This is a PyTorch implementation of PWStableNet: Learning Pixel-wise Warping Maps for Video Stabilization.
Source code and models will be opened soon!
If you have any questions, please contact with me: zmd1992@mail.ustc.edu.cn
The dataset for is the DeepStab dataset (7.9GB) http://cg.cs.tsinghua.edu.cn/download/DeepStab.zip thanks to Miao Wang [1].
The code will download the VGG-16 PyTorch base network weights at: https://download.pytorch.org/models/vgg16-397923af.pth automatically.
To train PWNet using the train script simply specify the parameters listed in ./lib/cfg.py
as a flag or manually change them.
The default parameters are set for the use of two NVIDIA 1080Ti graphic cards with 24G memory.
CUDA_VISIBLE_DEVICES=0,1 python3 main.py
we show an example videos to compare our PWNet with StabNet [1]
[1] M. Wang, G.-Y. Yang, J.-K. Lin, S.-H. Zhang, A. Shamir, S.-P. Lu, and S.-M. Hu, “Deep online video stabilization with multi-grid warp- ing transformation learning,” IEEE Transactions on Image Processing, vol. 28, no. 5, pp. 2283–2292, 2019