DetCo: Unsupervised Contrastive Learning for Object Detection
This work provides a significant improvement in unsupervised representation learning for object detection, benefiting researchers and practitioners in computer vision by offering a more efficient and effective pre-training method.
This paper introduces DetCo, a novel unsupervised contrastive learning approach that explores contrasts between global images and local patches to learn discriminative representations for object detection. DetCo achieves 57.4 mAP on PASCAL VOC with 100 epochs, matching MoCov2-800ep, and consistently outperforms supervised methods by 1.6/1.2/1.0 AP on Mask RCNN-C4/FPN/RetinaNet.
Unsupervised contrastive learning achieves great success in learning image representations with CNN. Unlike most recent methods that focused on improving accuracy of image classification, we present a novel contrastive learning approach, named DetCo, which fully explores the contrasts between global image and local image patches to learn discriminative representations for object detection. DetCo has several appealing benefits. (1) It is carefully designed by investigating the weaknesses of current self-supervised methods, which discard important representations for object detection. (2) DetCo builds hierarchical intermediate contrastive losses between global image and local patches to improve object detection, while maintaining global representations for image recognition. Theoretical analysis shows that the local patches actually remove the contextual information of an image, improving the lower bound of mutual information for better contrastive learning. (3) Extensive experiments on PASCAL VOC, COCO and Cityscapes demonstrate that DetCo not only outperforms state-of-the-art methods on object detection, but also on segmentation, pose estimation, and 3D shape prediction, while it is still competitive on image classification. For example, on PASCAL VOC, DetCo-100ep achieves 57.4 mAP, which is on par with the result of MoCov2-800ep. Moreover, DetCo consistently outperforms supervised method by 1.6/1.2/1.0 AP on Mask RCNN-C4/FPN/RetinaNet with 1x schedule. Code will be released at \href{https://github.com/xieenze/DetCo}{\color{blue}{\tt github.com/xieenze/DetCo}}.