Enhancing Object Detection Using Synthetic Examples
David C. Hughes, Hao Ji · 2021
Manual data annotation for training custom object detection can be a time-consuming and error-prone process. In this paper, we propose an automatic approach to generating synthetic, annotated images using differentiable neural rendering and 3D object models. We also investigate the possibility of using 3D adversarial object models to improve object detection accuracy. The experimental results show that the object detection models trained using both synthetic examples rendered from 3D object models and real data outperform the baseline model trained on only real data.