Chart question answering (CQA) requires multimodal large language models (MLLMs) to integrate visual comprehension with logical reasoning, yet current models struggle with accurate visual grounding and coherent reasoning chains. While extrinsic chain-of-thought prompting and visual cues significantly improve performance, current MLLMs lack intrinsic visual grounded reasoning capabilities, leading to inaccurate perception and reasoning disconnected from visual evidence. To address these limitations, we propose CURV, a curriculum learning framework that develops intrinsic visual reasoning capabilities by reformulating CQA as multi-step visual grounded reasoning, where each step coordinates logical reasoning with dynamic visual grounding through spatial attention concentration. To assist model learning, we further introduce CCQA, a three-level curriculum dataset with scalable synthetic generation across diverse chart types and reasoning patterns. Our curriculum systematically progresses from basic single-operation reasoning to complex multi-chart compositional tasks. Experiments demonstrate that CURV achieves up to $\uparrow20.50\%$ improvements over baselines and is generalizable to real-world benchmarks (up to $\uparrow12.30\%$) and out-of-domain multimodal reasoning tasks (up to $\uparrow10.20\%$), validating the effectiveness of internalizing visual reasoning with dynamic grounding for enhanced chart understanding capabilities. Code is available at: https://xhguo7.github.io/CURV/.
Alex: Welcome to another episode of ResearchPod. Today, we're exploring a framework called CURV, which aims to help artificial intelligence models better understand and interpret complex charts.
Sam: So this paper is asking why even advanced AI models still struggle to read a simple bar chart correctly?
Alex: Precisely. These models often produce what researchers call "hallucinations" — they generate confident-sounding answers that aren't actually supported by what's in the image. The core issue is that they aren't looking at the right parts of the chart when they reason through a problem.
Sam: And they try to solve chart questions in one big leap, rather than breaking it down step by step?
Alex: Exactly. They treat reading a chart as a single, flat task. But think about how you actually solve a maths problem on a whiteboard. You don't just stare at the whole page — you point to specific numbers as you work through each step. Human reasoning naturally moves back and forth between thinking and looking.
Sam: So the AI is skipping that back-and-forth entirely?
Alex: Right. The researchers call that back-and-forth "Visual Grounded Reasoning." The idea is that every step of the model's logic should be anchored to actual, specific evidence on the chart — not just a general impression of what the chart looks like.
Sam: So how does CURV actually force the model to do that? Does it just tell the AI to look harder at the image?
Alex: Not quite — it reformulates the task entirely. Instead of just asking for an answer, it requires the model to generate a sequence of reasoning steps. And crucially, each step must be paired with a specific visual focus: a coordinate or region on the chart that justifies that step.
Sam: So it's a two-part process for every single step of the logic?
Alex: Yes. For every logical thought, the model must also output a location on the chart — essentially pointing to the data it's using. If it can't point to anything, that reasoning step is flagged as unreliable. It's like requiring a student to show their work, but also to circle the specific number in the textbook they're drawing from.
Sam: And this is where the "curriculum" part of CURV comes in? Are they teaching the model like a student, starting with simple charts and working up to harder ones?
Alex: That's the strategy. They built a dataset called CCQA — Curriculum Chart Question Answering — which starts with basic, single-step tasks and gradually scales up to complex problems involving multiple charts at once.
Sam: So the model learns the basic "grammar" of charts before it has to tackle the really messy, multi-panel dashboards.
Alex: Exactly. And because the training is structured that way, the model eventually internalises the skill. It stops needing external prompts to look carefully — it just does it automatically.
Sam: How is this different from something like standard "Chain-of-Thought" prompting? That also breaks problems into steps.
Alex: Good question. Chain-of-Thought breaks a problem into a sequence of text sentences, which is useful. But it doesn't force the model to actually look at the image at each step. The model can write out a perfectly logical-sounding chain of reasoning while barely engaging with the visual data at all.
Sam: So CURV is different because it interleaves those text steps with actual coordinate points on the image?
Alex: Precisely. The researchers call the overall structure a "Meta-Learning Paradigm." Think of it as a four-part sequence for every question: first, the query; then the reasoning step; then the visual location that justifies it; and finally, the answer. The image is treated as a map to be navigated, not just a picture to glance at.
Sam: That actually explains why the gains hold up on charts the model has never seen before. It's not memorising specific charts — it's learning the process of how to look.
Alex: That is the key insight. By training on the mechanism of reasoning rather than just the final answers, the model becomes genuinely more reliable. The paper reports accuracy improvements of over twenty percent in some cases.
Sam: What are the limits of this approach? It sounds like it works well, but presumably it isn't perfect.
Alex: The researchers are candid about this. Because the training data is largely synthetic — meaning it was generated to follow standard chart formats — the model can struggle with unusual or irregular charts it hasn't encountered before. A hand-drawn sketch, or a highly unconventional visualisation, might trip it up.
Sam: So it's still bound by the vocabulary of chart types it was trained on?
Alex: Exactly. The current dataset covers seven common chart types, and it's focused on English-language charts, which limits how broadly it can be applied. These are real constraints the paper acknowledges.
Sam: What does the research point toward next?
Alex: The paper suggests the natural direction is toward what they call "agentic" systems — models that can autonomously use external tools, or collaborate with other AI agents, to tackle problems that go beyond reading a single chart. The goal is a model with a kind of flexible, self-directed gaze it can aim at any visual task.
Sam: So rather than a model that's been coached on specific chart types, you'd have one that can genuinely navigate visual information the way a skilled analyst would.
Alex: That's the long-term vision. And what CURV demonstrates is that the path there runs through the process of reasoning — making sure every logical step is tied to real visual evidence — rather than simply training on more data and hoping for better scores.
Sam: It's a meaningful shift in how we think about what these models actually need to learn.
Alex: It really is. Thanks for listening to ResearchPod.