Divide and Contrast: Self-supervised Learning from Uncurated Data
This addresses the challenge of scaling self-supervised learning to more realistic, uncurated data for computer vision applications, though it is incremental as it builds on existing contrastive methods.
The paper tackles the problem of self-supervised learning from uncurated image data like YFCC, where representation quality suffers due to diverse and heavy-tailed class distributions. It introduces Divide and Contrast (DnC), which improves performance on downstream tasks for less curated datasets while remaining competitive with state-of-the-art on curated ones.
Self-supervised learning holds promise in leveraging large amounts of unlabeled data, however much of its progress has thus far been limited to highly curated pre-training data such as ImageNet. We explore the effects of contrastive learning from larger, less-curated image datasets such as YFCC, and find there is indeed a large difference in the resulting representation quality. We hypothesize that this curation gap is due to a shift in the distribution of image classes -- which is more diverse and heavy-tailed -- resulting in less relevant negative samples to learn from. We test this hypothesis with a new approach, Divide and Contrast (DnC), which alternates between contrastive learning and clustering-based hard negative mining. When pretrained on less curated datasets, DnC greatly improves the performance of self-supervised learning on downstream tasks, while remaining competitive with the current state-of-the-art on curated datasets.