MATRIX 4 AI Data Sculpture

STATEMENT

 

Last year, Warner Bros. invited Ouchhh for creating a commission art piece about the Matrix 4 movie. The artwork would be installed simultaneously in Tokyo, London, NY, and San Francisco for the premiere of Matrix The project had to be canceled due to Covid regulations.

For the first time in the world, we created a data sculpture that can be experienced interior and exterior facade generated by ai.

In order to start and create a dataset to train the AI, the Matrix movie series which consists of three movies (The Matrix (1999), The Matrix Reloaded (2003), The Matrix Revolutions (2003)) are divided into frames as JPG sequence. After that process, the dataset which contains approximately 250000 images is created. The reason behind creating an image dataset to teach an AI is very similar to the very fundamental principle of filmmaking, from pictures to moving pictures. To analyze the data gathered from the Matrix movie series, and then create new pictures based on the dataset with the help of AI, some machine learning and deep learning techniques are used.

Since the GAN algorithm is not applicable to the whole dataset, it is first decided to create sub-datasets with images which have similar subjects and then train GANs on the sub-dataset separately. In order to create sub-datasets, first, it must have been understood how many different classes the main dataset consists of. The t-SNE algorithm was first used to understand the structure of the dataset in the three-dimensional universe visually. After understanding the dataset has 50 different classes, to divide the images into clusters, the K-Means algorithm is used, and eventually, sub-datasets are created. With those datasets, separate GAN algorithms started training.

For extracting the faces or postures from the images pre-trained CNN (Convolutional Neural Networks) algorithm from OpenCV library of python is used.