![]() |
![]() |
![]() |
![]() |
This is a data generator for 2D object detection with axis-aligned rectangular bounding boxes. Its main focus is on flexible and easily extensible data augmentation.
Main features:
- Lets you easily define arbitrary chains of image transformations.
- Provides most common photometric and geometric image transformations out of the box.
- Provides both random and deterministic versions of all image transformations.
- Provides flexible options for how to decide whether a random transformation of an image is "valid" (see tutorial).
- Provides parsers for ground truth annotations in the MS COCO (JSON) and Pascal VOC (XML) formats as well as for a wide range of CSV formats.
- Python 3.x
- Numpy
- OpenCV
- Beautiful Soup 4.x (to parse XML files)
Python 2 compatibility: This implementation seems to work with Python 2.7, but I don't provide any support for it. It's 2018 and nobody should be using Python 2 anymore.
Everything you need to know in order to use this data generator is explained in the following Jupyter notebook: