Self-Distillation Policy Optimization via Visual Feedback: Bridging Code and Visual Artifacts

• URL: http://arxiv.org/abs/2606.10334v1
• Date: Tue, 09 Jun 2026 02:28:39 GMT
• Title: Self-Distillation Policy Optimization via Visual Feedback: Bridging Code and Visual Artifacts
• Abstract summary: We study visual self-distillation for code-generated visual artifacts.<n>Visual-SDPO treats rendered visual feedback as privileged context for a weight-sharing teacher.<n>Visual-SDPO improves over the zero-shot base by more than 10 absolute points in the primary metric.
• Score: 4.312910873423553
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Code-generating large language models (LLMs) increasingly produce visual artifacts such as charts, web pages, and slides by writing programs that are executed by non-differentiable renderers, committing to code before observing the render. As a result, otherwise executable code often yields artifacts with visually salient defects, including overlapping elements, clipped text, broken alignment, low contrast, and overflow. We study visual-feedback self-distillation for code-generated visual artifacts. We propose Visual-SDPO, a self-distillation policy-optimization framework that treats rendered visual feedback as privileged context for a weight-sharing teacher and distills this feedback into a coding student. To make supervision spatially targeted rather than uniform, we introduce Visual-Grounded Code Credit Weighting, which traces each detected defect back to the code statements responsible for the affected elements and amplifies the distillation signal on those statements. A sequence-level GRPO (Group Relative Policy Optimization) term complements the dense token-level objective by rewarding executable, visually high-quality rollouts, while failed executions remain learnable through the self-distillation path by passing execution errors as privileged context to the teacher. We instantiate Visual-SDPO for chart, web/UI, and slide generation with a unified Qwen3-VL-8B-Instruct backbone. Across chart-to-code, UI-to-code, and slide-generation benchmarks (ChartMimic, Design2Code, and AeSlides), Visual-SDPO improves over the zero-shot base by more than 10 absolute points in the primary metric and over GRPO by at least 2.4 points, with fewer training steps and no added inference-time cost.

Related papers

• Rubrics as Visual-Repair Context for Self-Evolving UI-to-Code Generation [147.1183324267404]
  A local code edit propagates through layout, style, and component dependencies, correcting one visual mismatch while degrading regions that were previously faithful.<n>We present RubSE, a rubric-guided framework that uses rubrics to represent visual feedback as a structured visual-repair context.<n>RubSE substantially outperforms nave self-evolution in final-round and best-round settings, achieving more stable refinement and a higher trajectory-level performance ceiling.
  arXiv  Detail & Related papers  (2026-08-25T07:03:46Z)
• GALA: Multimodal Graph Alignment for Bug Localization in Automated Program Repair [40.228155198574356]
  We propose GALA, a framework that shifts multimodal Automated Program Repair (APR) from implicit semantic guessing to explicit structural reasoning.<n>GALA operates in four stages: it first constructs an Image UI Graph to capture visual elements and their structural relationships; then performs file-level alignment by cross-referencing this UI graph with repository-level structures.<n>By enforcing both semantic and relational consistency across modalities, GALA establishes a highly accurate visual-to-code mapping.
  arXiv  Detail & Related papers  (2026-04-09T11:06:25Z)
• OmniDiagram: Advancing Unified Diagram Code Generation via Visual Interrogation Reward [8.227370271724162]
  We introduce a novel visual feedback strategy named Visual Interrogation Verifies All (textscViva)<n>textscViva rewards the visual structure of rendered diagrams through a generative approach.<n>We construct M3$2$Diagram, the first large-scale diagram code generation dataset.
  arXiv  Detail & Related papers  (2026-04-07T07:10:24Z)
• Visual-ERM: Reward Modeling for Visual Equivalence [59.317480168347664]
  Visual Equivalence Reward Model (Visual-ERM) is a multimodal generative reward model that provides fine-grained, interpretable, and task-agnostic feedback.<n>Visual-ERM improves Qwen3-VL-8B-Instruct by +8.4 on chart-to-code and yields consistent gains on table and SVG parsing.<n>VisualCritic-RewardBench (VC-RewardBench) is a benchmark for judging fine-grained image-to-image discrepancies on structured visual data.
  arXiv  Detail & Related papers  (2026-03-13T17:58:14Z)
• Visual Self-Refine: A Pixel-Guided Paradigm for Accurate Chart Parsing [76.2602505940467]
  Existing models often struggle with visually dense charts, leading to errors like data omission, misalignment, and hallucination.<n>Inspired by the human strategy of using a finger as a visual anchor'' to ensure accuracy when reading complex charts, we propose a new paradigm named Visual Self-Refine (VSR)<n>The core idea of VSR is to enable a model to generate pixel-level localization outputs, visualize them, and then feed these visualizations back to itself, allowing it to intuitively inspect and correct its own potential visual perception errors.
  arXiv  Detail & Related papers  (2026-02-18T13:40:53Z)
• SVRepair: Structured Visual Reasoning for Automated Program Repair [17.545585659174773]
  Large language models (LLMs) have recently shown strong potential for Automated Program Repair (APR)<n>We propose textbfSVRepair, a multimodal APR framework with structured visual representation.<n> SVRepair first fine-tunes a vision-language model, textbfStructured Visual Representation (SVR), to uniformly transform heterogeneous visual artifacts into a emphsemantic scene graph.
  arXiv  Detail & Related papers  (2026-02-05T06:26:46Z)
• Aligning Text, Code, and Vision: A Multi-Objective Reinforcement Learning Framework for Text-to-Visualization [50.13408999553116]
  We propose RL-Text2Vis, the first reinforcement learning framework for Text2Vis generation.<n>Our method uses a novel multi-objective reward that jointly optimize textual accuracy, code validity, and visualization quality.<n>Our results establish GRPO as an effective strategy for structured, multimodal reasoning in visualization generation.
  arXiv  Detail & Related papers  (2026-01-08T04:29:07Z)
• Rethinking Prompt Design for Inference-time Scaling in Text-to-Visual Generation [63.042451267669485]
  We propose Prompt Redesign for Inference-time Scaling, a framework that adaptively revises the prompt during inference in response to scaled visual generations.<n>We introduce a new verifier, element-level factual correction, which evaluates the alignment between prompt attributes and generated visuals at a fine-grained level.<n>Experiments on both text-to-image and text-to-video benchmarks demonstrate the effectiveness of our approach.
  arXiv  Detail & Related papers  (2025-12-03T07:54:05Z)
• RECODE: Reasoning Through Code Generation for Visual Question Answering [68.86938437188964]
  We propose to leverage derendering -- the process of reverse-engineering visuals into executable code -- as a new modality for verifiable visual reasoning.<n>Our work demonstrates that grounding visual perception in executable code provides a new path toward more accurate and verifiable multimodal reasoning.
  arXiv  Detail & Related papers  (2025-10-15T17:05:37Z)
• ReLook: Vision-Grounded RL with a Multimodal LLM Critic for Agentic Web Coding [35.30369059154906]
  ReLook is an agentic, vision-grounded reinforcement learning framework.<n>It empowers an agent to close a robust generate--diagnose--refine loop.<n>It consistently outperforms strong baselines in vision-grounded front-end code generation.
  arXiv  Detail & Related papers  (2025-10-13T15:05:50Z)
• VisCoder: Fine-Tuning LLMs for Executable Python Visualization Code Generation [69.35779796364413]
  We present VisCode-200K, a large-scale instruction tuning dataset for Python-based visualization and self-correction.<n>It contains over 200K examples from two sources: (1) validated plotting code from open-source repositories, paired with natural language instructions and rendered plots; and (2) 45K multi-turn correction dialogues from Code-Feedback.
  arXiv  Detail & Related papers  (2025-06-04T13:24:44Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.