Search papers, labs, and topics across Lattice.
Flint is an intermediate language designed to streamline the creation of high-quality visualizations through concise, semantics-driven specifications, eliminating the need for users to manually configure low-level parameters. By employing a hierarchical data semantic model, Flint enables users to define the meanings of data fields, allowing the compiler to generate optimized visualization configurations that are adaptable across various libraries. The results indicate that Flint not only simplifies the visualization authoring process but also maintains high visual quality, making it a valuable tool for both human users and AI agents.
Flint transforms the way visualizations are created by allowing users to specify data meanings, leading to optimized and high-quality outputs without tedious configuration.
We present Flint, an intermediate language that enables authors to create high-quality visualizations from concise, semantics-driven specifications without explicitly configuring low-level parameters such as scales, axes, and formatting. Unlike prior systems that infer default configurations from surface-level data representations, often producing brittle choices, Flint introduces a hierarchical data semantic model that allows users to specify the meanings of data fields structurally and helps the compiler derive appropriate visualization configurations. From a concise specification, the system generates and optimizes library-agnostic visualization configurations and translates them into complete, executable specifications for multiple target grammars, including Vega-Lite, Apache ECharts, and Chart.js. We demonstrate that Flint simplifies the authoring process without compromising on visual quality, and it is an effective intermediate language for both humans and AI agents to create visualizations.
Tsinghua AI