Lune

CVPR Findings2026

Towards Generalization of Scene Text Tampering Localization via Causal Invariance

Huiru Shao, Bin Dong, Kaizhu Huang, Xiaowei Huang, Qiufeng Wang

Abstract

Image tampering localization models have achieved remarkable progress. However, evaluated on tampered scene texts, current models expose extremely limited generalization against divergent data shifts in real-world environments. We contend that this limitation arises since the model captures spurious correlations rather than the causal invariance between learned forensic representations and the confounders (e.g., content and tampering types). To tackle it, we propose a paradigm that includes confounding environment generation and invariant representation optimization. Based on this, we conduct confounding environment generation including confounder qualification, diversification and interventional data synthesis. Meanwhile, we leverage the regularization term for minimizing the variance of risks across observable and various confounding environments to guarantee causal invariance. We argue that our proposed paradigm is compatible with any architecture capable of disentangling semantic and forensic information which are elements for confounder qualification, such as the foundation model with light-weight adapters. To verify this, we evaluate the proposed method on the benchmark tampered scene text images datasets, Tampered-IC13 and OSTF, which are tampered with various generative methods and divergent contents on different image domains. Various experiments are compared to indicate that the proposed method is promising for generalization improvement.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Scalable Diffusion Models with Transformers cites High-Resolution Image Synthesis with Latent Diffusion Models
  • Scalable Diffusion Models with Transformers cites Learning Transferable Visual Models From Natural Language Supervision
  • TextDiffuser: Diffusion Models as Text Painters cites High-Resolution Image Synthesis with Latent Diffusion Models
  • TextDiffuser: Diffusion Models as Text Painters cites Learning Transferable Visual Models From Natural Language Supervision
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on21

  1. Learning Transferable Visual Models From Natural Language SupervisionAlec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh et al.ICML 2021 · 47,906 citations
  2. High-Resolution Image Synthesis with Latent Diffusion ModelsRobin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser et al.CVPR 2022 · 13,123 citations
  3. Scalable Diffusion Models with TransformersWilliam Peebles, Saining XieICCV 2023 · 5,568 citations
  4. Out-of-Distribution Generalization via Risk Extrapolation (REx)David Krueger, Ethan Caballero, Jörn-Henrik Jacobsen, Amy Zhang et al.ICML 2021 · 1,163 citations
  5. TextDiffuser: Diffusion Models as Text PaintersJingye Chen, Yupan Huang, Tengchao Lv, Lei Cui et al.NeurIPS 2023 · 290 citations

Related papers

  1. Revisiting Tampered Scene Text Detection in the Era of Generative AIChenfan Qu, Yiwu Zhong, Fengjun Guo, Lianwen JinAAAI 2025 · 20 citations
  2. DITL2: Dual-Stage Invariance Transfer Learning for Generalizable Document Image Tampering LocalizationSongze Li, Yunfei Guo, Shen Chen, Bin Li et al.ACM MM 2025
  3. Detective SAM: Adaptive AI-Image Forgery LocalizationGert Lek, Nicolas van Schaik, Chaoyi Zhu, Pin-Yu Chen et al.ICLR 2026
  4. Towards Better Robustness Against Natural Corruptions in Document Tampering LocalizationHuiru Shao, Kaizhu Huang, Wei Wang, Xiaowei Huang et al.AAAI 2025 · 1 citation
  5. Omni-IML: Towards Unified Interpretable Image Manipulation LocalizationChenfan Qu, Yiwu Zhong, Fengjun Guo, Lianwen JinICLR 2026 · 5 citations
Dusk over the sea between two cliffs drawn in fine vertical lines