Towards a category-extended object detector with limited data
This work tackles the practical problem of extending object detector categories with limited data for practitioners who need to update their models incrementally.
This paper addresses the challenge of progressively increasing object detector categories with limited data. They propose a scheme that includes a conflict-free loss for initial training and a retraining phase utilizing Monte Carlo Dropout for localization confidence and an overlap-weighted method for pseudo annotations.
Object detectors are typically learned on fully-annotated training data with fixed predefined categories. However, categories are often required to be increased progressively. Usually, only the original training set annotated with old classes and some new training data labeled with new classes are available in such scenarios. Based on the limited datasets, a unified detector that can handle all categories is strongly needed. We propose a practical scheme to achieve it in this work. A conflict-free loss is designed to avoid label ambiguity, leading to an acceptable detector in one training round. To further improve performance, we propose a retraining phase in which Monte Carlo Dropout is employed to calculate the localization confidence to mine more accurate bounding boxes, and an overlap-weighted method is proposed for making better use of pseudo annotations during retraining. Extensive experiments demonstrate the effectiveness of our method.