RULER: Rubric-Based RL Rewards for SVG Code Generation
RULER: Instance-aware Rubric Rewards for SVG Generation
Rubric-derived rewards offer a template for RL on any open-ended generation task where scalar metrics like CLIP trigger reward hacking, extending beyond SVG to other code-to-visual domains. The method needs no human preference labels, so small teams can replicate it with a VLM judge alone. Practitioners training generative models without ground truth should examine the rubric-design ablations, which the paper identifies as the decisive factor.
Researchers introduced RULER (Instance-aware Rubric Rewards for Reinforcement Learning), a method that converts each text instruction into a six-item rubric covering semantic, visual and stylistic axes, with a judge VLM scoring rendered rollouts for GRPO-based optimization.
The rubric is derived from text alone, so no paired SVG ground truth or human preference labels are needed.On MMSVG-Illustration and MMSVG-Icon, RULER lifts rubric scores from 0.
432/0.395 to 0.
693/0.683, surpassing dedicated SVG specialists and matching the much larger DeepSeek-V3.
Ablations identify rubric design as the key lever for RL on open-ended SVG generation.
RULER: Instance-aware Rubric Rewards for SVG Generation
Generating Scalable Vector Graphics (SVG) code from natural-language instructions is an open-ended task without absolute visual ground truth, leaving both evaluation and policy optimization without a faithful signal. Scalar metrics (CLIP, Aesthetic) calibrated on natural images transfer poorly to stylized vector content, and reusing them as RL rewards triggers reward hacking. We address both limitations with rubric-based scoring. We first establish empirically that prompting a vision-language judge with a multi-axis rubric correlates with human judgments far better than scalar metrics, both across samples and within instructions. Building on this finding, we introduce RULER (Instance-aware Rubric Rewards for Reinforcement Learning), which converts each instruction into an instance-aware rubric of six items spanning semantic, visual, and stylistic axes; a judge VLM scores rendered rollouts item-by-item, and the weighted satisfactions form a fine-grained reward optimized via Group Relative Policy Optimization. Because the rubric is derived from text alone, RULER requires neither paired SVG ground truth nor human preference labels. On MMSVG-Illustration and MMSVG-Icon, RULER lifts the rubric score from 0.432/0.395 to 0.693/0.683, surpassing dedicated SVG specialists and matching the substantially larger DeepSeek-V3, with ablations identifying rubric design as the active lever for RL on open-ended SVG generation. The project page is available at https://hangyuran.github.io/RULER/.
This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include RULER, DeepSeek-V3, SVG generation, reinforcement learning.
Primary source verification: published directly by arXiv: DeepSeek with first-party authority.
