Pascal VOC Image Augmentor

Script for image augmentation writing new Pascal VOC annotations, using the imgaug library.

Augment images present in dataset/images/ considering their Pascal VOC annotations in dataset/annotations/. This library is useful AFTER you've annotated your images with a tool such as labelImg.

ℹ️ This script was designed to consume a low amount of RAM. Images are processed one by one.

Usage

1. Place your images in dataset/images/ and their associated annotations in dataset/annotations/

   The image filenames must match their annotation filenames.

2. Configure the augmentation options as you wish in docker-compose.yml

3. Run the script

   docker-compose up augment

Visualize

You can run the following command then to visualize newly annotated images that were copied in dataset-augmented/

A red square will be drawn around the bounding boxes described in the annotations

Images will be saved into dataset-visualization/

docker-compose up visualize

Here is an example :

Top left you can see the original image, and then the augmented images. You can customize the augmentation parameters in augment.py at line 185.

References

1. Part of this project is an adaptation of the asetkn's tutorial on how to perform image augmentation with Pascal VOC annotated images