Video Extrapolation in Space and Time

Official PyTorch Implementation of paper "Video Extrapolation in Space and Time", ECCV 2022.

Yunzhi Zhang, Jiajun Wu

Stanford University

[arXiv] [Page]

Installation

First, install pytorch and torchvision following the official doc. Then run

pip install -r requirements.txt

Dataset

Follow the following links to download the original dataset:

1. KITTI;
2. RealEstate10K;
3. CATER;
4. Lin et al.;
5. Cloud.

Training

To train the model, run the following command:

python train.py configs/kitti/base.yaml -d kitti_city --single_gpu

Evaluation

To compute quantitative results, run the following command:

python compute_scores.py PATH_TO_CHECKPOINT

To generate qualitative results, run

python vest/inference/show_estate.py PATH_TO_CHECKPOINT

Acknowledgements

This codebase is built on top of Imaginare.