Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces ChartCynics, a dual-path framework that addresses the challenge of misleading charts by employing a "skeptical" reasoning paradigm. The framework separates perception from verification, utilizing a Diagnostic Vision Path for detecting structural anomalies and an OCR-Driven Data Path for ensuring numerical accuracy, while an Agentic Summarizer resolves cross-modal conflicts through a two-stage optimization process. Evaluations reveal that ChartCynics achieves significant performance improvements over existing models, establishing a new standard for robust chart interpretation in Vision-Language Models.
ChartCynics outperforms state-of-the-art models by nearly 29% in accurately interpreting misleading charts, showcasing the power of specialized agentic workflows.
Despite the success of Vision-Language Models (VLMs), misleading charts remain a significant challenge due to their deceptive visual structures and distorted data representations. We present ChartCynics, an agentic dual-path framework designed to unmask visual deception via a"skeptical"reasoning paradigm. Unlike holistic models, ChartCynics decouples perception from verification: a Diagnostic Vision Path captures structural anomalies (e.g., inverted axes) through strategic ROI cropping, while an OCR-Driven Data Path ensures numerical grounding. To resolve cross-modal conflicts, we introduce an Agentic Summarizer optimized via a two-stage protocol: Oracle-Informed SFT for reasoning distillation and Deception-Aware GRPO for adversarial alignment. This pipeline effectively penalizes visual traps and enforces logical consistency. Evaluations on two benchmarks show that ChartCynics achieves 74.43% and 64.55% accuracy, providing an absolute performance boost of ~29% over the Qwen3-VL-8B backbone, outperforming state-of-the-art proprietary models. Our results demonstrate that specialized agentic workflows can grant smaller open-source models superior robustness, establishing a new foundation for trustworthy chart interpretation.