Data Augmentation Mosaic Albumentation (YOLO format)

This repository contributes the code to construct, visualize and save augmented images, namely Mosaic, CutOut, Random Crop (keep bbox safe), Shearing, Rotation, etc. by using Albumentation and Mosaic implementation

Prerequisite

Please change the path in the 1st cell of each notebook to path in your local computer

Folder notebooks contains

• Mosaic_Augmentation.ipynb: construct, visualize and save mosaic images base on dataset folder
• Albumentation_Visualization.ipynb: Apply, visualize multiple augmentation techniques, such as Random Crop (keep bbox safe), Cut Out, Channel Dropping, Rotation, Shearing, Guassian Noise,...
• Kmeans_Cluster_Images.ipynb: Using Kmeans to cluster images based on images' features extracted from VGG19 model

Example

The library that we're using albumentation and Mosaic from Yolov5

Mosaic augmentation with YOLO label format

Channel Dropout augmentation with YOLO label format

Cut Out augmentation with YOLO label format

Random crop but keep bbox safe augmentation with YOLO label format

Reference

Albumentation

Yolov5