MURAL - Maynooth University Research Archive Library



    Evaluating Pruned Object Detection Networks for Real-Time Robot Vision


    O'Keefe, Simon and Villing, Rudi (2018) Evaluating Pruned Object Detection Networks for Real-Time Robot Vision. In: 2018 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC). IEEE, pp. 91-96. ISBN 9781538652213

    [img]
    Preview
    Download (330kB) | Preview


    Share your research

    Twitter Facebook LinkedIn GooglePlus Email more...



    Add this article to your Mendeley library


    Abstract

    Convolutional Neural Networks are the state of the art for computer vision problems such as classification and detection. Networks like YOLO and SSD have demonstrated excellent results on benchmark datasets such as the PASCAL VOC and COCO datasets. However these networks only run at real time with the support of powerful GPUs and are infeasible for use in low power embedded real-time robotic applications. Pruning has been shown to be an efficient technique for reducing the runtime computational cost of a neural network while maintaining performance in image classification tasks. In this work we evaluate the efficacy of pruning on the problem of object detection using a modified tiny-YOLO network. The network was trained on a custom object detection task and three pruning techniques were evaluated, including our contribution which specifically targets reducing the FLOPS in the network. The results show that pruning with our method followed by extended fine-tuning achieved a 4.5x reduction in FLOPS and a 7x reduction in parameters with no drop in accuracy.

    Item Type: Book Section
    Additional Information: Cite as: S. O'Keeffe and R. Villing, "Evaluating pruned object detection networks for real-time robot vision," 2018 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC), Torres Vedras, 2018, pp. 91-96, doi: 10.1109/ICARSC.2018.8374166.
    Keywords: Convolutional Neural Networks; object detection; real-time; pruning;
    Academic Unit: Faculty of Science and Engineering > Electronic Engineering
    Faculty of Science and Engineering > Research Institutes > Hamilton Institute
    Item ID: 13384
    Identification Number: https://doi.org/10.1109/ICARSC.2018.8374166
    Depositing User: Rudi Villing
    Date Deposited: 02 Oct 2020 16:53
    Publisher: IEEE
    Refereed: Yes
    URI:

    Repository Staff Only(login required)

    View Item Item control page

    Downloads

    Downloads per month over past year