Aligned Contrastive Loss for Long-Tailed Recognition
This work addresses the challenge of imbalanced data distributions in computer vision, which is a common problem in real-world applications, though it appears to be an incremental improvement over existing contrastive learning methods.
The paper tackles the long-tailed recognition problem by proposing an Aligned Contrastive Learning (ACL) algorithm that addresses gradient conflicts and imbalances in supervised contrastive learning, achieving new state-of-the-art performance on multiple benchmarks including CIFAR, ImageNet, Places, and iNaturalist datasets.
In this paper, we propose an Aligned Contrastive Learning (ACL) algorithm to address the long-tailed recognition problem. Our findings indicate that while multi-view training boosts the performance, contrastive learning does not consistently enhance model generalization as the number of views increases. Through theoretical gradient analysis of supervised contrastive learning (SCL), we identify gradient conflicts, and imbalanced attraction and repulsion gradients between positive and negative pairs as the underlying issues. Our ACL algorithm is designed to eliminate these problems and demonstrates strong performance across multiple benchmarks. We validate the effectiveness of ACL through experiments on long-tailed CIFAR, ImageNet, Places, and iNaturalist datasets. Results show that ACL achieves new state-of-the-art performance.