ssl-dataset 1.0.0

Python library for the Sinhala Sign Language (SSL) dataset — landmarks, skeleton images, and preprocessed arrays for 55 sign classes.

ssl-dataset

Python library for the Sinhala Sign Language (SSL) dataset — 55 sign classes, 4,060 samples, three ready-to-use sub-libraries.

Installation

pip install ssl-dataset

Dataset overview

Type	Classes	Samples per class	Total samples
Static signs (images)	0 – 36	100	3,700
Dynamic signs (videos)	37 – 54	20	360
Total	55		4,060

---

Sub-libraries

ssl_dataset.landmarks

Raw 3D hand landmark coordinates extracted by MediaPipe.

from ssl_dataset.landmarks import SSLLandmarkDataset

# LSTM / Transformer input — shape (N, 30, 63)
ds = SSLLandmarkDataset(split="train", format="lstm")
X, y = ds.load()
print(X.shape)   # (2842, 30, 63)
print(y.shape)   # (2842,)

# Raw structured format — shape (N, 30, 21, 3)
ds = SSLLandmarkDataset(split="train", format="raw")
X, y = ds.load()
print(X.shape)   # (2842, 30, 21, 3)

---

ssl_dataset.skeleton

28×28 RGB hand skeleton images generated from MediaPipe landmarks.

from ssl_dataset.skeleton import SSLSkeletonDataset

# CNN-LSTM input — shape (N, 30, 28, 28, 3)
ds = SSLSkeletonDataset(split="train", format="cnn_lstm")
X, y = ds.load()
print(X.shape)   # (2842, 30, 28, 28, 3)

# MLP input — shape (N, 70560)
ds = SSLSkeletonDataset(split="test", format="mlp")
X, y = ds.load()
print(X.shape)   # (609, 70560)

---

ssl_dataset.preprocessed

Pre-split, pre-labelled numpy arrays — load and train immediately.

from ssl_dataset.preprocessed import SSLPreprocessedDataset

X_train, y_train = SSLPreprocessedDataset("train").load()
X_val,   y_val   = SSLPreprocessedDataset("val").load()
X_test,  y_test  = SSLPreprocessedDataset("test").load()

print(X_train.shape)   # (2842, 30, 63)
print(y_train.shape)   # (2842, 55)  — one-hot encoded

# input_shape helper for model building
ds = SSLPreprocessedDataset("train")
print(ds.input_shape)  # (30, 63)

---

Split details

All sub-libraries use the same stratified split (seed = 42) matching the original thesis methodology:

Split	Samples	Percentage
Train	2,842	70%
Val	609	15%
Test	609	15%

---

Class labels

from ssl_dataset import CLASS_LABELS

print(CLASS_LABELS[0])   # 'අ'
print(CLASS_LABELS[37])  # 'ඈ'
print(CLASS_LABELS[54])  # 'මගේ'

---

Citation

If you use this dataset in your research, please cite:

Jayasha (2025). Sinhala Sign Language Dataset.
GitHub: https://github.com/jayashalakshani/ssl-dataset

License

This dataset and library are licensed under CC BY-NC 4.0 (Creative Commons Attribution Non-Commercial 4.0).

You are free to use, share, and adapt this dataset for research and educational purposes, as long as you give appropriate credit. Commercial use is not permitted.