PR-IQA creates geometrically consistent partial quality maps from reference overlaps and completes them via cross-attention to evaluate diffusion-synthesized views without ground truth, achieving full-reference accuracy and improving 3DGS reconstructions.
First, PR-IQA is currently trained using pseudo-GT quality maps derived from FR metrics, specifically DINOv2 feature similarity or SSIM
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
PR-IQA: Partial-Reference Image Quality Assessment for Diffusion-Based Novel View Synthesis
PR-IQA creates geometrically consistent partial quality maps from reference overlaps and completes them via cross-attention to evaluate diffusion-synthesized views without ground truth, achieving full-reference accuracy and improving 3DGS reconstructions.