Tittle: Aesthetic quantitative assessment of innovative design and element image
 
Background:
Graphic designers design images or poster with high aesthetic quality for advertising. It is known that designing is time-consuming and labor-intensive, as a result, some applications has been developed for automatic generation of designed images. However, not all these generated images are beautiful and the aesthetic quality assessment is important to help select the beautiful images.
 
Ordinary people judge the aesthetic quality of images mainly according to their subjective feelings. There are massive images of different aesthetic quality on the Internet, which provides enough data for image aesthetic assessment. Recently, CNN has been commonly used to assess the aesthetic quality and achieved a better performance than traditional methods. However, the aesthetic quality assessment of graphic designed images hasn’t gotten much attention.
 
we are planning to train a model to understand and predict the aesthetic quality of automatically generated images. This model will be used in the Ali Intelligent Design Platform to help select more beautiful images, which can improve the user experience.
 
Research Target:
For the aesthetic quality assessment of whole images, we plan to achieve the accuracy of more than 90% on the testing set, and the accuracy of more than 85% in the real test with the limitation of less then 500ms computing time.
 
For the aesthetic quality assessment of image elements, we plan to achieve the accuracy of more than 90% on the testing set, and the accuracy of more than 85% in the real test
with the limitation of less then 300ms computing time.
 
The researchers should propose the interpretable aesthetic features to describe banners inspired by art theories, base on that, design the labeling and evaluation specifications for elemental elements and whole images.