Research map / Image generation / Autoregressive and latent generation
Bi-Level Optimization for Self-Supervised AI-Generated Face Detection
Mian Zou, Nan Zhong, Baosheng Yu, Yibing Zhan, Kede Ma
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 |
|---|---|---|---|
| train the text encoder (not frozen) with joint embedding (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 81.65 -> 88.01 | Improved |
| add joint image-text embedding (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 85.93 -> 88.01 | Improved |
| add bi-level optimization of task weights (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 86.20 -> 88.01 | Improved |
| add pairwise ranking of ordinal EXIF tags (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 81.52 -> 88.01 | Improved |
| add focal weighting for categorical EXIF tags (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 81.87 -> 88.01 | Improved |
| use coarse-grained manipulation detection as primary task instead of binary real vs AI classification (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 86.75 -> 88.01 | Improved |
| use coarse-grained instead of fine-grained primary task (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 87.11 -> 88.01 | Improved |
| learn task weights instead of equal weighting (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 87.84 -> 88.01 | Improved |
| add manipulation detection tasks to EXIF tasks (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 81.80 -> 87.03 | Improved |
| add EXIF tasks to manipulation tasks (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 86.49 -> 88.01 | Improved |
| add fine-grained manipulation task (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 87.03 -> 88.01 | Improved |
| keep only top six weighted pretext tasks (ablation) | BLADES-OC · Cross-generator one-class detection | mAcc: 88.01 -> 87.62 | No clear effect |
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.
- AEROBLADE — AEROBLADE: Training-Free Detection of Latent Diffusion Images Using Autoencoder Reconstruction Error (CVPR 2024) · 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
- FSFM: A Generalizable Face Security Foundation Model via Self-Supervised Facial Representation Learning (CVPR 2025 · similarity 0.85)
- GIF: Generative Inspiration for Face Recognition at Scale (CVPR 2025 · similarity 0.84)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · similarity 0.84)
- AugGen: Synthetic Augmentation using Diffusion Models Can Improve Recognition (NeurIPS 2025 · similarity 0.84)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.84)
- A Sanity Check for AI-generated Image Detection (ICLR 2025 · 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.
