Research map / Image generation / Autoregressive and latent generation
PPM-CLIP: Probabilistic Prompt Modeling for Generalizable AI-Generated Image Detection
Xinyuan Wang, Yingxin Lai, Zhiming Luo, Zhihui Liu
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 Prompt Flow Module (probabilistic prompts) (ablation) | PPM-CLIP · GenImage mean | mAcc: 92.1 -> 99.6 | Improved |
| add PWCL patch-wise contrastive learning (ablation) | PPM-CLIP · GenImage mean | mAcc: 80.0 -> 99.6 | Improved |
| use frequency-based patch selection instead of random (ablation) | PWCL · GenImage mean | mAcc: 95.0 -> 99.6 | Improved |
| add Prompt Repository (ablation) | PPM-CLIP · GenImage mean | mAcc: 88.9 -> 99.6 | Improved |
| increase Monte Carlo samples from 2 to 10 (ablation) | PPM-CLIP · GenImage mean | mAcc: 98.6 -> 100.0 (FPS 50.51 -> 18.76) | Improved |
| increase Monte Carlo samples from 4 to 10 (ablation) | PPM-CLIP · GenImage mean | mAcc: 99.6 -> 100.0 | Improved |
| raise KL weight from 0.001 to 0.1 (ablation) | PPM-CLIP · GenImage mean | mAcc: 99.6 -> 97.1 | Hurt |
| lower KL weight from 0.001 to 0.0001 (ablation) | PPM-CLIP · GenImage mean | mAcc: 99.6 -> 92.3 | Hurt |
| use PPM-CLIP vs best baseline | PPM-CLIP · DRCT average | accuracy: 91.35 -> 95.72 (DRCT) | Improved |
| use PPM-CLIP vs UnivFD on inpainting (DR variants) | PPM-CLIP · DRCT DR variants | accuracy: ~51% -> 87.80% (SDv1/v2-DR) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LOTA — LOTA: Bit-Planes Guided AI-Generated Image Detection (ICCV 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- Any-Shift Prompting for Generalization over Distributions (CVPR 2024 · similarity 0.86)
- Aggregate-and-Adapt Natural Language Prompts for Downstream Generalization of CLIP (NeurIPS 2024 · similarity 0.85)
- CLIPTTA: Robust Contrastive Vision-Language Test-Time Adaptation (NeurIPS 2025 · similarity 0.85)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.85)
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 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.
