Hugging Face's logo

Datasets:
luciayen
/
CASL-W60-Landmarks

Modalities:
Document
Text
Formats:
parquet
Size:
1K - 10K
Libraries:
Datasets
pandas
Polars
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
landmarks
array 2D
label
stringclasses
60 values
[[0.0,0.0,-0.669442355632782,0.02090632915496826,-0.0494236946105957,-0.6345288157463074,0.033336997(...TRUNCATED)
G003
[[0.0,0.0,-0.6411384344100952,0.01911473274230957,-0.050091564655303955,-0.6025876402854919,0.031721(...TRUNCATED)
G003
[[0.0,0.0,-0.8624114990234375,0.019685328006744385,-0.039743900299072266,-0.8074209094047546,0.03111(...TRUNCATED)
G003
[[0.0,0.0,-0.5750025510787964,0.018960177898406982,-0.050794631242752075,-0.5283839106559753,0.03319(...TRUNCATED)
G003
[[0.0,0.0,-0.6330848336219788,0.01305478811264038,-0.031209170818328857,-0.6032307744026184,0.022035(...TRUNCATED)
G003
[[0.0,0.0,-0.648048460483551,0.021915137767791748,-0.041568756103515625,-0.5999647378921509,0.034074(...TRUNCATED)
G003
[[0.0,0.0,-0.7439996004104614,0.017694920301437378,-0.03605341911315918,-0.7021201848983765,0.027968(...TRUNCATED)
G003
[[0.0,0.0,-0.4445513188838959,0.014541178941726685,-0.03871840238571167,-0.4127747416496277,0.025458(...TRUNCATED)
G003
[[0.0,0.0,-0.5293881297111511,0.014319539070129395,-0.04600101709365845,-0.4991215765476227,0.026933(...TRUNCATED)
G003
[[0.0,0.0,-0.5919119715690613,0.014676511287689209,-0.036103278398513794,-0.5527224540710449,0.02426(...TRUNCATED)
G003
End of preview. Expand in Data Studio

CASL-W60-64-frames Landmarks

license: mit task_categories: - video-classification language: - en tags: - sign-language - mediapipe - skeleton - Central-African-Sign-Language pretty_name: CASL-W60 Landmarks size_categories: - 1K<n<10K

Dataset Card for CASL-W60 Landmarks

This dataset contains preprocessed holistic landmarks for Central African Sign Language (CASL). It is specifically designed for training sequence-based models like Transformers or LSTMs for Sign Language Recognition (SLR).

Dataset Details

  • Total Classes: 60 (Signs)
  • Data Format: NumPy arrays (.npy) converted to Parquet
  • Sequence Length: 64 frames (fixed)
  • Features per Frame: 225 landmarks (Pose + Left Hand + Right Hand)

Dataset Structure

Each sample in the dataset contains:

  • landmarks: A 2D array of shape (64, 225).
  • label: The string ID of the sign (e.g., G003).
  • split: Whether the sample belongs to the train or test set.

Feature Mapping (The 225 Vector)

The 225-dimensional vector for each frame is structured as follows:

  1. Pose (Indices 0-98): 33 landmarks $\times$ 3 coordinates (X, Y, Z)
  2. Left Hand (Indices 99-161): 21 landmarks $\times$ 3 coordinates (X, Y, Z)
  3. Right Hand (Indices 162-224): 21 landmarks $\times$ 3 coordinates (X, Y, Z)

Usage

You can load this dataset directly using the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("luciayen/your-dataset-name")

# Access a sample
sample = dataset['train'][0]
print(sample['landmarks'])
print(sample['label'])
Downloads last month
14
Size of downloaded dataset files:
266 MB
Size of the auto-converted Parquet files:
266 MB
Number of rows:
5,889

Models trained or fine-tuned on luciayen/CASL-W60-Landmarks