Research map / Image generation / Autoregressive and latent generation
Order within Chaos: Capturing Intrinsic Energy Anomalies for AI-Manipulated Image Forgery Localization
Yiming Wang, Baiqi Wu, Qingming Li, Jiahao Chen, Tong Zhang, Shouling Ji
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 |
|---|---|---|---|
| use LAD map instead of raw RGB input (ablation) | FLAME · SID+MagicBrush validation | IoU: 0.294 -> 0.567 | Improved |
| add SAM feature adapter (ablation) | FLAME · SID+MagicBrush validation | IoU: 0.313 -> 0.567 | Improved |
| add SAM refinement on top of coarse LAD-Net mask (ablation) | FLAME · SID+MagicBrush validation | IoU: 0.379 -> 0.567 | Improved |
| use LAD operator instead of average pooling (3x3) (ablation) | LAD map · SID+MagicBrush validation | IoU: 0.502 -> 0.567 | Improved |
| use LAD operator instead of min or max (3x3) (ablation) | LAD map · SID+MagicBrush validation | IoU: 0.439/0.451 -> 0.567 | Improved |
| increase LAD window size from 3x3 to 7x7 (ablation) | LAD map · SID+MagicBrush validation | IoU: 0.567 -> 0.457 | Hurt |
| use 5x5 window with Min operator instead of 3x3 (ablation) | Min operator · SID+MagicBrush validation | IoU: 0.439 -> 0.477 | Improved |
| fine-tune with 500 EditStream Qwen-Image-Edit images plus 20% replay | FLAME · Qwen-Image-Edit | IoU: 0.321 -> 0.482 | Improved |
| fine-tune on Qwen-generated EditStream data | FLAME · NanoBanana | IoU: 0.216 -> 0.391 | Improved |
| fine-tune with EditStream data (replay) | FLAME · MagicBrush and SID (ID) | IoU: 0.538 -> 0.507 (MagicBrush); 0.580 -> 0.569 (SID) | Hurt |
| apply JPEG compression (quality 75) | FLAME · MagicBrush | IoU: 0.538 -> 0.267 | Hurt |
| apply Gaussian noise (variance 5) | FLAME · MagicBrush | IoU: 0.538 -> 0.216 | Hurt |
Closest papers by meaning
- FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models (ICLR 2025 · similarity 0.87)
- 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 0.86)
- A Lottery Ticket Hypothesis Approach with Sparse Fine-tuning and MAE for Image Forgery Detection and Localization (AAAI 2025 · similarity 0.85)
- Training-Free Image Manipulation Localization Using Diffusion Models (AAAI 2025 · similarity 0.85)
- Less Supervision, Better Generalization: Weakly Supervised Fake Region Localization in Diffusion-Edited Images (NeurIPS 2026 · similarity 0.85)
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.
