SharoonArshad/asl-citizen-processed-200

ASL Citizen Processed Keypoints-200

This dataset was processed from the ASL Citizen keypoints-200 Kaggle dataset.

Purpose

This processed version is prepared for isolated American Sign Language recognition and encoder pretraining.

It is intended to be used as the first stage of a larger ASL-to-English pipeline:

1. Train isolated sign encoder on ASL Citizen.
2. Transfer the encoder into a CTC model.
3. Fine-tune on a continuous ASL dataset such as How2Sign.
4. Decode continuous signing into gloss/text.

Feature Format

Original PKL samples:

(T, 75, 4)

Where:

• T = number of frames
• 75 = selected landmarks/keypoints
• 4 = values per keypoint

Processed NumPy samples:

(num_samples, sequence_length, feature_dim)

Included Files

• train_features.npy
• train_labels.npy
• val_features.npy
• val_labels.npy
• test_features.npy
• test_labels.npy
• label_map.json
• id_to_label.json
• class_mapping.json
• metadata.json

Labels

Labels are stored as integer class IDs.

Use id_to_label.json to convert predicted class IDs back into readable ASL sign labels.

Loading Example

import numpy as np
import json

train_features = np.load("train_features.npy")
train_labels = np.load("train_labels.npy")

with open("id_to_label.json", "r", encoding="utf-8") as f:
    id_to_label = json.load(f)

print("Train features:", train_features.shape)
print("Train labels:", train_labels.shape)
print("First label:", id_to_label[str(train_labels[0])])

Notes

This dataset contains keypoint-based features, not raw videos.

The goal is to reduce training cost and make the dataset easier to use for sign-language recognition models.

Citation

This dataset is derived from ASL Citizen-related data. Please cite the original ASL Citizen dataset/source when using this processed version.