From Real to Synthetic and Back: Synthesizing Training Data for Multi-Person Scene Understanding

TL;DR

This paper introduces a pipeline for generating realistic multi-person images from synthetic data, reducing the domain gap, and improving training for scene understanding tasks like UV mapping and depth estimation.

Contribution

We propose a novel synthetic data generation pipeline using a cGAN to translate segmentation maps into realistic images, enhancing multi-person scene understanding.

Findings

Generated data improves model performance on UV mapping and depth estimation.

The pipeline effectively reduces the synthetic-to-real domain gap.

Models trained on synthetic data achieve competitive results on real datasets.

Abstract

We present a method for synthesizing naturally looking images of multiple people interacting in a specific scenario. These images benefit from the advantages of synthetic data: being fully controllable and fully annotated with any type of standard or custom-defined ground truth. To reduce the synthetic-to-real domain gap, we introduce a pipeline consisting of the following steps: 1) we render scenes in a context modeled after the real world, 2) we train a human parsing model on the synthetic images, 3) we use the model to estimate segmentation maps for real images, 4) we train a conditional generative adversarial network (cGAN) to learn the inverse mapping -- from a segmentation map to a real image, and 5) given new synthetic segmentation maps, we use the cGAN to generate realistic images. An illustration of our pipeline is presented in Figure 2. We use the generated data to train a multi-task model on the challenging tasks of UV mapping and dense depth estimation. We demonstrate the value of the data generation and the trained model, both quantitatively and qualitatively on the CMU Panoptic Dataset.