Data Enhancement

We develop robust cycles to improve synthetic datasets by verifying in real time the impact of synthetic data on performance.


Real Baseline

Split the real world labeled dataset by {train/val/test %} to determine the performance of the real world baseline.

Synthetic Only

Train and validate on our synthetic data and test on the same held out dataset used in baseline experiment.

Domain Adapted

Perform a domain adaptation step to enhance our synthetic data with real unlabeled images and train to get better results.

Transfer Learning

Apply transfer learning by fine tuning with a small subset of the labeled data to achieve the best performance.

Our Framework