Benchmark RadarChartSync
ChartSync evaluates image editing models on chart editing tasks, including text-only edits and visuo-logical cascading edits (VLCE) requiring synchronized text and geometry changes. The benchmark contains 870 expert-validated triplets across nine chart families, with objective visual metrics and a vision-language model judge for evaluation.
- Released
- 2026-07-11
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Existing image editing benchmarks often overlook structured data charts where data modifications require geometric synchronization. ChartSync provides a reproducible evaluation to assess models' ability to handle dependency-aware cascading updates, revealing capability gaps in open-source versus proprietary models.
Motivation
Generative image editing models struggle with structured statistical charts when data modifications require geometric synchronization.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.