Research map / Image generation / Autoregressive and latent generation
Towards Generalizable AI-Generated Image Detection via Image-Adaptive Prompt Learning
Yiheng Li, Zichang Tan, Guoqing Xu, Zhen Lei, Xu Zhou, Yang Yang
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 MLP-based adapters (ablation) | CLIP ViT-L/14 baseline · UniversalFakeDetect | mAcc: 84.19 -> 89.02 | Improved |
| add learnable tokens (ablation) | CLIP with adapters · UniversalFakeDetect | mAcc: 89.02 -> 91.83 | Improved |
| add image-adaptive prompts (ablation) | CLIP with adapters and tokens · UniversalFakeDetect | mAcc: 91.83 -> 94.88 | Improved |
| add optimal input selection (ablation) | IAPL · UniversalFakeDetect | mAcc: 94.88 -> 95.61 | Improved |
| add general condition to forgery-specific condition (ablation) | IAPL · UniversalFakeDetect | mAcc: 92.38 -> 93.32 | Improved |
| add learnable scaling factor (ablation) | IAPL · UniversalFakeDetect | mAcc: 93.32 -> 93.51 | Improved |
| add test-time token tuning (ablation) | IAPL · UniversalFakeDetect | mAcc: 93.51 -> 94.58 | Improved |
| add confidence selection (ablation) | IAPL · UniversalFakeDetect | mAcc: 94.58 -> 94.88 | Improved |
| use averaged entropy instead of pointwise entropy loss (ablation) | IAPL test-time tuning · UniversalFakeDetect | mAcc: 95.35 -> 95.61 | Improved |
| increase test-time tuning steps T from 1 to 3 (ablation) | IAPL · UniversalFakeDetect | mAcc: 94.47 -> 94.94 | Improved |
| insert adapters evenly at [4,24,4] instead of first six layers (ablation) | IAPL adapters · UniversalFakeDetect | mAcc: 78.27 -> 89.02 | Improved |
| insert adapters into more layers ([18,24,1]) instead of six evenly (ablation) | IAPL adapters · UniversalFakeDetect | mAcc: 89.02 -> 87.74 | Hurt |
Closest papers by meaning
- A Bias-Free Training Paradigm for More General AI-generated Image Detection (CVPR 2025 · similarity 0.86)
- Test-Time Adaptive Object Detection with Foundation Model (NeurIPS 2025 · similarity 0.86)
- Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection (CVPR 2024 · similarity 0.85)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.84)
- Breaking Semantic Artifacts for Generalized AI-generated Image Detection (NeurIPS 2024 · similarity 0.84)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · 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.
