MetaRefine: Score Refinement with Multi-Level Features for Image Assessment
What it is
MetaRefine is a lightweight architecture that improves image quality assessment scores by combining original image geometry (dimensions, aspect ratio) with multi-level visual features. It uses residual heads to predict score corrections on top of frozen existing models without altering their original scoring pipelines.
The problem it solves
Standard image assessment methods often resize, crop, or pad images, which can obscure original geometric and compositional cues that are critical for accurate quality evaluation.
What you could use it for
- Improving the accuracy of existing image quality assessment models
- Refining aesthetic and technical quality scores for image datasets
- Enhancing state-of-the-art assessment benchmarks with minimal parameter overhead
Why it matters
Significantly improves the correlation and error metrics of image quality assessment models with very low computational cost, potentially making existing models more robust to geometric variations.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.