Research map / Image generation / Autoregressive and latent generation
ADCD-Net: Robust Document Image Forgery Localization via Adaptive DCT Feature and Hierarchical Content Disentanglement
Kahim Wong, Jicheng Zhou, Haiwei Wu, Yain-Whar Si, Jiantao Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add DCT feature fused with RGB feature (ablation) | backbone · ablation subset | avg F1: .602 -> .661 | Improved |
| add focal (within-image contrastive) loss (ablation) | backbone + DCT · ablation subset | avg F1: .661 -> .682 | Improved |
| add adaptive DCT modulation by predicted alignment score (ablation) | backbone + DCT + focal · ablation subset | avg F1: .682 -> .719 | Improved |
| reconstruct DCT coefficients as well as the image in HCD (ablation) | ADCD-Net with HCD · ablation subset | avg F1: .715 -> .741 | Improved |
| add PPE on top of HCD (ablation) | ADCD-Net with HCD · ablation subset | avg F1: .741 -> .750 | Improved |
| add HCD to PPE model (ablation) | ADCD-Net with PPE · ablation subset | avg F1: .709 -> .750 | Improved |
| predicted DCT modulation score instead of fixed 1 (ablation) | ADCD-Net · Blur-7 | F1: .572 -> .707 | Improved |
| predicted DCT modulation score instead of fixed 0 (ablation) | ADCD-Net · JPEG-85 | F1: .717 -> .783 | Improved |
| fully using DCT features (score 1) versus disabling them (0) (ablation) | ADCD-Net · Noise-30 | F1: -.017 | Hurt |
| fully using DCT features (score 1) versus disabling them (0) (ablation) | ADCD-Net · Down-0.7 and JPEG-85 | F1: +.065, +.047 | Improved |
| add HCD (ablation) | ADCD-Net · text-BG bias statistic | Delta S_C: 0.937 -> 1.105 | Improved |
| add DCT traces (DTD vs TruFor) | TruFor · text-BG bias statistic | Delta S_C: 0.360 -> 0.773 | Improved |
Later papers that compared with this method
1 later accepted paper reports ADCD-Net as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- RiSE: Residual Subspace Expert for Generalizable Text-Centric Image Forgery Localization (NeurIPS 2026 · better on 4 of 5 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Toward Real-world Text Image Forgery Localization: Structured and Interpretable Data Synthesis (NeurIPS 2025 · similarity 0.86)
- Learning Discriminative Noise Guidance for Image Forgery Detection and Localization (AAAI 2024 · similarity 0.85)
- Towards Better Robustness Against Natural Corruptions in Document Tampering Localization (AAAI 2025 · similarity 0.84)
- CatmullRom Splines-Based Regression for Image Forgery Localization (AAAI 2024 · similarity 0.84)
- FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models (ICLR 2025 · similarity 0.84)
- ForensicHub: A Unified Benchmark & Codebase for All-Domain Fake Image Detection and Localization (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
