Unite the People: Closing the Loop Between 3D and 2D Human Representations

TLDR

3D models unify human body representations, and robust 2D estimation enables in‑the‑wild 3D fitting, but large‑scale labeled data for 2D estimators remains difficult to obtain. The authors propose a hybrid approach that extends SMPLify to generate high‑quality 3D body model fits across multiple human pose datasets. Human annotators then sort the fits as good or bad to create the UP‑3D dataset. The UP‑3D dataset enables training models that predict 31 segments and 91 landmarks, achieving state‑of‑the‑art 3D pose and shape estimation with an order‑of‑magnitude less training data and no gender or pose assumptions, and the dataset, code, and models are publicly available.

Abstract

3D models provide a common ground for different representations of human bodies. In turn, robust 2D estimation has proven to be a powerful tool to obtain 3D fits in-the-wild. However, depending on the level of detail, it can be hard to impossible to acquire labeled data for training 2D estimators on large scale. We propose a hybrid approach to this problem: with an extended version of the recently introduced SMPLify method, we obtain high quality 3D body model fits for multiple human pose datasets. Human annotators solely sort good and bad fits. This procedure leads to an initial dataset, UP-3D, with rich annotations. With a comprehensive set of experiments, we show how this data can be used to train discriminative models that produce results with an unprecedented level of detail: our models predict 31 segments and 91 landmark locations on the body. Using the 91 landmark pose estimator, we present state-of-the art results for 3D human pose and shape estimation using an order of magnitude less training data and without assumptions about gender or pose in the fitting procedure. We show that UP-3D can be enhanced with these improved fits to grow in quantity and quality, which makes the system deployable on large scale. The data, code and models are available for research purposes.