Deep Online Video Stabilization with Multi-Grid Warping Transformation Learning

https://ieeexplore.ieee.org/document/8554287

Prerequisites

• Linux
• Python 3
• NVIDIA GPU (12G or 24G memory) + CUDA cuDNN
• tensorflow-gpu==1.3.0
• numpy
• ...

Getting Started

Installation

Download data.zip at https://cg.cs.tsinghua.edu.cn/people/~miao/stabnet/data.zip.

This dataset does not contain flow information(set to 0). If you need to use data containing flow information, you can use the TVL1 algorithm to generate it.

unzip data.zip
mv data/models deep-online-video-stabilization/
mv data/datas deep-online-video-stabilization/
mv data/data deep-online-video-stabilization/
cd deep-online-video-stabilization-deploy
mkdir output

Testing

python3 -u deploy_bundle.py --model-dir ./models/v2_93/ --model-name model-80000 --before-ch 31 --deploy-vis --gpu_memory_fraction 0.9 --output-dir ./output/v2_93/Regular  --test-list /home/ubuntu/Regular/Regular/list.txt --prefix /home/ubuntu/Regular/Regular;

Training

python -u train_bundle_nobm.py

Dataset

DeepStab dataset (7.9GB) http://cg.cs.tsinghua.edu.cn/download/DeepStab.zip

Citation

If you find this useful for your research, please cite the following paper.

```
@ARTICLE{StabNet, 
    author={M. Wang and G. Yang and J. Lin and S. Zhang and A. Shamir and S. Lu and S. Hu}, 
    journal={IEEE Transactions on Image Processing}, 
    title={Deep Online Video Stabilization with Multi-Grid Warping Transformation Learning}, 
    year={2018}, 
    volume={}, 
    number={}, 
    pages={1-1}, 
}
```