User Guide

These sections demonstrate how to use plenoptic in a variety of contexts.

Users are encouraged to read the introductions for the synthesis method or model / metric they would like to use.

The Reproducing examples from the literature section includes examples of how to use plenoptic to reproduce results from published papers.

Finally, the Advanced Usage section describes some more complicated ways of interacting with plenoptic, which may be helpful for advanced users.

Synthesis method introductions

• Eigendistortions
  • Introduction
  • Example 1: Linear model, small 1D input “image”
  • Example 2: Which layer of ResNet is a better model of human visual distortion perception?
  • Appendix: More mathematical detail
• MAD Competition Conceptual Introduction
  • Full images!
• MAD Competition Usage
  • Basic usage
• Metamers
  • Basic usage
  • Reproducibility
  • More Advanced Options

Models and metrics

• Perceptual distance
  • References
  • SSIM (structural similarity)
  • MS-SSIM (multiscale structural similarity)
  • NLPD (normalized Laplacian pyramid distance)
  • Usage
  • Comparison of performance
• Steerable Pyramid
  • Introduction: Steerable Pyramid Wavelets
  • Boundary handling
  • Putting the “Steer” in Steerable Pyramid
  • Example Application: Frontend for Convolutional Neural Network
  • Model Training
• Portilla-Simoncelli Texture Model
  • Getting started
  • Understanding the Model
  • Technical details
  • Advanced usage

Reproducing examples from the literature

• Reproducing Berardino et al., 2017 (Eigendistortions)
  • What do eigendistortions tell us?
  • Input preprocessing
  • Synthesizing eigendistortions
  • Final thoughts
• Reproducing Wang and Simoncelli, 2008 (MAD Competition)
  • SSIM
  • MAD Competition

Advanced Usage

• Display and animate functions
  • General
  • Synthesis-specific
  • More complicated model representation plots
• Extending existing synthesis objects