arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2303.08360cs.CV
收藏

从单标签到多标签的知识蒸馏实证研究

Knowledge Distillation from Single to Multi Labels: an Empirical Study

OPPO Research Institute(OPPO研究院);Fudan University(复旦大学)

Youcai Zhang, Yuzhuo Qin, Hengwei Liu, Yanhao Zhang, Yaqian Li, Xiaodong Gu

摘要:

知识蒸馏(KD)在单标记图像分类中得到了广泛的研究。然而,其多标签分类的功效仍然相对未被探索。在这项研究中,我们首先研究了经典的KD技术,包括基于logit和基于特征的方法,多标签分类的有效性。我们的研究结果表明,基于logit的方法是不适合多标签分类,教师未能提供类别间的相似性信息或正则化效果的学生模型的训练。此外,我们观察到,基于特征的方法很难同时传达多个标签的紧凑信息。鉴于这些限制,我们建议合适的暗知识应该包含类信息,并与最终分类结果高度相关。为了解决这些问题,我们介绍了一种新的蒸馏方法的基础上类激活图(CAM),这是有效的和简单的实现。在广泛的设置范围内,基于CAM的蒸馏始终优于其他方法。

英文摘要:

Knowledge distillation (KD) has been extensively studied in single-label image classification. However, its efficacy for multi-label classification remains relatively unexplored. In this study, we firstly investigate the effectiveness of classical KD techniques, including logit-based and feature-based methods, for multi-label classification. Our findings indicate that the logit-based method is not well-suited for multi-label classification, as the teacher fails to provide inter-category similarity information or regularization effect on student model's training. Moreover, we observe that feature-based methods struggle to convey compact information of multiple labels simultaneously. Given these limitations, we propose that a suitable dark knowledge should incorporate class-wise information and be highly correlated with the final classification results. To address these issues, we introduce a novel distillation method based on Class Activation Maps (CAMs), which is both effective and straightforward to implement. Across a wide range of settings, CAMs-based distillation consistently outperforms other methods.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑