Research map / Image generation / Autoregressive and latent generation
A Debiased Reconstruction-based Framework for Training-Free Detection of AI-Generated Images
Sungik Choi, Hankook Lee, Jaehoon Lee, Robin Kim, Stanley Jungkyu Choi, Moontae Lee
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 rotation-based latent debiasing (S_latent vs f_latent) (ablation) | latent reconstruction error · GenImage | AUROC: 0.408 -> 0.667 | Improved |
| add image-level debiasing (S_image vs f_AE) (ablation) | image reconstruction error · GenImage | AUROC: 0.902 -> 0.969 | Improved |
| add image-level debiasing (S_image vs f_AE) (ablation) | image reconstruction error · LSUN-Bedroom | AUROC: 0.476 -> 0.464 | No clear effect |
| combine S_image and S_latent by multiplication (RDD) (ablation) | S_image · GenImage | AUROC: 0.969 -> 0.981 | Improved |
| combine S_image and S_latent by multiplication (RDD) (ablation) | S_image · LSUN-Bedroom | AUROC: 0.464 -> 0.940 | Improved |
| multiply instead of add scores (ablation) | RDD · GenImage | AUROC: 0.673 -> 0.981 | Improved |
| use rotation/low-pass instead of ColorJitter or RandomResizedCrop (ablation) | RDD · GenImage | AUROC: 0.907/0.841 -> 0.981 | Improved |
| vary Gaussian kernel size and sigma (ablation) | RDD · GenImage | AUROC: 0.966 - 0.982 | No clear effect |
| 45 or 180 degree rotation instead of 90 (ablation) | RDD · GenImage | AUROC: 0.983/0.984 | No clear effect |
| use RDD vs AEROBLADE | AEROBLADE · LSUN-Bedroom | AUROC: 0.476 -> 0.940 | Improved |
| use S_image for model attribution vs LatentTracer | LatentTracer · Attribution | AUROC: 50x faster; 0.292s vs 14.65s per sample | 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.
- AEROBLADE — AEROBLADE: Training-Free Detection of Latent Diffusion Images Using Autoencoder Reconstruction Error (CVPR 2024) · better on 3 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
- A Bias-Free Training Paradigm for More General AI-generated Image Detection (CVPR 2025 · similarity 0.89)
- Beyond Generation: A Diffusion-based Low-level Feature Extractor for Detecting AI-generated Images (CVPR 2025 · similarity 0.88)
- Training-free Detection of AI-generated images via Cropping Robustness (NeurIPS 2025 · similarity 0.88)
- BIAS-ID: A Framework for Analyzing Transformation Biases in AI-Generated Image Detectors (NeurIPS 2026 · similarity 0.88)
- Aligned Datasets Improve Detection of Latent Diffusion-Generated Images (ICLR 2025 · similarity 0.87)
- Breaking Semantic Artifacts for Generalized AI-generated Image Detection (NeurIPS 2024 · similarity 0.87)
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.
