Reproduce “VADB: A Large-Scale Video Aesthetic Database with Professional and Multi-Dimensional Annotations” 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
Qianqian Qiao, DanDan Zheng, Yihang Bo, Bao Peng, Heng Huang, Longteng Jiang, Huaye Wang, Jingdong Chen and 2 more
Researchers from Nanjing University and collaborators introduce VADB, a large-scale video aesthetic database with 10,490 professionally and multi-dimensionally annotated videos, alongside VADB-Net, a dual-modal assessment model. VADB-Net achieved superior performance in video aesthetic evaluation, with an SRCC of 0.93 and PLCC of 0.93 for overall scores, surpassing existing video quality assessment models.
Key insights
- Professional, multi-dimensional annotations (scores, comments, tags) are crucial for building comprehensive video aesthetic assessment datasets that capture both subjective and objective aspects.
- A CLIP-like dual-modal pre-training framework, leveraging video and rich text descriptions, effectively learns robust and transferable visual representations for complex video aesthetic understanding.
- Dedicated deep learning models designed for aesthetic judgment, rather than generic quality assessment, can achieve significantly higher performance in predicting nuanced human aesthetic evaluations of video.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.