Awesome

STRPM (CVPR2022)

Zheng Chang, Xinfeng Zhang, Shanshe Wang, Siwei Ma, Wen Gao.

Official PyTorch Code for "STRPM: A Spatiotemporal Residual Predictive Model for High-Resolution Video Prediction" [paper]

Requirements

• PyTorch 1.7
• CUDA 11.0
• CuDNN 8.0.5
• python 3.6.7

Installation

Create conda environment:

    $ conda create -n STRPM python=3.6.7
    $ conda activate STRPM
    $ pip install -r requirements.txt
    $ conda install pytorch==1.7 torchvision cudatoolkit=11.0 -c pytorch

Download repository:

    $ git clone git@github.com:ZhengChang467/STRPM.git

Test

For quick start, the pretrained models and the example data are needed to be downloaded to the following folders,

To test our model on the UCF Sport dataset, run the code in the PowerShell using:

    $ python STRPM_run.py --dataset ucfsport

The predictions from the UCF Sport dataset are shown as follows (ground truth (top) and predictions (bottom))

To test our model on the Human3.6M dataset, run the code in the PowerShell using:

    $ python STRPM_run.py --dataset human36m

To test our model on the SJTU4K dataset, run the code in the terminal using:

    $ python STRPM_run.py --dataset sjtu4k

The predicted results will be saved into path results/.

Train

We plan to share the train codes soon!

Citation

Please cite the following paper if you feel this repository useful.

@inproceedings{chang2022strpm,
  title={STRPM: A Spatiotemporal Residual Predictive Model for High-Resolution Video Prediction},
  author={Chang, Zheng and Zhang, Xinfeng and Wang, Shanshe and Ma, Siwei and Gao, Wen},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={13946--13955},
  year={2022}
}

License

See MIT License