Numerically Ranking The Artworks

Generative NFTs have been ranked on the rarity of traits. The same is; however, not possible for the 'The First Avatars' as each artwork is unique coming from a neural net. It will be possible to do a classification analysis at some point in the future to analyze which traits best fit which image portrait, but that is not an easy classification task. We look at some of the numerical ratings schemes proposed by GPT4.0.

Composition

An edge detection algorithm is used as a measure of composition. The idea being the density of edges would give an insight intro the complexity of the artwork. However the context needs to be understood. An expanding spiral would likely appear to be highly complex . We use the opencv Canny function.

Clarity

Laplacian variance is used in image analysis to understand 'clarity'. The Laplacian operator calculates the second derivative of the image’s intensity values. It emphasizes areas where the intensity changes quickly. A high laplacian variance indicates indicates the image has clear edges and fine details corresponding to a well-focused image.

Use of Colour

We focus on the hue channel, a histogram shows the frequency distribution of different colors in the image. High entropy indicates a diverse and complex color palette.

Visual Impact

Essentially refers to the brightness of the artwork

Clearly from the above sample images, the Composition analysis using Canny and the Use of Color using the Hue seem to be very different measures that can be used in rankings. Visual Impact and Clarity seem to be weaker measures. Both distributions - 'Composition' and 'Use of Color', exhibit a skew . This data is converted to a scale of 1 to 10 in the metadata for 'The First Avatars'.

Analysis of traits and rarities has its place. The lack of these provides the buyer with opportunities, for example, we have noticed that some portraits have a child portrait, some have double features (errors in AI engine), some have a trait like a pipe or a crown, some have goatees, etc., etc. These have not been documented as traits but are obviously visible. Let these be of additional value to the buyer. The use of the above rankings is very simplistic frankly, these ratings would not recognize the genius of a Rothko for example. A project of this depth (this is not a doodle project) will lead to a lot of analysis , there is a scope for a lot of art 'criticism' as each artwork is unique.

Research and choose wisely.

References

Conversations with GPT4.0