Logit knowledge distillation attracts increasing attention due to its practicality in recent studies. However, it often suffers inferior performance compared to the feature knowledge distillation. In this paper, we argue that existing logit-based methods may be sub-optimal since they only leverage the global logit output that couples multiple semantic knowledge. This may transfer ambiguous knowledge to the student and mislead its learning. To this end, we propose a simple but effective method, i.e., Scale Decoupled Distillation (SDD), for logit knowledge distillation. SDD decouples the global logit output into multiple local logit outputs and establishes distillation pipelines for them. This helps the student to mine and inherit fine-grained and unambiguous logit knowledge. Moreover, the decoupled knowledge can be further divided into consistent and complementary logit knowledge that transfers the semantic information and sample ambiguity, respectively. By increasing the weight of complementary parts, SDD can guide the student to focus more on ambiguous samples, improving its discrimination ability. Extensive experiments on several benchmark datasets demonstrate the effectiveness of SDD for wide teacher-student pairs, especially in the fine-grained classification task. Code is available at: https://github.com/shicaiwei123/SDD-CVPR2024


    Access

    Download


    Export, share and cite



    Title :

    Scale Decoupled Distillation


    Contributors:

    Publication date :

    2024


    Remarks:

    Accepted to CVPR2024 10 pages 6figure


    Type of media :

    Preprint


    Type of material :

    Electronic Resource


    Language :

    English




    Decoupled Knowledge Distillation

    Zhao, Borui / Cui, Quan / Song, Renjie et al. | ArXiv | 2022

    Free access

    Decoupled multi-scale distillation for medical image segmentation

    Zhang, Dingwen / Yu, Xiangchun | SPIE | 2024


    Empowering Object Detection: Unleashing the Potential of Decoupled and Interactive Distillation

    Qian, Fulan / Hong, Jiacheng / Yan, Huanqian et al. | IEEE | 2025


    Decoupled rotor

    European Patent Office | 2022

    Free access

    DECOUPLED ELECTRONIC BRAKE SYSTEM

    European Patent Office | 2024

    Free access