MNIST Object Detection dataset

A simple to generate a dataset for object detection on MNIST numbers. By default, the script generates a dataset with the following attributes:

• 10,000 images in train. 10,000 images in validation/test
• 10 Classes
• Maximum size of each digit: 100 pixels
• Minimum size of each digit: 15 pixels
• Between 1 and 20 digits per image.

Installation

Requires python>= 3.6

pip install -r requirements.txt

Generate dataset

python3 generate_dataset.py

You can also change a bunch of settings by writing:

python3 generate_dataset.py -h

Visualize dataset

python3 visualize_dataset.py data/mnist_detection/train/
.py