Vid-ODE - Official PyTorch Implementation

This repository provides the official PyTorch implementation of the following paper:

Vid-ODE: Continuous-Time Video Generation with Neural Ordinary Differential Equation
Sunghyun Park*¹, Kangyeol Kim*¹, Junsoo Lee¹, Jaegul Choo¹, Joonseok Lee², Sookyung Kim³, Edword Choi¹
¹KAIST, ²Google Research, ³Lawrence Livermore Nat’l Lab.
In AAAI 2021. (* indicates equal contribution)
Paper : https://arxiv.org/abs/2010.08188
Project : https://psh01087.github.io/Vid-ODE/

Abstract: Video generation models often operate under the assumption of fixed frame rates, which leads to suboptimal performance when it comes to handling flexible frame rates (e.g., increasing the frame rate of more dynamic portion of the video as well as handling missing video frames). To resolve the restricted nature of existing video generation models' ability to handle arbitrary timesteps, we propose continuous-time video generation by combining neural ODE (Vid-ODE) with pixel-level video processing techniques. Using ODE-ConvGRU as an encoder, a convolutional version of the recently proposed neural ODE, which enables us to learn continuous-time dynamics, Vid-ODE can learn the spatio-temporal dynamics of input videos of flexible frame rates. The decoder integrates the learned dynamics function to synthesize video frames at any given timesteps, where the pixel-level composition technique is used to maintain the sharpness of individual frames. With extensive experiments on four real-world video datasets, we verify that the proposed Vid-ODE outperforms state-of-the-art approaches under various video generation settings, both within the trained time range (interpolation) and beyond the range (extrapolation). To the best of our knowledge, Vid-ODE is the first work successfully performing continuous-time video generation using real-world videos.

Installation

Clone this repository:

git clone https://github.com/psh01087/Vid-ODE.git
cd Vid-ODE/

We support python3. Install the dependencies:

pip install -r requirements.txt

Downloading Datasets

After downloading the dataset and unzipping files, place them under the dataset folder in current directory. We preprocess all datasets for training our models.

Training

To train a model on specific dataset, run:

CUDA_VISIBLE_DEVICES=0 python main.py --phase train --dataset kth

All arguments used for this project are described in the function "get_opt()" in main.py. There are a lot of options to train our network on a wide range of datasets and also to evaluate various architectures for writing the paper. However, just for the purpose of executing the proposed method, the number of arguments that you need to change would be very limited. The following options will be what you need to concern:

--dataset : Specify the dataset to train, select among [kth, penn, mgif, hurricane].
--extrap : If you toggle this option, you can train the extrapolation model.
--irregular : If you toggle this option, you can train the model with irregularly sampled frames.

Evaluation

We evaluate our model using Structural Similarity (SSIM), Peak Signal-to-Noise Ratio (PSNR), and Learned Perceptual Image Patch Similarity (LPIPS). To evaluate a model on specific dataset, run:

CUDA_VISIBLE_DEVICES=0 python main.py --phase test_met --dataset kth --test_dir CHECKPOINT_DIR

Citation

If you find this work useful for your research, please cite our paper:

@article{park2020vid,
  title={Vid-ODE: Continuous-Time Video Generation with Neural Ordinary Differential Equation},
  author={Park, Sunghyun and Kim, Kangyeol and Lee, Junsoo and Choo, Jaegul and Lee, Joonseok and Kim, Sookyung and Choi, Edward},
  journal={arXiv preprint arXiv:2010.08188},
  booktitle={The Thirty-Fifth {AAAI} Conference on Artificial Intelligence, {AAAI} 2021},
  pages={online},
  publisher={{AAAI} Press},
  year={2021},
}