OBJECT-SELFIE-MACHINE

What is more important? The technological device, the person, or the selfie? The artwork raises this question when facing the audience to take a selfie as part of an artistic installation -it is not just any digital self-portrait-. It is the creation of a new visual context – born from the research archives of the different departments of the EPN (Escuela Politécnica Nacional del Ecuador) – in an analog-digital-interactive piece.

Through the use of a deep convolutional network and the training of the algorithm with photographs of thousands of people, the program recognizes the person and isolates them from its background; creating a new picture, a real-time techno collage with the ability to be shared instantly through Instagram @epnselfie

The idea stems from a personal approach to the selfie as an object of self-significance, which goes beyond interactivity with human beings or what they do or understand about it.

Technical description:
The technical challenge of this project was to identify and classify an image in real-time (semantic segmentation). The model used is based on the methodology of L. Chen, G Papandreou (2016). This model builds a maximum-hope (EM) estimator from a semi-supervised learning process. It uses a deep convolutional network and training data set with hand-labeled images.