VIEScore2: Unified Image Evaluation with Spatially Grounded Explanations
VIEScore2 scores generated images and localizes defects on a grid, beating Gemini-3-Flash on correlation.
VIEScore2 is a unified evaluator for image generation and editing that represents an image as an N×N grid and jointly predicts quality scores and defect locations. It is trained on 38K examples with score-only, localization-only, and joint supervision, then refined with GRPO using Dice overlap, score accuracy, and format-validity rewards. On the primary suite it reaches overall-score SRCC 0.601, compared with 0.491 for Gemini-3-Flash, and ranks among the top localization methods on most benchmarks.
- Predicts quality scores and defect cells in one pass
- Trained on 38K generation and editing examples
- GRPO rewards Dice overlap, score accuracy, and format validity
- Overall-score SRCC 0.601 versus 0.491 for Gemini-3-Flash
- Top three on five of six defect-localization benchmarks
- Vendors
- AI models
- VIEScore2Gemini-3-Flash
From the original article
1 min read at huggingface.co
Existing synthetic image evaluators typically provide only a scalar quality score and do not identify the image regions that support it. We introduce VIEScore2, a unified evaluator for image generation and editing tasks with optional conditioning images. VIEScore2 represents an image as an N x N grid and jointly predicts quality scores and defect locations in a single model pass.
By Xianda Du, Max Ku, Weiming Ren · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.