Fluid Consciousness
Key Words: Abstract Art, AI Painting, Generative Model, Image Aesthetics
Skills: StyleGAN, Pytorch, Python, Cloud GPU, Eye Tracking
Oct. 2022 - Sept. 2023
-
This is a study on the aesthetics of the image, with the aim of exploring the issue of painting emotions in this field.
-
The videos generated by our trained AI were showcased in the "Participation and Representation: The 5th China Community Aesthetic Education Action Plan" exhibition.
-
As lead technical support, I am responsible for training AI models, applying filtering and other post-processing methods to further improve image quality, and participating in eye-tracking experiments.
Try the online VR exhibition! (Scene 7, turn 180°)
Read the paper I wrote about the detailed research process!
Videos & Images
Concept
「Aesthetic , Perception and Emotion」
-
What to Explore: How would you describe the texture of emotional perception? What does it look like? What other ways are there for us to perceive others? Now Artificial Intelligence is helping us find some answers.
-
Our Work: We hypothesize that the aesthetic composition in abstract paintings can lead to different emotional evaluations. Using a trained StyleGAN3 generative model, we create abstract art images and videos with different compositions and continuous variation representation. Combined with experimental Neuroesthetic research, we correlate human cognitive perception with art composition in images.
-
The Meaning: we are exploring the visual perception and cognitive mechanisms behind human appreciation of art. Hopefully, our work inspires new forms of aesthetic connection and artistic communication.
Research Process
Core Tools
Ⅰ Style Mixing & Projector
-
Style Mixing blends two sources, and generate more images with combined features. In our case, the mixed images receive color from source A and composition from source B.
-
While style mixing allows us to get more diverse pictures, Projector provides ways to observe the continuous process of one image changing into another.
Ⅱ Interactive Visualization
-
Visualizer is an interactive model visualization tool that can be used to explore various characteristics of a trained model.
-
You can load different trained models and adjust parameters to manipulate them.
