After my experiences on Project Immersion, I got inspired to develop a machine learning classifier for my capstone project at the University of Puerto Rico at Arecibo. This linear classifier allows a computer to determine which of the following shot classifications best fits the framing of a person in the picture:

The classifier takes into consideration the proportions of people’s faces and bodies, in relation to the picture’s height. The resulting classifier’s code was published under the BSD 3-Clause License and links to it can be found here:

https://github.com/Cuperino/Tensorflow-Camera-Shot-Classificators

Additional to the classification code, I figured out that by normalizing these proportions with percentages, one can use a system of linear equations to create an equation that, when solved using a set of proportions, gives you the corresponding shot. The experimentation to achieve this was done in GNU Octave using MATLAB syntax. The resulting source code can be found at:

https://github.com/Cuperino/MATLAB-Camera-Shot-Classification-from-Linear-Equations

The dataset used for this research is a subset of images extracted from the Hollywood Human Actions (HOHA) dataset by Ivan Laptev, Marcin Marszałek, and Cordelia Schmid. This dataset is proprietary and free only for academic use. It can be found on Ivan Laptev’s website, at:
https://www.di.ens.fr/~laptev/actions/