LiDAR-based 3D object detection plays an essential role in autonomous driving. Although the detector trained on high-resolution data has much better performance than the same detector trained on low-resolution data, the high-resolution LiDAR cannot be widely used due to its high price. In this work, we propose a new distillation method called Focal Distillation to bridge the gap between high-resolution detector (teacher model) and low-resolution detector (student model). It consists of three essential components: focal classification distillation (FCD), focal regression distillation (FRD) and focal feature distillation (FFD). Taking the low-resolution data as input, the student model can learn discriminative features and produce more accurate results with the assistance of the teacher model trained on high-resolution data. We conducted extensive experiments to validate the effectiveness of Focal Distillation. Evaluated on the KITTI validation set, a typical SECOND model trained with Focal Distillation outperformed its non-distilled counterpart by 3.37%, 7.52%, 11.35% mAP on the category Car, Pedestrian, and Cyclist of moderate level, respectively. Moreover, the remarkable improvements observed on different models and different datasets further demonstrate the generalization ability of our proposed method.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Focal Distillation From High-Resolution Data to Low-Resolution Data for 3D Object Detection


    Beteiligte:
    Shan, Jiawei (Autor:in) / Zhang, Gang (Autor:in) / Tang, Chufeng (Autor:in) / Pan, Hujie (Autor:in) / Yu, Qiankun (Autor:in) / Wu, Guanhao (Autor:in) / Hu, Xiaolin (Autor:in)

    Erschienen in:

    Erscheinungsdatum :

    01.12.2023


    Format / Umfang :

    4708366 byte




    Medientyp :

    Aufsatz (Zeitschrift)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch



    Object detection in high-resolution radar data

    Danzer, Andreas / Universität Ulm | TIBKAT | 2022

    Freier Zugriff

    DEEP WATER HIGH RESOLUTION OBJECT DETECTION

    GILLOTT GRAHAM / HIBBEN TERENCE | Europäisches Patentamt | 2020

    Freier Zugriff

    Innovative focal plane segmentation for high-resolution planetary observation

    Hein, Grégoire / Ferrari, Marc / Tetaz, Nicolas | British Library Conference Proceedings | 2019


    Resolution Effects on Polarimetric High-Resolution X-Band Data

    Garestier, F. / Dubois-Fernandez, P. / Dupuis, X. et al. | British Library Conference Proceedings | 2005


    High-resolution superconductive serial analog-to-digital converter for on-focal plane data conversion [3360-04]

    Gupta, D. / Rylov, S. V. / Gaidarenko, D. V. et al. | British Library Conference Proceedings | 1998