RDR-KD: A Knowledge Distillation Detection Framework for Drone Scenes
- 8 May 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 21 (1545598X) , 1-5
Abstract
Drone object detection (DOD) with real-time deployment is a research hotspot. On the one hand, the performance of tiny object detection is closely related to the ground detection capability of the drone platform. Existing methods are keen on designing complex networks to enhance the accuracy of tiny objects, which significantly increases computational costs. On the other hand, the limited drone hardware resources urgently require lightweight models for deployment. To address the dilemma of balancing detection accuracy and computational efficiency, we propose a regenerated-decoupled-responsive knowledge distillation (RDR-KD) framework specifically for drone scenes. First, we design the Regenerated Distillation and the Decoupled Distillation to fully transfer the tiny object feature information from the teacher model to the student model. Meanwhile, we devise the logit-based Responsive Distillation based on focal loss and efficient intersection over union (EIoU) to alleviate class imbalance. Finally, we conduct extensive experiments on the VisDrone2019 dataset. The experimental results demonstrate that the proposed RDR-KD framework improves AP and ${\mathrm{AP}}_{S}$ of the student model by 3.3% and 2.9% respectively, which outperforms other state-of-the-art distillation frameworks.Keywords
Funding Information
- National Natural Science Foundation of China General Program (82272130)
- Aeronautical Science Foundation (2023Z019072001)
- National Natural Science Foundation of China Key Program (U22A20103)
This publication has 16 references indexed in Scilit:
- Bridging Cross-task Protocol Inconsistency for Distillation in Dense Object Detection2023
- From Knowledge Distillation to Self-Knowledge Distillation: A Unified Approach with Normalized Loss and Customized Soft Labels2023
- AMD: Adaptive Masked Distillation for Object Detection2023
- Masked Generative Distillation2022
- Focal and Global Knowledge Distillation for Detectors2022
- Decoupled Knowledge Distillation2022
- Localization Distillation for Dense Object Detection2022
- Channel-wise Knowledge Distillation for Dense Prediction2021
- Distilling Object Detectors via Decoupled Features2021
- Focal Loss for Dense Object Detection2017