# Learning to Train with Synthetic Humans

## TL;DR

This paper investigates the use of synthetic humans for training neural networks in multi-person 2D pose estimation, finding that augmented real datasets with synthetic data and an adversarial training framework improve generalization to real data.

## Contribution

It introduces a novel adversarial student-teacher framework that selectively uses synthetic samples to enhance training for pose estimation.

## Key findings

- Augmented real datasets with synthetic humans perform best.
- Not all synthetic samples are equally informative for training.
- The student-teacher framework outperforms standard training on synthetic data.

## Abstract

Neural networks need big annotated datasets for training. However, manual annotation can be too expensive or even unfeasible for certain tasks, like multi-person 2D pose estimation with severe occlusions. A remedy for this is synthetic data with perfect ground truth. Here we explore two variations of synthetic data for this challenging problem; a dataset with purely synthetic humans and a real dataset augmented with synthetic humans. We then study which approach better generalizes to real data, as well as the influence of virtual humans in the training loss. Using the augmented dataset, without considering synthetic humans in the loss, leads to the best results. We observe that not all synthetic samples are equally informative for training, while the informative samples are different for each training stage. To exploit this observation, we employ an adversarial student-teacher framework; the teacher improves the student by providing the hardest samples for its current state as a challenge. Experiments show that the student-teacher framework outperforms normal training on the purely synthetic dataset.