Get Started
Home
Topics
Search
Library
Research questionHow should chart-to-code models handle latent data that rendered charts do not uniquely determine?Rendered charts may preserve only aggregates, proportions, projections, or selected contour levels, leaving multiple underlying programs visually compatible. Supervising against one reference script can therefore penalize code that is consistent with everything visible in the chart.
AI
Code Generation & Program Synthesis
Computer Vision
Evaluation & Benchmarks
Multimodal Models
Research Paper
Latest papersRecent research connected to this question, newest first.What You See Is What You Get: Observation-Aligned Supervision for Chart-to-Code GenerationApplies to chart-to-code generation across box, histogram, pie, 3D scatter, and contour charts. The evidence covers controlled experiments with multiple vision-language models, measuring observable-value and end-to-end recovery; the contour study also reports a trade-off involving representational compactness.research paper · Sep 3, 2026
Related questions
How can visual generative models infer latent rules and produce globally logically consistent images?How can process simulators preserve latent-state uncertainty when generating trajectories from partial event logs?How can sparse SDF observations constrain a frozen INR’s latent code to complete plausible unseen 3D geometry?How can graph topology inference from node observations be unified with generating new graph instances?