Machine Learning Art
Architectural morphing
a new transfer learning strategy | Demo
In architecture, what is morphing?
Morphing is when an item gradually alters its shape to take on a new one. The essential procedure of morphing is the selection of two objects and the assignment of n, the number of intermediate phases. In n stages, the first item transforms into the second.
- May 2022 — AI art tools update can be found ➡️ HERE ⬅️
Transfer Learning of StyleGAN
StyleGAN’s transfer learning has lately shown considerable promise in various applications, particularly domain translation. Previous systems used a source model during transfer learning by swapping or freezing weights, but they had limitations in terms of visual quality and source control. In other words, they necessitate additional computationally intensive models with limited control steps that hinder a seamless transition. To circumvent these restrictions, the authors offer a novel strategy. First, they provide a simple feature matching loss to increase generation quality instead of switching or freezing. In addition, they train a target model with the suggested technique, FixNoise, to maintain source features exclusively in a disentangled subspace of a target feature space to regulate the degree of source features.
Architectural morphing and domain translation are two examples of multi-domain features that may be used with the approach.
Fix the Noise
The new transfer learning approach described by the authors for controlling conserved source domain characteristics in the target model. FixNoise, when used in conjunction with a basic feature matching loss, correctly separates the source and target features in the target model’s feature space. As a result, their technique can regulate the source features in a single model via noise interpolation. This indicates that their method is more computationally efficient than other ways. Additionally, experimental findings show that the suggested technique surpasses prior picture quality and smooth transition efforts.
@article{lee2022fix,
title={Fix the Noise: Disentangling Source Feature for Transfer Learning of StyleGAN},
author={Lee, Dongyeun and Lee, Jae Young and Kim, Doyeon and Choi, Jaehyun and Kim, Junmo},
journal={arXiv preprint arXiv:2204.14079},
year={2022}
}
Project Page:
https://arxiv.org/pdf/2204.14079.pdf
DEMO:
https://github.com/LeeDongYeun/FixNoise/blob/main/demo.ipynb
Keywords: computer vision, Artificial Intelligence, datasets, Machine Learning, AI art, art, digital art, 3D, generative, 3D modeling, Architectural morphing, Pattern Recognition , architecture, StyleGAN2
I invite you to explore the concept of “AI creativity” by reading and learning from the many articles found on 🔵 MLearning.ai 🟠
- Check out my instagram with new material every week
- If you enjoyed this, follow me on Medium for more
- Want to collaborate? Let’s connect on LinkedIn
- https://linktr.ee/datasculptor
- 3D Machine Learning generated model on sketchfab
Data Scientists must think like an artist when finding a solution when creating a piece of code. Artists enjoy working on interesting problems, even if there is no obvious answer.
All our writers (members) receive the opportunity to be promoted on our social media, which increases the popularity of articles published on MLearning.ai