Data Augmentor

A python data augmentation tool for deep learning

MIT License Copyright (c) 2018 Distributed Robotic Exploration and Mapping Systems Laboratory, ASU

Zhiang Chen, Nov 2018

1. xml2mask.py

xml2mask.py provides methods of generating masks from xml files downloaded from web-based annotation tool LabelMe. It can

• generate and combine all masks on a single layer and save as .jpg files. e.g. mask.shape = [width, height]
• generate all masks on individual layers and save as .npy files. e.g. mask.shape = [nm_objects, width, height]

2. rename.py

rename.py provides methods of renaming images and the corresponding annotations by numerical order. It supports to

• rename images and annotations by numerical order
• add annotations with prefix "label_"
• convert with .jpg and .png
• work with .npy

3. augmentation.py

augmentation.py augments images and corresponding annotations with same rules, which include

• resizing
• left-right flipping
• up-down flipping
• rotating
• zooming-in and zooming-out

4. generateCSV.py

generateCSV.py generates CSV files for RetinaNet https://github.com/DREAMS-lab/keras-retinanet. See the test main for how to use it: https://github.com/DREAMS-lab/data_augmentor/blob/master/generateCSV.py#L67.

5. processTiff.py

processTiff.py provides methods of processing GeoTiff. It can

• split large GeoTiff to small GeoTiff
• convert GeoTiff to PNG

6. labelboxReader.py

labelReader.py provides methods of generating masks from json files downloaded from web-based annotation tool LabelBox.

This is dependent on GDAL python. To install the related packages:

$ sudo add-apt-repository ppa:ubuntugis/ppa && sudo apt-get update
$ sudo apt-get install gdal-bin