Co-Learning Meets Stitch-Up for Noisy Multi-label Visual Recognition

TL;DR

This paper introduces a novel co-learning framework with Stitch-Up augmentation to effectively handle noisy, long-tailed, multi-label visual data, improving robustness and accuracy in real-world scenarios.

Contribution

It proposes a Stitch-Up augmentation method and a heterogeneous co-learning framework specifically designed for noisy, long-tailed multi-label visual recognition tasks.

Findings

Outperforms baseline methods on VOC-MLT-Noise and COCO-MLT-Noise benchmarks.

Achieves more robust label noise reduction in long-tailed multi-label data.

Demonstrates significant improvements in recognition accuracy under noisy conditions.

Abstract

In real-world scenarios, collected and annotated data often exhibit the characteristics of multiple classes and long-tailed distribution. Additionally, label noise is inevitable in large-scale annotations and hinders the applications of learning-based models. Although many deep learning based methods have been proposed for handling long-tailed multi-label recognition or label noise respectively, learning with noisy labels in long-tailed multi-label visual data has not been well-studied because of the complexity of long-tailed distribution entangled with multi-label correlation. To tackle such a critical yet thorny problem, this paper focuses on reducing noise based on some inherent properties of multi-label classification and long-tailed learning under noisy cases. In detail, we propose a Stitch-Up augmentation to synthesize a cleaner sample, which directly reduces multi-label noise by stitching up multiple noisy training samples. Equipped with Stitch-Up, a Heterogeneous Co-Learning framework is further designed to leverage the inconsistency between long-tailed and balanced distributions, yielding cleaner labels for more robust representation learning with noisy long-tailed data. To validate our method, we build two challenging benchmarks, named VOC-MLT-Noise and COCO-MLT-Noise, respectively. Extensive experiments are conducted to demonstrate the effectiveness of our proposed method. Compared to a variety of baselines, our method achieves superior results.