Image Data Augmentation Toolkit and Visualizer

This is a tool for visualizing and analyzing image data augmentation methods.

Back to the main page

About

Image Augmentation Toolkit and Visualizer is an open-source web application designed to visualise and analyse the effects of image data augmentation techniques using visual outputs and quantitative metrics. The toolkit was developed as a practical resource for researchers, practitioners and educators, and aims to bridge the gap between theoretical augmentation methods and their actual impact on image structures, perceptual quality and pixel-level characteristics.

Data augmentation is a crucial component of machine learning pipelines, improving model generalisation, robustness and fairness. However, the way in which different augmentation techniques manipulate image content, such as structural details, contrast, sharpness and perceptual similarity, is often understood intuitively rather than based on measurable evidence. This toolkit provides an interactive, standardised platform on which users can visually and numerically explore and compare augmentation methods.

Key Features

  • Visual demonstrations: Interactively view original and augmented images to see how each transformation modifies the image.
  • Comprehensive Metrics:
    • Structural Similarity (SSIM)
    • Perceptual Distance (LPIPS)
    • Mean Squared Error (MSE)
    • PSNR (Peak Signal-to-Noise Ratio)
    • Contrast and Sharpness Preservation
    • Histogram Correlation
    • Edge Density Ratio
    • Processing Runtime
    • Contrast and Sharpness Preservation
    • Histogram Correlation
    • Edge Density Ratio
    • Processing Runtime
  • Augmentation Techniques Catalogue: Includes traditional augmentation methods such as rotation, scaling and flipping, as well as modern, mixing-based techniques such as Mixup, CutMix, AugMix, PixMix, FMix and PuzzleMix.
  • Quantitative Visualisation of Changes: Use visual dashboards to compare the effects of augmentation methods on perceptual similarity, structure and pixel intensity distributions.
  • Web-based and lightweight: It runs directly in modern browsers with no complex setup required.
  • Dockerised architecture: Simplified deployment using Docker Compose for seamless local and production environments.

Project Objectives

  • Provide a visual and interactive platform to help understand data augmentations beyond implementation-level code.
  • Enable easy and meaningful quantitative comparisons between augmentation methods.
  • Facilitate reproducible research experiments and serve as an educational resource.
  • Contribute to the open-source community with a modular and extensible toolkit.

Open Source and Academic Research

This open-source toolkit will be accompanied by a research article detailing the system design, metric rationale and comparative study of augmentation methods. We welcome contributions, feedback and collaborative extensions from the research and developer communities.

Contact

This project developed by Mevlüt Kagan Balga and Fatih Başçiftçi. For any questions, feedback or collaboration, please contact us at kaganbalga@gmail.com or bascfitci@selcuk.edu.tr.