This respository is implementation of the proposed method in LRW-1000: A Naturally-Distributed Large-Scale Benchmark for Lip Reading in the Wild. Our paper can be found here.

• python 3.6.7
• pytorch 1.0.0.dev20181103
• scipy 1.1.0

This model is pretrained on LRW with RGB lip images(112×112), and then tranfer to LRW-1000 with the same size. We train the model end-to-end.

You can train the model as follow:

python main.py --data_root "data path" --index_root "index root"

Where the data_root and index_root specifys the "LRW-1000 data path" and "label path" correspondly.
All the parameters we use is set as default value in args.py.You can also pass parameters through console just like:

python main.py --gpus 0,1 --batch_size XXX --lr 1e-4 --data_root "data path" --index_root "index root" ...

Note:
Please pay attention that you may need modify the code in dataset.py and change the parameters data_root and index_root to make the scripts work just as expected.

If this repository was useful for your research, please cite our work:

@article{shuang18LRW1000,
  title={LRW-1000: A Naturally-Distributed Large-Scale Benchmark for Lip Reading in the Wild},
  author={Shuang Yang, Yuanhang Zhang, Dalu Feng, Mingmin Yang, Chenhao Wang, Jingyun Xiao, Keyu Long, Shiguang Shan, Xilin Chen},
  booktitle={arXiv},
  year={2018}
}