Lifelong Object Detection

By Wang Zhou, Shiyu Chang, Norma Sosa, Hendrik Hamann, David D. Cox. 2020

48 citations recorded by the source corpus.

Recent advances in object detection have benefited significantly from rapid developments in deep neural networks. However, neural networks suffer from the well-known issue of catastrophic forgetting, which makes continual or lifelong learning problematic. In this paper, we leverage the fact that new training classes arrive in a sequential manner and incrementally refine the model so that it additionally detects new object classes in the absence of previous training data. Specifically, we consider the representative object detector, Faster R-CNN, for both accurate and efficient prediction. To prevent abrupt performance degradation due to catastrophic forgetting, we propose to apply knowledge distillation on both the region proposal network and the region classification network, to retain the detection of previously trained classes. A pseudo-positive-aware sampling strategy is also introduced for distillation sample selection. We evaluate the proposed method on PASCAL VOC 2007 and MS COCO benchmarks and show competitive mAP and 6x inference speed improvement, which makes the approach more suitable for real-time applications. Our implementation will be publicly available.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Da-Wei Zhou
2. Han-Jia Ye
3. Bernt Schiele
4. De-chuan Zhan
5. Liyuan Wang
6. Ziwei Liu
7. Andrea Vedaldi
8. Dacheng Tao

Citations in the corpus

• Bridging Non Co-occurrence with Unlabeled In-the-wild Data for Incremental Object Detection (2021, neutral): “al [40] proposes an incremental version of Faster R-CNN that distills selected anchors and proposals with the PseudoPositive-Aware Sampling strategy.”
• Wanderlust: Online Continual Object Detection in the Real World (2021, background): “Most of the current research [32, 41, 49] re-purpose existing static datasets such as VOC and COCO to evaluate continual object detection.”
• Continual semi-supervised learning through contrastive interpolation consistency (2021, background): “However, when the labeling stage involves human intervention (as holds in a number of computer vision tasks such as classification, object detection [4], etc.”
• Class-Incremental Learning: A Survey (2023, background): “Similarly, [262] applies knowledge distillation on both the region proposal network and the region classification network.”
• Augmented Box Replay: Overcoming Foreground Shift for Incremental Object Detection (2023, methodology): “Implementation Details: Similar to [7,9,22,28,39,47,54, 59, 65], we use the Faster R-CNN [51] architecture with a Resnet-50 [24] backbone pretrained on ImageNet [11].”
• Continual Detection Transformer for Incremental Object Detection (2023, methodology): “Inspired by this, recent IOD methods extended the KD framework to other detectors ( e.g ., Faster-RCNN [43] and GFL [26]) by adding KD terms on the intermediate feature maps [13,56,59] and region proposal networks [8,15,38].”
• Continual Object Detection via Prototypical Task Correlation Guided Gating Mechanism (2022, unknown): “Distinct from existing distillation-based continual learning methods [34, 39, 45, 53], we use feature-level knowledge distillation to avoid the training collapse of the sparse gated model rather than alleviating catastrophic forgetting.”