Corpus ID: 203626844

Training Kinetics in 15 Minutes: Large-scale Distributed Training on Videos

@article{Lin2019TrainingKI,
  title={Training Kinetics in 15 Minutes: Large-scale Distributed Training on Videos},
  author={J. Lin and Chuang Gan and S. Han},
  journal={ArXiv},
  year={2019},
  volume={abs/1910.00932}
}
  • J. Lin, Chuang Gan, S. Han
  • Published 2019
  • Computer Science, Engineering
  • ArXiv
  • Deep video recognition is more computationally expensive than image recognition, especially on large-scale datasets like Kinetics [1]. Therefore, training scalability is essential to handle a large amount of videos. In this paper, we study the factors that impact the training scalability of video networks. We recognize three bottlenecks, including data loading (data movement from disk to GPU), communication (data movement over networking), and computation FLOPs. We propose three design… CONTINUE READING
    3 Citations
    Building BROOK: A Multi-modal and Facial Video Database for Human-Vehicle Interaction Research
    • 1
    • PDF
    1.4 The Future of Computing: Bits + Neurons + Qubits
    • D. Gil, W. Green
    • Mathematics, Computer Science
    • 2020 IEEE International Solid- State Circuits Conference - (ISSCC)
    • 2020
    The Future of Computing: Bits + Neurons + Qubits
    • 4
    • PDF

    References

    SHOWING 1-10 OF 24 REFERENCES
    TSM: Temporal Shift Module for Efficient Video Understanding
    • 167
    • PDF
    Highly Scalable Deep Learning Training System with Mixed-Precision: Training ImageNet in Four Minutes
    • 192
    • PDF
    Scaling SGD Batch Size to 32K for ImageNet Training
    • 224
    • PDF
    ImageNet Training in Minutes
    • 208
    • PDF
    Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
    • 1,311
    • Highly Influential
    • PDF
    Large-Scale Video Classification with Convolutional Neural Networks
    • 4,172
    • PDF
    Horovod: fast and easy distributed deep learning in TensorFlow
    • 356
    • PDF
    Large Batch Training of Convolutional Networks
    • 131
    • PDF
    Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset
    • 1,931
    • Highly Influential
    • PDF