Research map / Image generation / Autoregressive and latent generation
MIRAGE: Towards AI-Generated Image Detection in the Wild
OuCheng Huang, Manxi Lin, Jiexiang Tan, Xiaoxiong Du, Yang Qiu, Junjun Zheng, Xiangheng Kong, Yuning Jiang, Bo Zheng
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 reasoning (H-R) to Qwen2.5-VL-7B (ablation) | Qwen2.5-VL-7B · MIRAGE OOD-C | accuracy (%): 69.07 -> 72.36 | Improved |
| use analytic reasoning (A-R) instead of heuristic (H-R) (ablation) | Qwen2.5-VL-7B · MIRAGE ID | accuracy (%): 96.12 -> 93.90 | Hurt |
| add explicit reflection step (HA-R-Af) over guess-reason-answer (ablation) | MIRAGE-R1 · MIRAGE OOD-C | accuracy (%): 75.25 -> 78.13 | Improved |
| use inference-time adaptive thinking instead of always final answer (ablation) | MIRAGE-R1 · MIRAGE OOD-C | accuracy (%): 78.13 -> 78.46 | Improved |
| add reasoning (HA-R) to Qwen2.5-VL-7B (ablation) | Qwen2.5-VL-7B · MIRAGE ID | accuracy (%): 96.67 -> 97.26 | No clear effect |
| train AIDE on MIRAGE instead of GenImage | AIDE · Chameleon (fake subset) | accuracy (%): 26.80 -> 45.32 | Improved |
| train AIDE on MIRAGE instead of GenImage | AIDE · Chameleon (all) | accuracy (%): 65.77 -> 60.69 | Hurt |
| use MIRAGE-R1 instead of Effort under JPEG QF=50 | MIRAGE ID · MIRAGE ID under JPEG QF=50 | accuracy (%): 84.54 -> 95.43 | Improved |
| use MIRAGE-R1 instead of Gemini2.5-Pro | in-the-wild AIGI detection · MIRAGE OOD-C | accuracy (%): 82.30 -> 78.46 | Hurt |
Closest papers by meaning
- Is Artificial Intelligence Generated Image Detection a Solved Problem? (NeurIPS 2025 · similarity 0.87)
- BIAS-ID: A Framework for Analyzing Transformation Biases in AI-Generated Image Detectors (NeurIPS 2026 · similarity 0.86)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.86)
- WildFake: A Large-Scale and Hierarchical Dataset for AI-Generated Images Detection (AAAI 2025 · similarity 0.85)
- CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI (CVPR 2025 · similarity 0.85)
- A Sanity Check for AI-generated Image Detection (ICLR 2025 · 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.
