Singa (Sign Language Recognition)

Singa is a project focused on sign language recognition using deep learning techniques. This README provides instructions on setting up the virtual environment, downloading datasets, running TensorBoard, and recording your own dataset.

Setting up the Virtual Environment

1. Initialize the Virtual Environment:

   • Run the setup script to create and configure the virtual environment.

   ./script/setup

2. Activate the Virtual Environment:

   • Use the activation script to start the virtual environment.

   ./script/activate

   • After activating the virtual environment for the first time, install the project requirements:

   pip install -r requirements.txt
   

Downloading the Dataset

• Execute the dataset script to download the required dataset from our dataset in Kaggle.

  ./script/dataset

• Additionally, we have another dataset available, which is not processed into one file dataset-cleaned.

  Additional Dataset Information

  • Both datasets consist of 20 labels.
  • Each label has a total of 120 videos, resulting in 2400 videos in total.
  • Since both datasets have already been augmented (flipped only), the total number of .npy files is 4800

  If you need access to the raw (video) dataset, you can request it via my email: [email protected]

Running TensorBoard

• Use the following command to start TensorBoard and visualize the training logs.

  tensorboard --logdir=./logs/train_log_directory

  • Replace train_log_directory with the actual name of the directory where your log files are stored.

Recording your own dataset

• If you want to add your own dataset, refer to the detailed instructions in RECORDING-DATASET.md.

Additional Notes:

• Ensure you have the Kaggle API installed. The Kaggle API should already be installed via pip:

  pip install kaggle

• The scripts assume that you have the necessary permissions to execute them. If you encounter permission issues, you may need to adjust the permissions using:

  chmod +x ./script/script_name

• The logs directory should contain the TensorFlow event files generated during model training, which TensorBoard uses for visualization.

By following these instructions, you can set up your environment, download the necessary datasets, monitor your model training, and record custom datasets for improving the sign language recognition model. For further details on dataset recording, refer to the RECORD_DATASET.md file.