Continual 3D Convolutional Neural Networks for Real-time Processing of Videos

被引:3
|
作者
Hedegaard, Lukas [1 ]
Iosifidis, Alexandros [1 ]
机构
[1] Aarhus Univ, Dept Elect & Comp Engn, Aarhus, Denmark
来源
关键词
3D CNN; Human activity recognition; Efficient; Stream processing; Online inference; Continual inference network;
D O I
10.1007/978-3-031-19772-7_22
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We introduce Continual 3D Convolutional Neural Networks (Co3D CNNs), a new computational formulation of spatio-temporal 3D CNNs, in which videos are processed frame-by-frame rather than by clip. In online tasks demanding frame-wise predictions, Co3D CNNs dispense with the computational redundancies of regular 3D CNNs, namely the repeated convolutions over frames, which appear in overlapping clips. We show that Continual 3D CNNs can reuse preexisting 3D-CNN weights to reduce the per-prediction floating point operations (FLOPs) in proportion to the temporal receptive field while retaining similar memory requirements and accuracy. This is validated with multiple models on Kinetics-400 and Charades with remarkable results: CoX3D models attain state-of-the-art complexity/accuracy trade-offs on Kinetics-400 with 12.1-15.3x reductions of FLOPs and 2.3-3.8% improvements in accuracy compared to regular X3D models while reducing peak memory consumption by up to 48%. Moreover, we investigate the transient response of Co3D CNNs at start-up and perform extensive benchmarks of on-hardware processing characteristics for publicly available 3D CNNs.
引用
收藏
页码:369 / 385
页数:17
相关论文
共 50 条
  • [31] Lightweight convolutional neural network for real-time 3D object detection in road and railway environments
    Mauri, A.
    Khemmar, R.
    Decoux, B.
    Haddad, M.
    Boutteau, R.
    JOURNAL OF REAL-TIME IMAGE PROCESSING, 2022, 19 (03) : 499 - 516
  • [32] Lightweight convolutional neural network for real-time 3D object detection in road and railway environments
    A. Mauri
    R. Khemmar
    B. Decoux
    M. Haddad
    R. Boutteau
    Journal of Real-Time Image Processing, 2022, 19 : 499 - 516
  • [33] Action Recognition in Videos with Spatio-Temporal Fusion 3D Convolutional Neural Networks
    Wang, Y.
    Shen, X. J.
    Chen, H. P.
    Sun, J. X.
    PATTERN RECOGNITION AND IMAGE ANALYSIS, 2021, 31 (03) : 580 - 587
  • [34] Special issue on real-time 3D imaging and processing
    Sen Wang
    Journal of Real-Time Image Processing, 2012, 7 : 1 - 2
  • [35] Special issue on real-time 3D imaging and processing
    Wang, Sen
    JOURNAL OF REAL-TIME IMAGE PROCESSING, 2012, 7 (01) : 1 - 2
  • [36] REAL-TIME PROCESSING PIPELINE FOR 3D IMAGING APPLICATIONS
    Chaikalis, D. P.
    Sgouros, N. P.
    Maroulis, D. E.
    2009 16TH INTERNATIONAL CONFERENCE ON DIGITAL SIGNAL PROCESSING, VOLS 1 AND 2, 2009, : 607 - 612
  • [37] Stereoscopic cameras for the real-time acquisition of panoramic 3D images and videos
    Gurrieri, Luis E.
    Dubois, Eric
    STEREOSCOPIC DISPLAYS AND APPLICATIONS XXIV, 2013, 8648
  • [38] Real-time 3D
    Coco, D
    COMPUTER GRAPHICS WORLD, 1995, 18 (12) : 22 - +
  • [39] Convolutional neural networks for real-time epileptic seizure detection
    Achilles, Felix
    Tombari, Federico
    Belagiannis, Vasileios
    Loesch, Anna Mira
    Noachtar, Soheyl
    Navab, Nassir
    COMPUTER METHODS IN BIOMECHANICS AND BIOMEDICAL ENGINEERING-IMAGING AND VISUALIZATION, 2018, 6 (03): : 264 - 269
  • [40] Convolutional Neural Networks for Real-Time and Wireless Damage Detection
    Avci, Onur
    Abdeljaber, Osama
    Kiranyaz, Serkan
    Inman, Daniel
    DYNAMICS OF CIVIL STRUCTURES, VOL 2, IMAC 2019, 2020, : 129 - 136