Include50 islr dataset

Indian Sign Language Recognition (ISLR) 50-Class Dataset ATTRIBUTION AND ORIGINAL WORK NOTICEThis dataset is not our original creation. It is a derivative processed subset derived from the original INCLUDE and INCLUDE-50 datasets developed by Sridhar et al. We referenced the original dataset and followed the data extraction and processing pipeline to structure, filter, and preprocess the dataset for model training. Original Dataset Record:https://zenodo.org/records/4010759 Original Dataset Citation:Sridhar, A., Ganesan, R. G., Kumar, P., & Khapra, M. (2020). INCLUDE: A Large Scale Dataset for Indian Sign Language Recognition [Dataset]. Zenodo. ACM Multimedia 2020 (ACMMM2020). https://doi.org/10.1145/3394171.3413528 DATASET OVERVIEWThis repository contains the processed 50-class Indian Sign Language Recognition (ISLR) dataset. It includes raw sign language videos organized by semantic categories, dataset indexing files, class label mappings, and preprocessed MediaPipe Holistic landmark feature representations ready for training deep learning models. DIRECTORY AND FILE OVERVIEW 1. include_50_dataset/ (Directory)A hierarchical directory containing raw sign language video clips (.MOV and .MP4 formats) organized into 15 semantic category folders and sub-divided into class-specific folders.- Categories: Adjectives, Animals, Clothes, Colours, Days_and_Time, Electronics, Greetings, Home, Jobs, Means_of_Transportation, People, Places, Pronouns, Seasons, and Society.- Directory Structure: include_50_dataset/<Category_Folder>/<Class_Folder>/<Video_File> 2. include-50_dataset.zipA compressed ZIP archive containing the full include_50_dataset/ directory to facilitate full dataset downloading and extraction. 3. include-50_class_map.jsonA JSON file mapping each of the 50 distinct sign language vocabulary words to a unique integer class index from 0 to 49. 4. include-50_dataset.csv & include50_scie_final_index.csvMetadata index files for the dataset containing 943 video sample records. These files define dataset structure, file locations, and official benchmark train/val/test split allocations.- Total Samples: 943- Split Breakdown: • Train: 675 samples • Validation: 77 samples • Test: 191 samples 5. raw_train_test_val_processed_dataset.h5A preprocessed HDF5 data file containing normalized MediaPipe Holistic landmark tracking coordinates extracted frame-by-frame from the videos. Data is split into train, val, and test groups matching the official benchmark splits.- Feature Coordinates (X_raw): Formatted with shape (samples, 60, 543, 3) • 60: Fixed sequence length (padded via last-frame duplication) • 543: MediaPipe Holistic landmark points per frame (468 face, 33 pose, 21 left hand, 21 right hand landmarks) • 3: Spatial coordinates (x, y, z)- Data Tensor Dimensions across official splits: • Train split: X_raw shape (675, 60, 543, 3), y shape (675,) • Validation split: X_raw shape (77, 60, 543, 3), y shape (77,) • Test split: X_raw shape (191, 60, 543, 3), y shape (191,) Landmarks were extracted using the INCLUDE-50 dataset. Further details are available in the corresponding research work. INCLUDE 50 Research Work Citation @inproceedings{10.1145/3394171.3413528,author = {Sridhar, Advaith and Ganesan, Rohith Gandhi and Kumar, Pratyush and Khapra, Mitesh},title = {INCLUDE: A Large Scale Dataset for Indian Sign Language Recognition},year = {2020},isbn = {9781450379885},publisher = {Association for Computing Machinery},doi = {10.1145/3394171.3413528},numpages = {10},series = {MM '20}}