Reproduce “DIVA-VQA: Detecting Inter-frame Variations in UGC Video Quality” 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
Xinyi Wang, Angeliki Katsenou, David Bull
DIVA-VQA introduces a No-Reference Video Quality Assessment model that assesses user-generated content by detecting inter-frame variations through a patch difference fragmentation strategy. This approach achieved an average SRCC of 0.898 and PLCC of 0.906 across four major UGC datasets, demonstrating enhanced accuracy and generalization for complex video distortions.
Key insights
- Leveraging inter-frame variations through patch difference analysis effectively reveals motion and compression artifacts in both dynamic and static regions, aligning with human perceptual changes.
- Combining a 3D-CNN (SlowFast) for global motion with a 2D-Transformer (Swin Transformer) for detailed spatial features provides a robust and balanced approach to capture complex spatio-temporal distortions in UGC.
- A moderate patch size (e.g., 16x16 for 224x224 input) for fragmentation is optimal for balancing the capture of motion variations and fine-grained details without introducing excessive redundancy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.