Research map / Image generation / Autoregressive and latent generation
OmniAID: Decoupling Semantic and Artifacts for Universal AI-Generated Image Detection in the Wild
Yuncheng Guo, Junyan Ye, Chenjue Zhang, Hengrui Kang, Haohuan Fu, Conghui He, Weijia Li
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 Universal Artifact Expert to the semantic experts (ablation) | OmniAID · Chameleon (trained on GenImage SD v1.4) | accuracy (%): 66.07 -> 77.35 | Improved |
| add Object/Scene semantic expert e1 (ablation) | OmniAID · Chameleon (trained on GenImage SD v1.4) | accuracy (%): 68.11 -> 77.35 | Improved |
| add Human/Animal semantic expert e0 (ablation) | OmniAID · Chameleon (trained on GenImage SD v1.4) | accuracy (%): 70.80 -> 77.35 | Improved |
| combine both semantic experts with the artifact expert instead of a single expert (ablation) | OmniAID · GenImage (trained on SD v1.4) | accuracy (%): 84.38 -> 95.94 | Improved |
| train on Mirage-Train instead of GenImage-SD v1.4 | OmniAID · Mirage-Test | accuracy (%): 51.10 -> 88.39 | Improved |
| train on Mirage-Train instead of GenImage-SD v1.4 | OmniAID · Chameleon | accuracy (%): 77.35 -> 91.42 | Improved |
| train on Mirage-Train instead of GenImage-SD v1.4 | Effort · GenImage | accuracy (%): 91.10 -> 85.00 | Hurt |
| train on Mirage-Train instead of GenImage-SD v1.4 | AIDE · Mirage-Test | accuracy (%): 31.25 -> 76.78 | Improved |
| train on Mirage-Train instead of GenImage-SD v1.4 | OmniAID · GenImage | accuracy (%): 95.94 -> 97.24 | Improved |
| use SVD initialization instead of standard LoRA initialization (ablation) | OmniAID experts · OOD benchmarks | accuracy (%) | Improved |
Closest papers by meaning
- Breaking Semantic Artifacts for Generalized AI-generated Image Detection (NeurIPS 2024 · similarity 0.87)
- OmniVL-Guard: Towards Unified Vision-Language Forgery Detection and Grounding via Balanced RL (ICML 2026 · 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 0.85)
- WildFake: A Large-Scale and Hierarchical Dataset for AI-Generated Images Detection (AAAI 2025 · similarity 0.85)
- A Bias-Free Training Paradigm for More General AI-generated Image Detection (CVPR 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.
