Skip to content

RECORDING-DATASET.md

Latest commit

 

History

History
33 lines (22 loc) · 2.03 KB

RECORDING-DATASET.md

File metadata and controls

33 lines (22 loc) · 2.03 KB

Dataset Recording Instructions for Singa (Sign Language Recognition)

This document provides detailed instructions on how to record your own dataset using the slr_collecting_data.ipynb Jupyter notebook. This is essential for creating and customizing a dataset for training the sign language recognition model in the Singa project.

How to Record Your Own Dataset

  1. Running the Notebook:

    • Open and run the Jupyter notebook file slr_collecting_data.ipynb.

  2. Adjusting Parameters:

    • Videos per Label: You can change the number of videos per label/action by adjusting the videos_per_label variable. The default value is 60.
    • Frames per Video: You can change the number of frames per video by adjusting the frames_per_video variable. The default value is 60.

  3. Using Mediapipe TaskVision:

Recording Session Details

  • Video Capture:

    • The script will capture the specified number of videos (60 by default) per label action.
    • Each video will consist of 60 frames by default.
    • There will be a pause of approximately 1.5 seconds between each video capture to allow the actor to adjust for the next video.

  • Debugging:

    • For debugging purposes, you can set debug_landmark=True to visualize the keypoints detected by Mediapipe TaskVision. This helps the actor verify if their sign actions are being detected properly.

  • Exiting the Recording Session:

    • To quit the recording session, press q on the keyboard.

By following these instructions, you can efficiently record and customize your dataset for training the sign language recognition model. Ensure that the parameters are set according to your requirements and refer to the Mediapipe documentation for additional customization options.