Header logo is ps

Learning to Train with Synthetic Humans


Conference Paper


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, as well as 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. 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 this student-teacher framework outperforms all our baselines.

Author(s): David T. Hoffmann and Dimitrios Tzionas and Michael J. Black and Siyu Tang
Book Title: German Conference on Pattern Recognition (GCPR)
Pages: 609--623
Year: 2019
Month: September
Publisher: Springer International Publishing

Department(s): Perceiving Systems
Research Project(s): Learning from Synthetic Data
Bibtex Type: Conference Paper (inproceedings)
Paper Type: Conference

DOI: https://doi.org/10.1007/978-3-030-33676-9_43

ISBN: 978-3-030-33676-9
URL: https://ltsh.is.tue.mpg.de
Attachments: pdf


  title = {Learning to Train with Synthetic Humans},
  author = {Hoffmann, David T. and Tzionas, Dimitrios and Black, Michael J. and Tang, Siyu},
  booktitle = {German Conference on Pattern Recognition (GCPR)},
  pages = {609--623},
  publisher = {Springer International Publishing},
  month = sep,
  year = {2019},
  url = {https://ltsh.is.tue.mpg.de},
  month_numeric = {9}