Latvian Sign Language Landmark Corpus

  Authors

Neimane, Liene Krista

  Item identifier

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

 Project URL

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

 Date issued

2025

 Type

corpus, text

 Size

332 items, 1 file

 Language(s)

Latvian Sign Language

 Description

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.

 Publisher

Liene Krista Neimane

 Subject(s)

Latvian Sign Language MediaPipe landmarks

 Collection(s)

Language resources and tools of AiLab IMCS UL

Show full item record