Skip to content

Latest commit

 

History

History
59 lines (44 loc) · 2.09 KB

README.MD

File metadata and controls

59 lines (44 loc) · 2.09 KB

License: CC BY-NC 4.0

V4D: 4D Convolutional Neural Networks for Video-level Representation Learning

Code for the ICLR 2020 paper V4D: 4D Convolutional Neural Networks for Video-level Representation Learning

Performance compared with SOTA methods on Kinetics

Model Backbone Top1 Top5
ARTNet with TSN ARTNet ResNet18 70.7 89.3
ECO BN-Inception+3D ResNet18 70.0 89.4
S3D-G S3D Inception 74.7 93.4
Nonlocal Network 3D ResNet50 76.5 92.6
SlowFast SlowFast ResNet50 77.0 92.6
I3D I3D Inception 72.1 90.3
Two-stream I3D I3D Inception 75.7 92.0
I3D-S Slow pathway ResNet50 74.9 91.5
Ours V4D V4D ResNet50 77.4 93.1

Requirements

  • Python >=3.6
  • PyTorch >=1.3
  • torchvision that matches the PyTorch installation.

Train on Kinetics and Mini-kinetics

./scripts/train_kinetics.sh
./scripts/train_minikinetics.sh

Test pretrained model on Mini-Kinetics-200 and Kinetics (download our trained model from Google)

./scripts/test_kinetics.sh
./scripts/test_minikinetics.sh

Contact

For any questions, please feel free to reach

If you use this method or this code in your research, please cite as:

@inproceedings{zhang2020v4d,
title={V4D: 4D Convolutional Neural Networks for Video-level Representation Learning},
author={Zhang, Shiwen and Guo, Sheng and Huang, Weilin and Scott, Matthew R and Wang, Limin},
booktitle={Proceedings of International Conference on Learning Representations},
year={2020}
}

License

V4D is CC-BY-NC 4.0 licensed, as found in the LICENSE file. It is released for academic research / non-commercial use only. If you wish to use for commercial purposes, please contact [email protected].