PyTorch implementation of our paper, Revisiting Hierarchical Approach for Persistent Long-Term Video Prediction, which will be presented in ICLR 2021. Please check out more qualitative results in our project page.
REPO_DIR='/path/to/clone/this/repo' git clone https://www.github.com/1Konny/hierarchicalvideoprediction $REPO_DIR cd $REPO_DIR bash scripts/dependency.sh cd $REPO_DIR/image_generator python scripts/download_models_flownet2.py python scripts/download_flownet2.py cd $REPO_DIRCheck it out in this link
CUDA_VISIBLE_DEVICES='0,1,2,3' bash scripts/train_structure_generator.sh $DATASET, where DATASET can be one of KITTI or Cityscapes.
CUDA_VISIBLE_DEVICES='0' bash scripts/test_structure_generator.sh $DATASETCUDA_VISIBLE_DEVICES='0,1,2,3' bash scripts/train_image_generator.sh $DATASETStep 6: Extract RGB-level predictions using the trained image generator and predictions from the structure generator.
CUDA_VISIBLE_DEVICES='0' bash scripts/test_image_generator.sh $DATASET@inproceedings{ lee2021revisiting, title={Revisiting Hierarchical Approach for Persistent Long-Term Video Prediction}, author={Wonkwang Lee and Whie Jung and Han Zhang and Ting Chen and Jing Yu Koh and Thomas Huang and Hyungsuk Yoon and Honglak Lee and Seunghoon Hong}, booktitle={International Conference on Learning Representations}, year={2021}, url={https://openreview.net/forum?id=3RLN4EPMdYd} }