DETReg: Unsupervised Pretraining with Region Priors for Object Detection
This addresses the need for more effective unsupervised pretraining in object detection, particularly for scenarios with limited labeled data, though it is incremental as it builds on existing DETR detectors.
The paper tackles the problem of self-supervised pretraining for object detection by introducing DETReg, which pretrains the entire detection network including localization and embedding components, resulting in improved performance on benchmarks like COCO, PASCAL VOC, and Airbus Ship, especially in low-data regimes such as with only 1% of labels.
Recent self-supervised pretraining methods for object detection largely focus on pretraining the backbone of the object detector, neglecting key parts of detection architecture. Instead, we introduce DETReg, a new self-supervised method that pretrains the entire object detection network, including the object localization and embedding components. During pretraining, DETReg predicts object localizations to match the localizations from an unsupervised region proposal generator and simultaneously aligns the corresponding feature embeddings with embeddings from a self-supervised image encoder. We implement DETReg using the DETR family of detectors and show that it improves over competitive baselines when finetuned on COCO, PASCAL VOC, and Airbus Ship benchmarks. In low-data regimes DETReg achieves improved performance, e.g., when training with only 1% of the labels and in the few-shot learning settings.