zentardev/handwritten-digit-dataset

Handwritten Digit Dataset

This dataset contains a collection of handwritten digits (0-9) contributed by users through an interactive web-based drawing application. The dataset is continuously updated, reflecting real-world human handwriting variability.

Dataset Details

The images are pre-processed to match the standard machine learning format for digit recognition:

• Dimensions: 28x28 pixels.
• Format: Grayscale (single channel).
• Processing: Each digit is cropped to its bounding box, centered, and scaled to fit the standard 20x20 model input space with padding.
• Source: Crowdsourced here.

Data Structure

The dataset is split into two subsets to facilitate robust model training and evaluation:

1. Training Set (train.csv)

Contains samples where the user successfully identified the digit, or the system confirmed the user's intent. These represent "clean" samples of various writing styles.

2. Test Set (test.csv)

Contains samples where the user failed the challenge or where the model's prediction did not match the user's intent. This set is invaluable for identifying model weaknesses, handling ambiguous handwriting, and evaluating performance on "edge cases."

Column	Description
image_base64	The 28x28 pixel image encoded in Base64.
prediction	The digit value (0-9) predicted by the model.
correct	Boolean indicating if the sample was a successful match.
true_label	The target of the digit predicted by the model.

Usage

You can easily load this dataset using the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("zentardev/handwritten-digit-dataset")

# Accessing training data
train_data = dataset["train"]

# Accessing test/evaluation data
test_data = dataset["test"]

Contribution & Automation

This dataset is live. It is automatically synced from the application's database to this repository using a GitHub Actions pipeline. If you would like to contribute, please visit our app and help us classify digits!

License

This project is licensed under the MIT License. Feel free to use this data for training, research, or personal projects.