CAL-ODWC: continual adaptation learning for object detection under weather changes

Abstract

We address object detection task in adverse weather conditions. Generally, adverse weather conditions degrade object detection performance. To overcome this drawback, we propose Continual Adaptation Learning (CAL). Specifically, we utilize Developmental Memory (DM) and Regularization Method (RM) to resolve catastrophic forgetting problem. DM and RM was invented for classification task. We research a optimal method to apply DM and RM to object detection task. In result, CAL-ODWC has improved performance.