In this paper, a lightweight Vision Transformer (ViT) model based on Support Vector Machine (SVM) optimization is proposed, which aims to improve the accuracy and efficiency of the model in object detection and segmentation tasks, especially in resource-constrained environments. In this paper, SVM is used to optimize the structure and hyperparameters of the ViT model, which reduces the computational complexity and storage requirements of the model and maintains high detection and segmentation accuracy. Experimental results show that the average detection accuracy (mAP) of the optimized model on the COCO dataset is increased by 2.8 percentage points compared with the standard ViT model, and the inference speed is increased by 18%. In addition, the IoU of the optimized model in the segmentation task is increased by 2.7%, and the average segmentation time is reduced by 20%. These results prove the effectiveness of the SVM optimization strategy in the design of lightweight models and provide new ideas for the practical deployment of ViT models.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Research on Structure and Hyperparameter Optimization of Lightweight ViT Model Based on SVM Optimization in Object Detection


    Beteiligte:
    Zhang, Yuanlong (Autor:in) / Wu, Qi (Autor:in) / Liu, Guiran (Autor:in) / Tian, Jinhao (Autor:in) / Liang, Shenrui (Autor:in) / Zhang, Cheng (Autor:in)


    Erscheinungsdatum :

    23.10.2024


    Format / Umfang :

    514116 byte




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch



    Driver stress levels detection system using hyperparameter optimization

    Rastgoo, Mohammad Naim / Nakisa, Bahareh / Rakotonirainy, Andry et al. | Taylor & Francis Verlag | 2024



    A Hyperparameter Quality Assessment Method for UAV Object Detection Based on IER Rule

    Kang, Xiao / Mu, Quanqi / Han, Wence et al. | Springer Verlag | 2023