Renoir is a project aimed at recognizing the styles of digital artwork. It originated as an assignment for the 'Introduction to Software Convergence' course at Kyunghee University's Department of Software Convergence.
Downplaying the issue of 'copying' is not something artists take lightly. There are actual cases where illustrators have suffered financially because their artistic style was mimicked.
Generative AI creates new data based on datasets used in its training. It can be seen as generating new data by copying existing ones. Therefore, it is nearly impossible for creators to view generative art AI positively.
As a result, we concluded that a program capable of verifying whether generative art AI has learned a specific artistic style is needed. However, Project Renoir initially focuses on enabling artificial intelligence to recognize artistic styles.
- Backbone : VGG16
- Cost function : Binary Cross Entropy
- Backbone : Inception
- Cost function : Binary Cross Entropy
- Sadly, a code of Renoir 2.0 doesn't exist.
- Backbone : VGG19 without fully connected layer and with Batch Normalization
- Cost function : CosFace
- You can download the .pth file at this link : https://drive.google.com/file/d/11rjLBSiwp_CJ8oYXAP0U6Q9O3VMC8fZ3/view?usp=sharing
First Graph
- It includes training accuracy and test accuracy during epochs 1 to 22.
Second Graph
- It includes training accuracy and test accuracy during epochs 21 to 28.
if you want to read more information, visit my webpage. Please note that the posts on my webpage are written in Korean.
link : https://carpal-money-c50.notion.site/Renoir-362407490c6b403681048c471e52226f?pvs=4
Maybe I will create a GUI for Renoir soon...