Dataset Card for INCLUDE

Dataset Summary

This dataset contains all videos in the INCLUDE dataset. As huggingface does not support video uploads at this time, the HF dataset contains metadata about each video such as the parent class, the video class, the path to the video and whether its a part of the INCLUDE-50 dataset (use include_50==True to get only include_50 videos). The videos themselves can be downloaded from Zenodo using the provided bash script.

Video download instructions

1. Copy paste the following code into a file. Save the file as download_script.sh

#!/bin/bash

# The base URL for the API request
base_url="https://zenodo.org/api/records/4010759"

# Fetch the JSON metadata from Zenodo
response=$(curl -s "$base_url")

# Parse JSON to extract file URLs and names using jq
echo "$response" | jq -r '.files[] | .links.self + " " + .key' | while read -r file_url file_name
do
  # Use curl to download each file and save it with the respective name
  echo "Downloading $file_name from $file_url..."
  curl -o "$file_name" "$file_url"
  echo "$file_name downloaded."
done

echo "All files downloaded."

# Loop through all zip files in the current directory
for file in *.zip; do
    # Unzip each file into a directory with the same name as the zip file without the extension
    unzip "${file%.zip}"
done

echo "All files unzipped."

2. Make the above file executable by opening a terminal window and running chmod +x download_script.sh
3. Execute the above file in the directory you want to store the videos by running ./download_script.sh

Use the video_paths mentioned in the huggingface 'video_path' column to access the corresponding video.

Supported Tasks

This dataset was created for the goals of education and Isolated Sign Language Recognition, but may be used for other purposes.

Data Splits

The dataset is split into train and test splits as described in the paper

Dataset Creation

All details on dataset creation can be found in the INCLUDE paper.

Personal and Sensitive Information

These videos represent real people and their unique ways of communication. We urge all users of the dataset to respect their privacy - do not use these videos for any purposes that might infringe on the privacy or dignity of the individuals featured. Please ensure that the usage of these videos aligns with ethical standards and promotes understanding and inclusivity.

Discussion of Biases

India is a diverse country, and does not have one uniform sign language. The videos in this dataset were shot in Chennai, Tamil Nadu. However, they are not the only representation of "Indian Sign Language", as ISL varies from place to place across the country.

Citation Information

If you use this dataset, please cite the following work:

@inproceedings{sridhar_include:_2020, address = {New York, NY, USA}, series = {{MM} '20}, title = {{INCLUDE}: {A} {Large} {Scale} {Dataset} for {Indian} {Sign} {Language} {Recognition}}, isbn = {9781450379885}, shorttitle = {{INCLUDE}}, url = {https://doi.org/10.1145/3394171.3413528}, doi = {10.1145/3394171.3413528}, urldate = {2024-07-16}, booktitle = {Proceedings of the 28th {ACM} {International} {Conference} on {Multimedia}}, publisher = {Association for Computing Machinery}, author = {Sridhar, Advaith and Ganesan, Rohith Gandhi and Kumar, Pratyush and Khapra, Mitesh}, month = oct, year = {2020}, pages = {1366--1375}, }

Contributions

Thanks to @Rohith, @Gokul and @Advaith for adding this dataset. For any further questions, please reach out to advaithsridhar08@gmail.com