Research map / Image generation / Autoregressive and latent generation
ForgeLens: Data-Efficient Forgery Focus for Generalizable Forgery Image Detection
Yingjian Chen, Lei Zhang, Yakun Niu
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 WSGM to frozen CLIP-ViT (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc: 81.38 -> 94.52 | Improved |
| add WSGM to frozen CLIP-ViT (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.AP: 90.14 -> 98.12 | Improved |
| add FAFormer to frozen CLIP-ViT (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc: 81.38 -> 87.89 | Improved |
| add FAFormer on top of WSGM (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc: 94.52 -> 94.99 | Improved |
| add FAFormer on top of WSGM (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.AP: 98.12 -> 98.83 | Improved |
| use WSGM instead of Adapter (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc: 85.49 -> 94.52 | Improved |
| use WSGM instead of LoRA (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc: 86.96 -> 94.52 | Improved |
| reduce training data from 100% to 1% (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc / Avg.AP | No clear effect |
| apply Gaussian blur to train and test images (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc: 94.99 -> 81.98 | Hurt |
| apply Gaussian noise to train and test images (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc: 94.99 -> 82.64 | Hurt |
| apply JPEG compression or random cropping (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc: 94.99 -> 91.62 (JPEG), 94.99 -> 91.17 (crop) | Hurt |
| set WSGM count 12 with reduction factor 4 vs 24 WSGMs with factor 8 (ablation) | ForgeLens · UniversalFakeDetect (train ProGAN 4-class) | Avg.Acc: 90.12 -> 94.99 | Improved |
Later papers that compared with this method
2 later accepted papers report ForgeLens as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Layer Consistency Matters: Elegant Latent Transition Discrepancy for Generalizable Synthetic Image Detection (CVPR 2026 · better on 6 of 6 shared results)
- GLARE: Towards Generalizable Detection of Latent Diffusion Images with Global-Local Reconstruction Error (ECCV 2026 · better on 2 of 3 shared results)
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
- Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection (CVPR 2024 · similarity 0.89)
- Forensics-Bench: A Comprehensive Forgery Detection Benchmark Suite for Large Vision Language Models (CVPR 2025 · similarity 0.88)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · similarity 0.88)
- Forensics Adapter: Adapting CLIP for Generalizable Face Forgery Detection (CVPR 2025 · similarity 0.87)
- OmniVL-Guard: Towards Unified Vision-Language Forgery Detection and Grounding via Balanced RL (ICML 2026 · similarity 0.87)
- SUMI-IFL: An Information-Theoretic Framework for Image Forgery Localization with Sufficiency and Minimality Constraints (AAAI 2025 · similarity 0.86)
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.
