Reproduce “Non-Aligned Reference Image Quality Assessment for Novel View Synthesis” 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
Abhijay Ghildyal, Rajesh Sureddi, Nabajeet Barman, Saman Zadtootaghaj, Alan Bovik
A new framework, NOVA, was developed for Non-Aligned Reference Image Quality Assessment (NAR-IQA) in Novel View Synthesis (NVS) applications. This model accurately assesses the perceptual quality of synthesized novel views using non-aligned reference images, demonstrating superior performance and robustness to spatial misalignment on both a newly constructed NVS NAR-IQA benchmark and an existing NVS-QA benchmark.
Key insights
- Training on synthetically generated, localized distortions within Temporal Regions of Interest (TROIs) enables a model to generalize effectively to diverse and unseen NVS artifacts and scene layouts.
- Supervised contrastive learning, guided by established Full-Reference IQA metrics, can create a perceptually aligned embedding space that remains robust despite spatial misalignments and viewpoint changes.
- Leveraging a LoRA-enhanced DINOv2 foundation model within this framework allows for high performance and generalization in non-aligned reference image quality assessment for novel view synthesis.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.