Research map / Detection and segmentation / Open-set and few-shot detection
Penny-Wise and Pound-Foolish in AI-Generated Image Detection
Yabin Wang, Zhiwu Huang, Zhou Su, Adam Prügel‐Bennett, Xiaopeng Hong
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 the class-aware binary term (lambda2=1) to the semantic-preserving term (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 72.46 -> 80.83 | Improved |
| add the semantic-preserving term (lambda1=1) to the class-aware binary term (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 76.20 -> 80.83 | Improved |
| add both loss terms vs neither (lambda 0,0 -> 1,1) (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 71.15 -> 80.83 | Improved |
| use lambda 1,1 instead of 0.5,0.5 (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 76.73 -> 80.83 | Improved |
| increase lambdas to 2,2 (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 80.83 -> 76.09 | Hurt |
| use ViT-L/14 instead of ViT-B/16 backbone (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 75.42 -> 80.83 | Improved |
| use OpenAI instead of OpenCLIP ViT-L/14 (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 79.69 -> 80.83 | No clear effect |
| increase prompt length from 1 to 16 (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 68.38 -> 80.83 | Improved |
| increase prompt depth from 1 to 8 (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 78.30 -> 80.83 | Improved |
| train with 20 classes instead of 2 (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 71.56 -> 80.83 | Improved |
| train with 20 classes instead of 16 (ablation) | PoundNet · ForenSynths/DIF/Ojha average | ACC: 76.71 -> 80.83 | Improved |
Closest papers by meaning
- D^3: Scaling Up Deepfake Detection by Learning from Discrepancy (CVPR 2025 · similarity 0.87)
- Exploring Unbiased Deepfake Detection via Token-Level Shuffling and Mixing (AAAI 2025 · similarity 0.87)
- Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection (AAAI 2025 · similarity 0.86)
- PRPO: Paragraph-level Policy Optimization for Vision-Language Deepfake Detection (ICML 2026 · similarity 0.85)
- WildFake: A Large-Scale and Hierarchical Dataset for AI-Generated Images Detection (AAAI 2025 · similarity 0.84)
- Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning (ICLR 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.
