What is PixPlot? (DH Tools) – YouTube

Introduction by OpenMethods Editor (Marinella Testori):

This short video teaser summarizes the main characteristics of PixPlot, a Python-based tool for clustering images and analyzing them from a numerical perspective as well as its pedagogical relevance as far as
machine learning is concerned.

The paper “Visual Patterns Discovery in Large Databases of Paintings”, presented at the Digital Humanities 2016 Conference held in Poland,
can be considered the foundational text for the development of the PixPlot Project at Yale University.

As the authors Lenardo, Seguin and Kaplan highlight in their abstract, there may be an increasing demand for accessing visual data by means of innovative techniques, not exclusively based on the exploitation of metadata.

Among these techniques, there is that of Convolutional Neural Networks (CNN), about which an introduction can be found in Indolia et al. (2018).
Practically speaking, the CNN approach enhances the traditional parameters for object classification by means of a multiple-layer deep learning method which has found its first, ideal application in visual art history.

Peter Leonard, in his contribution regarding “Semantic Image Clustering with Neural Networks”, describes the development of PixPlot
as well as its main features and potentialities, particularly: a) the exploitation of the so-called ‘penultimate layer’ of image visualization in the network; b) the improvement of the image reduction in its suitability for a computer-screen visualization; c) a high-performance user experience.