The Zombie Formalist on Social Media

I’ve been doing reflecting and discussing my envisioned use of social media for the Zombie Formalist and the issue is much more complex than I had expected. The purpose of using social media is that ‘likes’ would be one way the machine could determine the ‘value’ of compositions and use that in the training process and model what is liked on social media. After some discussion with a social media savvy person, I came up with three possible strategies for my use of social media, in order of preference:

  1. The Zombie Formalist has it’s own social media profile and all content generated is uploaded.
  2. I create a satirical identity for a company that makes the “Zombie Formalist” as a tech gadget (not artwork it itself) that has a social media presence. The profile would appear to exist only to ‘sell’ the product.
  3. I create a social media profile for myself where the Zombie Formalist output is one component in of a social media presence for my practise in general.

Only #1 allows for social media to be used “in the loop” to attribute value to compositions. #2 and #3 would be more promotional mechanisms, but not be literally connected to the Zombie Formalist hardware as I have envisioned. #2 would be a lot of work in developing marketing and branding; this is an interesting approach, but the required investment makes that a separate project that requires much more time; I could always revisit this when the rest of the project is complete or in a future iteration. #3 would be a very standard use of social media and while it would provide promotional value for my practise, it does not actually have anything to do with this particular project. As it’s a major part of the concept of the work (social media determining value) #1 is the priority.

I was initially inclined to select Instagram because of it’s image-centrism and how it’s used by artists for both promotion and sales. Unfortunately it is not suitable for #1 for a few different reasons: On a technical level, Instagram does not allow posting through the public API, only through the official app and through “partners” who are presumably licensed to allow uploading of content independently of the app. On the social level, from what I understand, success on Instagram means highly curated high-quality content with a strong emphasis on individual brand. Since the Zombie Formalist will generate a lot of mediocre images, where the social media audience defines value, that means that Instagram fits best with options #3, or perhaps #2, but rules it out for #1.

#2 could work well on Facebook also, but Facebook seems to be have a unified API with Instagram and no longer allows creating posts algorithmically (except presumably for those who pay for licenses). When it comes to #1 it seems the only technically and socially viable option is Twitter. The wildness of twitter and the permissive API seem to be much better fits for #1. No wonder there are so many bots on there! The text orientation and the way images are treated on Twitter are not ideal; I find the seemingly arbitrary wide-screen cropping of thumbnails and the compositional emphasis on metadata (hashtags, etc.) particularly unpleasant to deal with… I wonder if there are ways to renter Twitter differently to be more… well… Instagram looking. At least this reflection gives me a direction to work within and I can work on some code and perhaps experiment with uploading (a subset?) of my labelled data-set and see how that works.

Enclosure Fabrication

I have finally gotten a rough sketch of the design for the Zombie Formalist; see the images below for details. The idea is that a minimal structure would be waterjet cut and bent from a single sheet of metal that would hold the screen and parts; a wood frame would slide over that to occlude the technology and make the whole thing appear like a normal contemporary art frame. I’ve approached a few fabricators and will post as that aspect of the project moves along.

DNN Face Detection Confidence — Part 4

I ran the OpenCV DNN-based face detector while I was working and the results are much better than I previous saw with the jetson-inference example. I presume the difference in the performance is due to the use of a different model. The following plot shows my face run (red) on top of the noFace run from the previous post (blue). The face mean confidence was .935 (compared to the mean noFace confidence of 0.11) and there is a clear gap between confidence where a face is present and where no faces are present, as shown in the plot. It seems this is the method I should use; I’ll try integrating it into my existing code and see how problematic the ability to recognize face profiles is.