Teach YOLO to Remember: A Self-Distillation Approach for Continual Object Detection
This addresses the problem of forgetting in real-time object detection for incremental data scenarios, offering a tailored solution for YOLO, though it is incremental as it builds on prior continual learning methods.
The paper tackles catastrophic forgetting in YOLO-based continual object detection by introducing YOLO LwF, a self-distillation approach with replay memory, achieving state-of-the-art performance with mAP improvements of +2.1% on VOC and +2.9% on COCO.
Real-time object detectors like YOLO achieve exceptional performance when trained on large datasets for multiple epochs. However, in real-world scenarios where data arrives incrementally, neural networks suffer from catastrophic forgetting, leading to a loss of previously learned knowledge. To address this, prior research has explored strategies for Class Incremental Learning (CIL) in Continual Learning for Object Detection (CLOD), with most approaches focusing on two-stage object detectors. However, existing work suggests that Learning without Forgetting (LwF) may be ineffective for one-stage anchor-free detectors like YOLO due to noisy regression outputs, which risk transferring corrupted knowledge. In this work, we introduce YOLO LwF, a self-distillation approach tailored for YOLO-based continual object detection. We demonstrate that when coupled with a replay memory, YOLO LwF significantly mitigates forgetting. Compared to previous approaches, it achieves state-of-the-art performance, improving mAP by +2.1% and +2.9% on the VOC and COCO benchmarks, respectively.