SMPL users.

Instructors

The tutorial is taught by a team from the Max Planck Institute for Intelligent Systems, who have spent years developing 3D human body models for vision, graphics and learning. The MPI team is joined by Meshcapade GmbH, which supplies industry with SMPL and supports numerous applications.

Schedule:

Morning session: The basics, FAQs, how-to

Something to watch while you wait for the tutorial to start:

10:00-11:20 Introduction to the tutorial and learning objectives. Overview of 3D body models, the history, mesh registration, linear blend skinning, SMPL and related models. Instructor: Michael Black