Clustering Visual Similar Objects for Enhanced Synthetic Image Data for Object Detection

Object detection often struggles with accurately identifying visually similar parts, a challenge commonly faced in industrial applications. To address this issue, we propose a clustering methodology based on the visual similarity of 3D object models. This approach is particularly effective when integrated with synthetic image generation, as both processes rely on 3D models. In this case study, we observed more than a 20% increase in classification performance on two different object detector architectures on a validation dataset when training an object detector on visually similar groups rather than on all classes, suggesting the potential of our method as a baseline for a multi-stage classification scheme.