Multi-View Correlation Distillation for Incremental Object Detection
This addresses the problem of updating object detection models with new classes without access to old data, which is crucial for real-world applications where data privacy and storage are concerns, though it is an incremental improvement over existing methods.
The paper tackles catastrophic forgetting in incremental object detection by proposing Multi-View Correlation Distillation (MVCD), which uses correlation distillation losses from multiple views to regularize learning, achieving effective mitigation of forgetting and learning of new classes as demonstrated on VOC2007 and COCO datasets.
In real applications, new object classes often emerge after the detection model has been trained on a prepared dataset with fixed classes. Due to the storage burden and the privacy of old data, sometimes it is impractical to train the model from scratch with both old and new data. Fine-tuning the old model with only new data will lead to a well-known phenomenon of catastrophic forgetting, which severely degrades the performance of modern object detectors. In this paper, we propose a novel \textbf{M}ulti-\textbf{V}iew \textbf{C}orrelation \textbf{D}istillation (MVCD) based incremental object detection method, which explores the correlations in the feature space of the two-stage object detector (Faster R-CNN). To better transfer the knowledge learned from the old classes and maintain the ability to learn new classes, we design correlation distillation losses from channel-wise, point-wise and instance-wise views to regularize the learning of the incremental model. A new metric named Stability-Plasticity-mAP is proposed to better evaluate both the stability for old classes and the plasticity for new classes in incremental object detection. The extensive experiments conducted on VOC2007 and COCO demonstrate that MVCD can effectively learn to detect objects of new classes and mitigate the problem of catastrophic forgetting.