Yuet Lee

Breathe (2023)

The work ‘Breathe’ is built upon a conditional variational autoencoder that was trained by an original dataset composed of the artist's handwriting letters. As the model undergoes more training epochs, its outputs gradually become sharper and more defined, instead of simply selecting the ‘best’ outcome(the most recognizable outcome) based on standard efficiency criteria. This work emphasis the evolution and difference of the generative letters from earlier and later epochs, it mixes letters produced from different epochs. This creates an effect that resembles breathing, as the letters come in and out of focus, becoming more or less recognizable. Arranging the produced letters from earlier to later epochs to form a motion of transformation of letters from less recognizable to more recognizable which follows the behavior of breathing. Human breathing follows a certain pattern, breathing in at first then holding the breathe at the peak then breathing out at last, the length of the actions of breathing in and out is longer than the moment of holding the breathe. By following this pattern, arrange 2 pieces of outcome from the earlier epoch to be the motion of breathing in and out respectively, and 1 piece of outcome from the later epoch to be the moment of holding the breathe in between breathing in and out. 

The text "Breathe like everyone does" showed as a command to the audience, which demands conformity with a general standard in the society. The rhythm of the letters can be seen as a model that instructs viewers on the proper way of breathing. However, this demand for conformity carries an ironic tone, satirizing the prevailing pressure to embrace machine learning as the definitive technology of the future.

Project Detail ​

640*360 px | 01:29 | Single-channel video Python 3.11 | Adobe Premier Pro | Adobe Illustrator ​

Reference Code ​

[1] CVAE Implementation Hujinsen. (2018, May 20). Pytorch_VAE_CVAE/CVAE.ipynb at master · hujinsen/pytorch_VAE_CVAE. GitHub. https://github.com/hujinsen/pytorch_VAE_CVAE/blob/master/CVAE.ipynb [2] Labeling and data loader Subramanyam, V. S. (2021, January 29). Creating a custom dataset and Dataloader in Pytorch. Medium. https://medium.com/analytics-vidhya/creating-a-custom-dataset-and-dataloader-in-pytorch-76f210a1df5d ​

Project Advisor ​

Prof. Héctor RODRIGUEZ