Indian Sign Language Dataset

Welcome to the Custom Indian Sign Language Dataset! This dataset has been meticulously curated with the aim of enhancing the accuracy and robustness of ISL gesture recognition systems. By sharing this dataset, we aspire to contribute to the research community and empower fellow researchers and data scientists to explore and innovate in the domain of sign language recognition.

Dataset Overview: This dataset comprises a collection of ISL gesture images, meticulously captured and processed to simulate real-world scenarios. The original images feature a diverse range of gestures performed by myself and a few friends, representing a broad spectrum of sign language expressions.

Controlled Noise Addition: To introduce realism and variability akin to real-world communication scenarios, each image underwent controlled noise addition. This process included the incorporation of various background types such as blurry, messy, and colorful, aiming to emulate dynamic environmental conditions. By intentionally introducing controlled noise, we aimed to equip the dataset with the ability to train models that can effectively handle the challenges posed by different backgrounds and lighting conditions.

Image Specifications: All images in this dataset have been standardized to a size of 126x126 pixels. This uniformity ensures consistency in data presentation and facilitates ease of preprocessing and model development. The images are stored in a format that ensures seamless integration into various machine learning frameworks.

We encourage you to explore, analyze, and leverage this dataset for your research, experimentation, and development endeavors. By utilizing this dataset, you'll gain access to a unique collection of ISL gesture images that have been thoughtfully curated to enhance the accuracy and adaptability of sign language recognition models.

License: Please note that this dataset is provided under the [License Type], which outlines the terms and conditions for its use and distribution. Kindly ensure that you review and adhere to the license specifications when utilizing the dataset.

We invite you to join us in advancing the field of sign language recognition by leveraging this custom dataset as a valuable resource for your research and innovation. Together, let's work towards creating more inclusive and accessible technology solutions for the deaf community.

ISL Gestures Landmarks Dataset

The dataset contains ISL Gestures for Alphabet classes A-Z, each containing ~2k records. The total entries make up to around 50k records.

Purpose of this dataset

So you have searched all across the web for a good ISL dataset, but didn't find a decent one yet? Did your models also overfit while working on those datasets? well, not anymore! I'm introducing my own dataset.

A Quick History

Other available ISL datasets are in the form of images, not providing enough diversity to even produce a decent model. All the time - at least in my case - models got overfit. Therefore, I've created my own dataset. Tried and tested, and also being my first dataset, it produces a good model!

Features of the Dataset

1. Hand Landmarks are captured rather than images

2. Each class (A-Z) contains ~2,000 records of hand landmarks captured at different angles, rotations & perspectives while maintaining the front view.

3. Additionally, Interchanged hands in single-hand gestures (e.g., C, V, I, L, O, etc.) as well as non-symmetrical gestures (e.g., D, E, F, H, J, etc.) to maintain robustness and add variability

4. The data, after capturing got cleaned enough to provide you the base allowing possibility of further modifications if you want.

5. Feature engineered and produced one column - uses_two_hands - boosting R^2 score to >99% !

The dataset contains word-level Indian Sign Language videos in .mp4 format. These signs were sourced from the official Indian Sign Language YouTube channel, maintained by the government body. Four contributors created 60 unique signs, each with 60 corresponding videos. Additionally, 30 videos of still signs were included to improve model accuracy and avoid misinterpretation.

Paper link - https://link.springer.com/chapter/10.1007/978-981-97-6992-6_19

Indian Sign Language Alphabet Dataset: Overview

About the Dataset

The Indian Sign Language Alphabet Dataset is a curated collection of hand gesture images representing the alphabets (A–Z) in Indian Sign Language (ISL). It was specifically created to support the development of Gesture-Speech, a groundbreaking project that translates Indian Sign Language gestures into spoken words, making communication more accessible for the hearing and speech-impaired communities.

The dataset is intended for researchers, developers, and educators to build robust machine learning models for recognizing ISL gestures and advancing assistive technologies.

Project Inspiration: Gesture-Speech

The dataset was developed as part of the Gesture-Speech project, an open-source initiative hosted on GitHub. Gesture-Speech focuses on real-time translation of Indian Sign Language gestures into spoken language using deep learning and computer vision. The project’s mission is to: - Empower differently-abled individuals by breaking communication barriers. - Promote inclusivity through assistive AI-driven technologies. - Enable the use of ISL in everyday interactions.

The dataset serves as the backbone of Gesture-Speech’s training pipeline, allowing the development of accurate and efficient recognition models.

Key Features

1. Dataset Highlights:

   • Classes: 26 classes for each English alphabet (A-Z), represented by corresponding ISL gestures.
   • Images: 1,200 images per class, totaling 31,200 images.
   • Diversity: Includes hand gestures from individuals with varying hand shapes, sizes, and skin tones, ensuring model robustness.
   • Real-world Conditions: Captured in different lighting and background conditions to simulate practical usage scenarios.

2. Integration with Gesture-Speech:

   • Fully compatible with TensorFlow and PyTorch frameworks used in the Gesture-Speech project.
   • Pre-processed and labeled for easy model training and testing.

Dataset Structure

The dataset follows a directory-based structure for classification tasks:

Indian_Sign_Language_Alphabet/
│── a/
│  ├── 0.jpg
│  ├── 1.jpg
│  ├── ...
│── b/
│  ├── 0.jpg
│  ├── 1.jpg
│  ├── ...
│── ...
│── z/

Applications

• Gesture Recognition Models: Train deep learning models (e.g., CNNs) to recognize ISL alphabets.
• Assistive Technology: Power systems like Gesture-Speech to convert gestures into speech for improved accessibility.
• Education Tools: Develop interactive platforms to teach and learn Indian Sign Language.
• Human-Computer Interaction: Enable gesture-controlled applications for robotics and automation.

Technical Details

1. Resolution: Images are resized to 384x384 pixels for consistency.
2. File Format: Images are stored in .jpg format.
3. Dataset Size: 334MB.

Gesture-Speech: GitHub Project

The Gesture-Speech project can be found on GitHub. Key highlights of the project: - Uses TensorFlow and OpenCV to process hand gestures in real-time. - Translates ISL gestures into spoken English or Hindi phrases using a deep learning pipeline. - Open-source and extensible, encouraging contributions and further development.

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. Users can: - Share: Copy and redistribute the material in any medium or format. - Adapt: Remix, transform, and build upon the material for any purpose, even commercially. - Attribution: Proper credit must be given to the dataset creator.

Acknowledgments

The dataset and the Gesture-Speech project are dedicated to promoting inclusivity and enabling new opportunities for the differently-abled. Special thanks to contributors and volunteers who helped create this dataset by providing diverse samples of ISL gestures.

Dataset Details: The INCLUDE dataset has 4292 videos (the paper mentions 4287 videos but 5 videos were added later). The videos used for training are mentioned in train.csv (3475), while that used for testing is mentioned in test.csv (817 files). Each video is a recording of 1 ISL sign, signed by deaf students from St. Louis School for the Deaf, Adyar, Chennai. INCLUDE50 has 766 train videos and 192 test videos. Train-Test Split: Please download the train-test split for INCLUDE and INCLUDE50 from here: Train-Test Split Publication Link: https://dl.acm.org/doi/10.1145/3394171.3413528 AI4Bharat website: https://sign-language.ai4bharat.org/ Download Instructions For ease of access, we have prepared a Shell Script to download all the parts of the dataset and extract them to form the complete INCLUDE dataset. You can find the script here: http://bit.ly/include_dl Paper Abstract: Indian Sign Language (ISL) is a complete language with its own grammar, syntax, vocabulary and several unique linguistic attributes. It is used by over 5 million deaf people in India. Currently, there is no publicly available dataset on ISL to evaluate Sign Language Recognition (SLR) approaches. In this work, we present the Indian Lexicon Sign Language Dataset - INCLUDE - an ISL dataset that contains 0.27 million frames across 4,287 videos over 263 word signs from 15 different word categories. INCLUDE is recorded with the help of experienced signers to provide close resemblance to natural conditions. A subset of 50 word signs is chosen across word categories to define INCLUDE-50 for rapid evaluation of SLR methods with hyperparameter tuning. The best performing model achieves an accuracy of 94.5% on the INCLUDE-50 dataset and 85.6% on the INCLUDE dataset

Please use the train-test split mentioned in the description

Here is a professionally formatted README.md file designed for your Hugging Face dataset. This file is optimized to help you rank on search engines while clearly linking back to your Kaggle repository.

  Indian Sign Language: Hindi (ISL-Hindi)





  India’s First Comprehensive Image Dataset for Hindi Character Recognition






  Overview

This dataset is the first of its kind, specifically curated to address the lack of resources for regional Indian languages in the field of… See the full description on the dataset page: https://huggingface.co/datasets/KRISH09bha/Hindi-Indian-Sign-language-dataset-ISL.

This dataset consists of the Indian sign language of all the alphabets and numbers in Indian hand recognition given by ISRTC(Indian Sign Research and Training Center). This dataset is in black white background for faster computing and for getting better accuracy while training the dataset.

Please give credit to this dataset if you download it.

This dataset contains depth data collected through Intel RealSense Depth Camera D435i. Data corresponding to Indian Sign Language (ISL) gesture of Weekdays (Sunday-Saturday) is used.

Data is stored as comma separated values. Each line corresponds to a sign.

Indian Sign Language Detection

## Overview

Indian Sign Language Detection is a dataset for object detection tasks - it contains A Z 0 9 annotations for 1,748 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

LIGHTscrn/Indian-Sign-language-landmarks-30frames dataset hosted on Hugging Face and contributed by the HF Datasets community

This dataset contains MP4 video clips of Indian Sign Language (ISL) gestures, intended for use in machine learning, gesture recognition, and accessibility-focused projects. It includes over 3000 short videos featuring alphabets (A–Z), numbers (0–9), and common words like “Hello” and “Thank You.” To meet upload limits, the dataset is split across multiple ZIP files, each with around 1000 videos. All files are in .mp4 format, with consistent resolution and duration. This dataset is useful for building ISL recognition models, real-time sign detection, and inclusive communication tools. Unzip all parts to access the full set.

y

Indian Sign Language_2024

## Overview

Indian Sign Language_2024 is a dataset for object detection tasks - it contains A annotations for 1,010 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Introduction to the Indian Sign Language (ISL) Model

This project focuses on creating a precise and real-time detection model for Indian Sign Language (ISL) gestures. We utilize the YOLO-NAS Object Detection model with the coco/14 checkpoint and leverage Roboflow for data preprocessing, which includes annotating various images and balancing classes. The training, validation, and testing phases are executed using a Google Colab notebook, followed by live detection implementation in PyCharm. This model aims to facilitate seamless communication for the deaf and hard-of-hearing community by translating ISL gestures into text and speech.

This dataset contains 36 classes representing Indian Sign Language (ISL) characters, including digits (0–9) and alphabets (A–Z). Each class has 1,000 images, resulting in a total of 36,000 labeled samples.

The dataset is designed to support research and development in:

Computer Vision: Hand gesture recognition

Deep Learning: Image classification and CNN-based models

Human-Computer Interaction: Enabling communication tools for the deaf and hard-of-hearing community

Sign Language Translation Systems

Dataset Details:

Classes: 36 (0–9, A–Z)

Images per class: 1,000(some might need pre-processing)

Format: JPG

Use cases: Training, validation, and testing of sign language recognition models

This dataset can be used to build, train, and benchmark machine learning models for gesture recognition tasks. It contributes to bridging the communication gap by empowering developers and researchers to create real-world applications such as sign language interpreters, accessibility tools, and educational platforms.

Indian Words Sign Language

## Overview

Indian Words Sign Language is a dataset for object detection tasks - it contains Indian Words Sign Language T57q annotations for 1,095 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Indian Sign Language_40

## Overview

Indian Sign Language_40 is a dataset for object detection tasks - it contains A 8AcR A 8hEz annotations for 1,405 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

This dataset contains Indian Sign Language (ISL) hand gesture images for alphabets A to Z. Each class is stored in separate folders and includes multiple image samples captured under consistent lighting conditions. The dataset is suitable for training machine learning and deep learning models such as CNN, LSTM, and hybrid architectures for sign language recognition.

This video and gloss-based dataset has been meticulously crafted to enhance the precision and resilience of ISL (Indian Sign Language) gesture recognition and generation systems. Our goal in sharing this dataset is to contribute to the research community, providing a valuable resource for fellow researchers to explore and innovate in the realm of sign language recognition and generation.Overview of the Dataset: Comprising a diverse array of ISL gesture videos and gloss datasets. The term "gloss" in this context often refers to a written or spoken description of the meaning of a sign, allowing for the representation of sign language in a written form. The dataset includes information about the corresponding spoken or written language and the gloss for each sign. Key components of a sign language gloss dataset include ISL grammar that follows a layered approach, incorporating specific spatial indices for tense and a lexicon with compounds. It follows a unique word order based on noun, verb, object, adjective, or part of a question. Marathi sign language follows the subject-object-verb (SOV) form, facilitating comprehension and adaptation to regional languages. This Marathi sign language gloss aims to become a medium for everyday communication among deaf individuals. This dataset reflects a careful curation process, simulating real-world scenarios. The original videos showcase a variety of gestures performed by a professional signer capturing a broad spectrum of sign language expressions. Incorporating Realism with green screen with controlled lighting conditions. All videos within this dataset adhere to pixels, ensuring uniformity for data presentation and facilitating streamlined pre-processing and model development stored in a format compatible with various machine and Deep learning frameworks, these videos seamlessly integrate into the research pipeline

Indian Sign Language Fingerspelling Landmark Dataset

A self-collected dataset of 21 3D hand landmark coordinates for static Indian Sign Language (ISL) fingerspelling recognition, extracted using MediaPipe Hands.

Overview

This dataset contains wrist-relative MediaPipe hand landmark features for 23 static ISL fingerspelling signs. It was collected to support real-time ISL recognition on consumer webcam hardware without GPU acceleration.

Letters H, J, and Y are excluded as they involve hand movement (dynamic signs) requiring sequence modelling.

Signs Covered

A B C D E F G I K L M N O P Q R S T U V W X Z (23 static signs)

Dataset Structure

Each row represents one captured frame with both hands detected.

Columns: 127 total - left_lm0_x to left_lm20_z — 63 features for left hand
(x, y, z coordinates of 21 MediaPipe landmarks, wrist-relative) - right_lm0_x to right_lm20_z — 63 features for right hand
(x, y, z coordinates of 21 MediaPipe landmarks, wrist-relative) - label — ISL sign letter (A–Z excluding H, J, Y)

Feature encoding: - All coordinates are wrist-relative (landmark 0 subtracted from all) - Missing hand = 63 zeros (for one-handed signs) - Values rounded to 6 decimal places

Files

Collection Details

• Contributors: 6 training contributors + 1 held-out test contributor
• Samples: 300 samples per sign for base contributors, 200 additional samples per weak sign from targeted contributors
• Hardware: Standard laptop webcams (no special equipment)
• Conditions: Varied indoor lighting (normal, bright, dim)
• Hand tracking: MediaPipe Hands 0.10.9, max_num_hands=2
• Normalisation: Wrist-relative (landmark 0 as origin)

Intended Use

• ISL sign language recognition research
• Gesture classification benchmarking
• Transfer learning for low-resource sign language datasets

Collection Script

The data collection script is available at: ISL-FingerSpell GitHub Repository

Citation

If you use this dataset, please cite: [Deka et al.], "ISL-FingerSpell: Real-Time Indian Sign Language Fingerspelling Recognition Using MediaPipe Hand Landmarks", 2026.

Acknowledgements

Hand landmark extraction powered by MediaPipe Hands: Zhang et al., "MediaPipe Hands: On-device Real-time Hand Tracking", arXiv:2006.10214, 2020.