2023 Fiscal Year Research-status Report
Language-driven visual content customization using construction sequences.
| Project/Area Number |
23K16921
|
|---|
| Research Institution | The University of Tokyo |
|---|
Principal Investigator |
沈 奕超 東京大学, 大学院情報理工学系研究科, 助教 (40969119)
|
|---|
| Project Period (FY) |
2023-04-01 – 2028-03-31
|
| Keywords | computer graphics / typography / computer vision / icon colorization / virtual try-on |
|---|
| Outline of Annual Research Achievements |
In the past one year, I have been working on various projects related to this funding, including designing high-usability icons, icon colorization, procedural shape generation, and foundation model for typography. These projects lead to four journal papers and one conference short paper. The result outcomes exhibits distinct contributions that we focus on providing novel computational models and foundation models for graphic design elements, such as fonts and icons.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
In the past year, I developed a foundation model for typography which is fundamentally change how multilingual font applications can be achieved. This model can be used to do language-driven font retrieval and vector glyph optimization. Meanwhile, I also developed another foundation model for connecting text descriptions with icons. This foundation model can then be used to design high-usability icons that reduce the effort of icon designers.
These two foundation models are very unique compared to other existing foundation models where our models focus on graphic design domains rather than natural scene images.
|
| Strategy for Future Research Activity |
The development of the foundation model of typography (i.e., FontCLIP) provides a great foundation for me to proceed on related research projects, such as utilize language input to design variable fonts and sentence generation. Meanwhile, with the foundation model for icon, I plan to combine it with a large language model to generate diverse icons that users can use for various graphic design tasks.
|
| Causes of Carryover |
In the past year, since the computational resources, including the machines and GPUs are still suffice for the project need, so I did not spend that much for purchasing new devices. In this year, since I plan to work improving the foundation models that facilitate language driven content generation, so I plan to purchase new computational devices to utilize the research fundings.
|
