---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: label
    dtype:
      class_label:
        names:
        - A
        - B
        - C
        - D
        - E
        - F
        - G
        - H
        - I
        - K
        - L
        - M
        - 'N'
        - O
        - P
        - Q
        - R
        - S
        - T
        - U
        - V
        - W
        - X
        - 'Y'
  splits:
  - name: train
    num_bytes: 0
    num_examples: 0
  - name: test
    num_bytes: 0
    num_examples: 0
  download_size: 0
  dataset_size: 0
configs:
- config_name: default
  data_files:
  - split: train
    path: train/*
  - split: test
    path: test/*
license: mit
tags:
- computer-vision
- image-classification
- sign-language
- asl
- mnist
- gesture-recognition
task_categories:
- image-classification
task_ids:
- multi-class-image-classification
paperswithcode_id: asl-mnist
size_categories:
- 10K<n<100K
language:
- en
multilinguality:
- monolingual
source_datasets:
- original
annotations_creators:
- expert-generated
pretty_name: american-sign-language-mnist
---

# ASL-MNIST Dataset Card

![](ASL.png)

## Dataset Description

The American Sign Language MNIST dataset contains static hand gesture images representing letters of the ASL alphabet. This dataset provides image classification data for training models to recognize ASL hand poses.

## Dataset Details

- **Total samples**: Varies by split (training and test sets available)
- **Image format**: JPG
- **Number of classes**: 24 (out of 26 alphabet letters)
- **Image type**: Grayscale hand gesture photographs
- **Split**: Training and test sets provided

## Classes

The dataset includes 24 ASL letters:
A, B, C, D, E, F, G, H, I, K, L, M, N, O, P, Q, R, S, T, U, V, W, X, Y

### Missing Letters

J (index 9) and Z (index 25) are excluded because they require hand motion in ASL. Static images cannot capture these motion-based letters.

## Dataset Structure

```
asl-mnist/
├── train/
│   ├── train_0.jpg
│   ├── train_1.jpg
│   └── ...
├── test/
│   ├── test_0.jpg
│   ├── test_1.jpg
│   └── ...
├── train.csv
└── test.csv
```

## Label Mapping

The dataset uses numeric indices that map to letters:

```python
asl_labels = {
    0: 'A', 1: 'B', 2: 'C', 3: 'D', 4: 'E', 5: 'F', 6: 'G', 7: 'H', 8: 'I',
    10: 'K', 11: 'L', 12: 'M', 13: 'N', 14: 'O', 15: 'P', 16: 'Q', 17: 'R',
    18: 'S', 19: 'T', 20: 'U', 21: 'V', 22: 'W', 23: 'X', 24: 'Y'
}
```

## Data Fields

Each sample contains:
- `image`: The hand gesture image file
- `label`: Classification label (letter A-Y, excluding J and Z)
- `split`: "train" or "test"

## Usage

This dataset is suitable for:
- Image classification tasks
- Computer vision model training
- ASL recognition research
- Educational projects on sign language

## Limitations

- Contains only static poses (no motion-based letters)
- Limited to 24 out of 26 alphabet letters
- Single hand gestures only
- Grayscale images may limit color-based feature learning

## Citation

If you use this dataset in your research, please cite the original ASL-MNIST dataset, available through Kaggle 
https://www.kaggle.com/datasets/datamunge/sign-language-mnist/data 

## Technical Implementation

The dataset has been processed and formatted for use with FiftyOne, a computer vision dataset management tool. The implementation includes:
- Automatic label mapping from numeric indices to letters
- Train/test split preservation
- Metadata computation for efficient querying
- Support for filtering unknown labels