Literature Review: Synthetic Data

We are actively developing classifiers to detect objects of interest in human rights research, using synthetically generated data to supplement the unavailability of real, annotated data in many cases. For an overview of this project's relationship to mtriage, see our blog post 'Computer Vision in Triple Chaser'.

In December 2019 we presented a paper at NeurIPS detailing metrics on our approach so far in partnership with engineers at ElementAI, titled 'Objects of violence: synthetic data for practical ML in human rights investigations'.

This wiki overviews relevant literature relating to and extending from this research.

Most Relevant

• Training Deep Networks with Synthetic Data: Bridging the Reality Gap by Domain Randomization, NVIDIA and University of Toronto. May 2018.
• Structured Domain Randomization: Bridging the Reality Gap by Context-Aware Synthetic Data, NVIDIA (similar authors). May 2019.

Further Reading

• Syn2Real: A New Benchmark for Synthetic-to-Real Visual Domain Adaptation