Unverified50% confidenceFactExact time
WeGenEditBench's evaluation pipeline first establishes item-level deductions and visual evidence, then aggregates them into scores across four dimensions: editing execution, content preservation, visual quality, and task-specific correctness.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe combination of WeGenEditBench and WeGenEvoStudio provides an interpretable evaluation path, where every score is backed by atomic requirements and visual evidence, and every failure point has a corresponding data construction and adaptation mechanism.73% similarUnverifiedEvaluation of graphical abstracts should place more emphasis on visual impact, information density, and readability of conclusions; evaluation of technical roadmaps focuses on the logical coherence of the workflow, clarity of stage division, and semantic consistency of arrows and nodes; evaluation of illustrative figures emphasizes detail fidelity, annotation completeness, and reproducibility.69% similarUnverifiedThe researchers built ReGap-Bench, a paired benchmark containing five types of visual reasoning tasks, to test the effect of external reference on model performance.68% similarUnverifiedIGMBench evaluation adopts four categories of deterministic checks: executability, behavior, preservation, and visual.68% similarUnverified锚定样本技术在评估提示词中加入已由人类专家标注明确分数的示例，覆盖评分量表各个档位，可帮助裁判模型准确理解评分标准68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1125243API
curl https://kongchang.com/api/v1/knowledge/claims/1125243MCP
get_claim(id=1125243)