2020 Fiscal Year Research-status Report
Understanding Concrete and Abstract Representations in Art
Project/Area Number |
20K19822
|
Research Institution | Osaka University |
Principal Investigator |
GARCIA・DOCAMPO NOA 大阪大学, データビリティフロンティア機構, 特任研究員(常勤) (80870005)
|
Project Period (FY) |
2020-04-01 – 2022-03-31
|
Keywords | art analysis / visual understanding / deep neural networks / transfer learning / question answering |
Outline of Annual Research Achievements |
During the first year of research, we collected a dataset for visual question answering on art. The dataset was presented an international workshop of computer vision for art (VisArt) in one of the most important computer vision conferences (ICCV) in front of worldwide experts. Additionally, we conducted extensive experiments on the dataset and presented a new model to address visual question answering on art, obtaining outstanding performance.
|
Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
We already collected the dataset for visual question answering in art and performed extensive evaluation of different baselines and models. We also applied these techniques to automatic description of artworks and visual question answering for natural images.
|
Strategy for Future Research Activity |
For future research, we plan keep exploring better models for high-level representation of art and apply them to visual question answering, image description generation and semi-supervised learning. We are currently studying emotions associated to art using image representations.
|
Causes of Carryover |
Because of COVID-19 travel restrictions, I could not use the funding as planned to attend conferences and give talks about our research. For next year, I will use the funding on: 1) traveling to international conferences, 2) paying journal publication fees, 3) dataset annotations, and 4) GPU machine purchase.
|