# SketchVL: Policy Optimization via Fine-Grained Credit Assignment for Chart Understanding and More Source: [SketchVL: Policy Optimization via Fine-Grained Credit Assignment for Chart Understanding and More]() Collection index: [llms.txt]() Data updated: 2026-01-13T05:00:04.831Z Authors: Muye Huang, Lingling Zhang, Yifei Li, Yaqiang Wu, Jun Liu Published: 2026-01-09T10:13:01.000Z ## Abstract Charts are high-density visual carriers of complex data and medium for information extraction and analysis. Due to the need for precise and complex visual reasoning, automated chart understanding poses a significant challenge to existing Multimodal Large Language Models (MLLMs). Many MLLMs trained with reinforcement learning (RL) face the challenge of credit assignment. Their advantage estimation, typically performed at the trajectory level, cannot distinguish between correct and incorrect reasoning steps within a single generated response. To address this limitation, we introduce SketchVL, a novel MLLM that optimized with FinePO, a new RL algorithm designed for fine-grained credit assignment within each trajectory. SketchVL's methodology involves drawing its intermediate reasoning steps as markers on the image and feeding the annotated image back to itself, creating a robust, multi-step reasoning process. During training, the FinePO algorithm leverages a Fine-grained Process Reward Model (FinePRM) to score each drawing action within a trajectory, thereby precisely assigning credit for each step. This mechanism allows FinePO to more strongly reward correct tokens when a trajectory is globally successful, and more heavily penalize incorrect tokens when the trajectory is globally suboptimal, thus achieving fine-grained reinforcement signals. Experiments show that SketchVL learns to align its step-level behavior with the FinePRM, achieving an average performance gain of 7.23\\% over its base model across chart datasets, natural image datasets, and mathematics, providing a promising new direction for training powerful reasoning models. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Chart-based Question Answering on EvoChart-QA](): Chart-based Question Answering; EvoChart-QA; Accuracy: 58.64. - [Chart Question Answering on ChartBench](): Chart-based Question Answering; ChartBench; Accuracy: 65.11. - [Chart-based Question Answering on ChartQA Pro](): Chart-based Question Answering; ChartQA Pro; Accuracy: 52.62. - [Chart Question Answering on ChartQA](): Chart Question Answering; ChartQA; Accuracy: 83.96. - [Multimodal Understanding on MMStar](): Multimodal Understanding; MMStar; Accuracy: 57.13. - [Mathematical Reasoning on MathVista (Accuracy)](): Mathematical Reasoning; MathVista; Accuracy: 63.5. - [Visual Question Answering on PlotQA](): Visual Question Answering; PlotQA; Accuracy (v1): not recorded. ## Original sources and implementation [Original paper]() [PDF]()