VIEScore2 scores images and pinpoints defects on a grid
VIEScore2: Unified Image Evaluation with Spatially Grounded Explanations
Computer Vision and Pattern Recognition
Summary
Images made or changed by computers are usually given a single quality score, but it's hard to know what parts of the image caused that score. This work introduces VIEScore2, a system that looks at a picture in a grid layout and gives both a quality score and marks spots where defects may be. It learns from many examples with different types of feedback, and it explains its assessment in an easy-to-read way. The system performs better than similar existing tools, both for scoring image quality and for showing where problems are.
What this means in practice
- •For digital imaging teams: Identify quality issues and their locations in generated or edited images to improve visual content workflows.
- •For quality assurance teams: Automate detection and explanation of defects in images to reduce manual inspection efforts in media production.
Authors
Xianda Du, Max Ku, Weiming Ren, Zhi Rui Tam, Chunlin Ren, Ping Nie, Min-Hung Chen, Wenhu Chen
Abstract
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. Its text-native grid representation provides a common interface for heterogeneous spatial supervision and enables directly verifiable post-training objectives. We train on 38K examples spanning score-only, localization-only, and joint supervision across generation and editing tasks. Starting from supervised fine-tuning, we further apply GRPO to improve defect localization using rewards that combine cell-level Dice overlap, score accuracy, and output-format validity. A parameter-free parser converts the structured predictions into readable explanations. On the primary suite, VIEScore2 achieves an overall-score SRCC of 0.601, compared with 0.491 for Gemini-3-Flash, the strongest zero-shot general-purpose VLM baseline under matched inputs. For defect localization, VIEScore2 outperforms both general-purpose VLMs and specialized spatial evaluators on three of six benchmarks in per-image grid IoU and ranks among the top three on five, including datasets beyond its training sources.