LALip: Large Language Model Aided Lip-Reading

Quickstart

1. Download preprocessed dataset

2. Extract data.zip in the root project directory

3. Run train.py from the root project directory

Troubleshooting

• Ensure that train.py data_path arg value points to the correct directory

Notes

• The preprocessed dataset contains some invalid data points. These occurences are handled in the dataloader by ignoring these datapoints. If a batch size of 64 is set, and some batches contain less than 64, this is why.