Latvian Sign Language Landmark Corpus

  Autori

Neimane, Liene Krista

  Vienuma identifikators

http://hdl.handle.net/20.500.12574/139

 Projekta URL

https://github.com/LucyferYX/LZV-CSLR

 Izdošanas datums

2025

 Tips

corpus, text

 Lielums

332 items, 1 file

 Valoda(-s)

Latvian Sign Language

 Apraksts

The corpus contains MediaPipe-extracted landmark data representing 45 Latvian Sign Language signs. It includes 33 alphabet letters (a-z), 11 numbers (0-10), and a pause, which were captured from videos featuring several different signers. The collection covers both isolated signs and sign combinations forming complete words or short sentences. For each video frame, 543 landmarks were obtained using MediaPipe Holistic: pose (0-32), face (33-500), left hand (501-521), and right hand (522-542). The dataset is provided as .npy files containing the landmark data suitable for machine learning. A supplementary Jupyter notebook is also provided, showcasing example on how the landmarks can be loaded and visualized as a 2D human skeleton.

 Izdevējs

Liene Krista Neimane

 Tēma(-s)

Latvian Sign Language MediaPipe landmarks

 Kolekcija (s)

Language resources and tools of AiLab IMCS UL

Rādīt pilnu ierakstu