Reproduce “Personalized Image Aesthetics Assessment with Rich Attributes” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuzhe Yang, Liwu Xu, Leida Li, Nan Qie, Yaqian Li, Peng Zhang, Yandong Guo
The OPPO Research Institute and Xidian University created the Personalized Image Aesthetics database with Rich Attributes (PARA), a comprehensive dataset designed to advance personalized image aesthetics assessment. PARA includes 31,220 images with diverse annotations, encompassing objective image attributes, human-oriented subjective attributes, and desensitized subject information, empirically showing that incorporating subjective attributes significantly improves PIAA model performance over unconditional approaches.
Key insights
- Existing Personalized Image Aesthetics Assessment (PIAA) datasets lack diverse annotations, particularly regarding the 'subject aspect,' which limits the development of robust personalized models.
- Human-oriented subjective attributes, such as personality traits and emotional responses, significantly influence individual aesthetic judgments and can be leveraged to enhance PIAA.
- The inherent subjectivity of aesthetic judgment results in higher score variance for images with low to mid-range aesthetic appeal, while high-aesthetic images tend to have more consensus.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.