README.md

metadata

license: other
pretty_name: ASL Keypoint Dataset - 84 Classes
task_categories:
  - video-classification
  - time-series-classification
tags:
  - asl
  - sign-language
  - mediapipe
  - holistic
  - keypoints
  - isolated-sign-recognition

ASL Keypoint Dataset - 84 Classes

This dataset contains preprocessed MediaPipe Holistic keypoint sequences for isolated sign language classification.

Dataset Version

This version keeps only classes with at least 50 training samples. In your current filtered version, this produced 84 classes.

Files

• train_features_50plus.npy
• train_labels_50plus.npy
• val_features_50plus.npy
• val_labels_50plus.npy
• id_to_label_50plus.json
• label_to_id_50plus.json
• old_to_new_label_ids_50plus.json
• new_to_old_label_ids_50plus.json
• class_counts_after_filtering_50plus.csv
• filter_metadata_50plus.json

Important Note

The label IDs have been remapped after filtering. During inference, use id_to_label_50plus.json to convert model output IDs back to sign labels.

Metadata

{
  "hf_repo_id": "SharoonArshad/training_model23",
  "repo_type": "dataset",
  "min_samples": 50,
  "local_repo_path": "/kaggle/working/training_model23_raw",
  "original_train_samples": 64284,
  "filtered_train_samples": 6770,
  "original_val_samples": 8188,
  "filtered_val_samples": 862,
  "original_num_classes": 2310,
  "filtered_num_classes": 84,
  "train_features_shape": [
    6770,
    50,
    204
  ],
  "val_features_shape": [
    862,
    50,
    204
  ]
}