Existing graph layout algorithms are usually not able to optimize all the aesthetic properties desired in a graph layout. To evaluate how well the desired visual features are exhibited in a graph layout, many readability metrics were introduced in the past decade. However, the calculation of these readability metrics often requires access to the node and edge coordinates and is usually computationally inefficient, especially for dense graphs. Importantly, when the node and edge coordinates are not accessible, it becomes impossible to evaluate the graph layouts quantitatively. We propose a novel deep-learning based approach to assess the readability of layouts by directly using graph images. A convolutional neural network architecture is proposed, trained on a benchmark dataset of graph...[ Read more ]

Existing graph layout algorithms are usually not able to optimize all the aesthetic properties desired in a graph layout. To evaluate how well the desired visual features are exhibited in a graph layout, many readability metrics were introduced in the past decade. However, the calculation of these readability metrics often requires access to the node and edge coordinates and is usually computationally inefficient, especially for dense graphs. Importantly, when the node and edge coordinates are not accessible, it becomes impossible to evaluate the graph layouts quantitatively. We propose a novel deep-learning based approach to assess the readability of layouts by directly using graph images. A convolutional neural network architecture is proposed, trained on a benchmark dataset of graph images, which is composed of synthetically-generated and graphs created by sampling from real large networks. The proposed method is quantitatively compared to traditional methods and qualitatively assessed with the help of a case study. This work is a first step towards using deep learning based approach to evaluate images from the visualization field quantitatively.