Research map / Image generation / Autoregressive and latent generation
Enabling Your Forensic Detector Know How Well It Performs on Distorted Samples
Bin Li, Haoyu Li, Haodong Li, Jiaming Zhong, Changsheng Chen, Jiangqun Ni, bo.cao
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 DACOM confidence routing instead of logit-based confidence routing | Logit Calib. Top-1 routing · Seen distortions | mean accuracy: +5.82% in mean accuracy | Improved |
| add DACOM confidence routing instead of logit-based confidence routing | Logit Calib. Top-1 routing · Seen distortions | mean AP: +8.98% in mean AP | Improved |
| add DACOM confidence routing instead of logit-based confidence routing | Logit Calib. Top-1 routing · Unseen distortions | mean accuracy: +2.62% in mean accuracy | Improved |
| add DACOM confidence routing instead of logit-based confidence routing | Logit Calib. Top-1 routing · Unseen distortions | mean AP: +8.45% in mean AP | Improved |
| use DACOM Top-1 routing instead of the single best detector | CD · Seen distortions | average accuracy: 84.85 -> 95.40 | Improved |
| route with DACOM instead of using worst-case detector | C2P* · Seen distortions (worst subset) | accuracy: 81.10 -> 90.90 | Improved |
| add DACOM confidence filtering instead of logit confidence at 40% filtered | Logit Calib. · Cross-dataset single distortion | BAcc: 73.71 -> 77.32 | Improved |
| add DACOM confidence filtering instead of logit confidence at 40% filtered | Logit Calib. · Cross-dataset single distortion | EER: 27.50 -> 19.87 | Improved |
| choose FR-IQA target SSIM vs FSIM vs DISTS vs MS-SSIM for DACOM (ablation) | DACOM · Seen distortions | average accuracy: 95.13 -> 95.40 | No clear effect |
| filter using no-reference IQA (ARNIQA) instead of DACOM (ablation) | DACOM · Cross-dataset single distortion | BAcc: 77.32 -> 69.83 | Hurt |
Closest papers by meaning
- BIAS-ID: A Framework for Analyzing Transformation Biases in AI-Generated Image Detectors (NeurIPS 2026 · similarity 0.87)
- A Bias-Free Training Paradigm for More General AI-generated Image Detection (CVPR 2025 · similarity 0.86)
- Stay-Positive: A Case for Ignoring Real Image Features in Fake Image Detection (ICML 2025 · similarity 0.85)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.84)
- Seeing What Matters: Generalizable AI-generated Video Detection with Forensic-Oriented Augmentation (NeurIPS 2025 · similarity 0.84)
- Forensic Self-Descriptions Are All You Need for Zero-Shot Detection, Open-Set Source Attribution, and Clustering of AI-generated Images (CVPR 2025 · similarity 0.83)
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.
