Facial Landmark Detection Using Synthetic Data

By Roey Ron / Datagen Research Team

In this benchmark report, we show how we use synthetic data, generated with the Datagen Platform, to:

1. Train a model with synthetic data and achieve results that are comparable to training with real data only, while using a test set composed of real images.
2. Combine synthetic data with different sizes of real datasets for an additional performance gain on real test data.
3. Train a model with our synthetic data and fine-tune with only 50 real data points – to get results which are equivalent to training with more than 250 real labeled images.

Download this free benchmark report to learn about:

• our experiment design and results
• how to deal with the domain gaps that arise when mixing synthetic and real data
• our facial landmark detection training recipe