Reproduce “VizDefender: Unmasking Visualization Tampering through Proactive Localization and Intent Inference” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sicheng Song, Yanjie Zhang, Zixin Chen, Huamin Qu, Changbo Wang, Chenhui Li
VizDefender is an end-to-end framework designed to detect and interpret maliciously tampered data visualizations by combining a semi-fragile watermark for precise localization and an MLLM-based module for inferring manipulation intent. The framework demonstrates high accuracy in identifying altered regions and explaining tampering methods, while maintaining the visual quality of protected charts.
Key insights
- A formative study identified nine common visualization tampering types, categorizing them into 'data manipulation' and 'visual encoding manipulation,' which formed the basis for the MLLM's rule-based reasoning.
- Invertible Neural Networks, combined with posterior estimation, enable a semi-fragile watermark that is sensitive to malicious edits yet robust to benign operations, a balance crucial for visualization integrity.
- Using contour lines as visual prompts and applying component-to-method mapping rules for MLLM inference significantly mitigates hallucination and improves accuracy in semantically interpreting visual tampering.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.