Unsupervised Object-Level Representation Learning from Scene Images
This addresses the challenge of unsupervised representation learning for scene images, which is crucial for applications in computer vision but has been limited by reliance on curated data, offering a more general-purpose approach.
The paper tackles the problem of self-supervised learning on complex scene images with multiple objects, where existing methods rely on object-centric priors from curated datasets like ImageNet. It introduces Object-level Representation Learning (ORL), which significantly improves performance on COCO, surpassing supervised ImageNet pre-training on several downstream tasks and showing better results with more unlabeled scene data.
Contrastive self-supervised learning has largely narrowed the gap to supervised pre-training on ImageNet. However, its success highly relies on the object-centric priors of ImageNet, i.e., different augmented views of the same image correspond to the same object. Such a heavily curated constraint becomes immediately infeasible when pre-trained on more complex scene images with many objects. To overcome this limitation, we introduce Object-level Representation Learning (ORL), a new self-supervised learning framework towards scene images. Our key insight is to leverage image-level self-supervised pre-training as the prior to discover object-level semantic correspondence, thus realizing object-level representation learning from scene images. Extensive experiments on COCO show that ORL significantly improves the performance of self-supervised learning on scene images, even surpassing supervised ImageNet pre-training on several downstream tasks. Furthermore, ORL improves the downstream performance when more unlabeled scene images are available, demonstrating its great potential of harnessing unlabeled data in the wild. We hope our approach can motivate future research on more general-purpose unsupervised representation learning from scene data.