Meijia Huang, Yingjie Yin, Shihao Wang +1 more
A beautiful chart is worth a lot, because people remember it. Generating one automatically means satisfying several constraints at the same time, and they pull against each other: the numerical geometry has to stay accurate, text has to render exactly rather than approximately, each label has to attach to the right mark, and the whole thing has to hold a coherent artistic style.
Current text-to-image and image editing models tend to break on the combination. The failures listed are specific: distorted geometry, hallucinated text, labels bound to the wrong marks, and stylisation heavy enough to hurt readability. Any one of those turns a chart from information into decoration.
That is the reason this is a hard benchmark rather than a cosmetic one. A chart with invented numbers is worse than no chart, because it carries the authority of looking like data. ArtChart is built to measure whether the mathematical integrity survives the styling.
Artistic charts make data memorable and visually engaging, but generating them faithfully demands simultaneously preserving numerical geometry, rendering exact in-image text, binding labels to correct marks, and maintaining coherent artistic style. Current text-to-image and image editing models frequently fail on these coupled constraints, producing distorted geometries, hallucinated text, misbound labels, or over-stylized marks that undermine readability and mathematical…
Behaviour-Conditioned Neural Processes for Adaptive Residential Short-Term Load Forecasting
arXiv (cs.AI) · July 17, 2026An Exam for Active Observers
arXiv (cs.LG) · July 17, 2026PRISA: Proactive Infrastructure LiDAR Framework for Intersection Safety Assessment
arXiv (cs.CV) · July 17, 2026CLIFE: Camera-LiDAR Fusion Framework for Edge-Deployable Roadside VRU Perception