Pair DETR: Contrastive Learning Speeds Up DETR Training
This addresses a key bottleneck in DETR training for object detection applications, offering significant speed improvements while maintaining or enhancing accuracy.
The paper tackles the slow convergence problem of DETR object detection by using a representation learning technique that detects objects as paired keypoints with contrastive learning, resulting in at least 10x faster convergence than original DETR and higher Average Precision scores on MS COCO.
The DETR object detection approach applies the transformer encoder and decoder architecture to detect objects and achieves promising performance. In this paper, we present a simple approach to address the main problem of DETR, the slow convergence, by using representation learning technique. In this approach, we detect an object bounding box as a pair of keypoints, the top-left corner and the center, using two decoders. By detecting objects as paired keypoints, the model builds up a joint classification and pair association on the output queries from two decoders. For the pair association we propose utilizing contrastive self-supervised learning algorithm without requiring specialized architecture. Experimental results on MS COCO dataset show that Pair DETR can converge at least 10x faster than original DETR and 1.5x faster than Conditional DETR during training, while having consistently higher Average Precision scores.