Realistically distributing object placements in synthetic training data improves the performance of vision-based object detection models
This addresses the challenge of synthetic-to-real domain adaptation for vision-based object detection, but it is incremental as it focuses on a specific aspect of data distribution.
The study tackled the problem of improving object detection models by making synthetic training data more realistic, specifically through better object placement distribution, and achieved a substantial improvement in performance when tested on real data.
When training object detection models on synthetic data, it is important to make the distribution of synthetic data as close as possible to the distribution of real data. We investigate specifically the impact of object placement distribution, keeping all other aspects of synthetic data fixed. Our experiment, training a 3D vehicle detection model in CARLA and testing on KITTI, demonstrates a substantial improvement resulting from improving the object placement distribution.