Research map / Image generation / Autoregressive and latent generation
RA-Det: Towards Universal Detection of AI-Generated Images via Robustness Asymmetry
Xinchang Wang, Yunhao Chen, Yuechen Zhang, Congcong Bian, Zihao Guo, Xingjun Ma, Hui Li
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 discrepancy branch (ablation) | RA-Det · 16 generators average | accuracy (%): 85.73 -> 93.47 | Improved |
| add low-level residual branch (ablation) | RA-Det · 16 generators average | accuracy (%): 88.62 -> 93.47 | Improved |
| add semantic branch (ablation) | RA-Det · 16 generators average | accuracy (%): 90.34 -> 93.47 | Improved |
| use median filter instead of Gaussian filter for residual map (ablation) | RA-Det residual branch · 16 generators average | accuracy (%): 90.58 -> 93.47 | Improved |
| use learned DRP instead of Gaussian noise (ablation) | RA-Det probing · 16 generators average | accuracy (%): 86.11 -> 93.47 | Improved |
| use learned DRP instead of Gaussian blur (ablation) | RA-Det probing · 16 generators average | accuracy (%): 87.07 -> 93.47 | Improved |
| add contrastive robustness loss Lra to BCE (ablation) | RA-Det · 16 generators average | accuracy (%): 90.86 -> 93.47 | Improved |
| use DINOv3 ViT-L/16 instead of CLIP ViT-L/14 backbone (ablation) | RA-Det · 16 generators average | accuracy (%): 91.85 -> 93.47 | Improved |
| vary contrastive margin gamma 0.1 to 2.0 (ablation) | RA-Det · 16 generators average | accuracy (%): 93.47 (0.1), 90.96 (0.25), 92.67 (0.5), 93.47 (1.0), 92.31 (2.0) | No clear effect |
| train on SDv1.4 or ProGAN+SDv1.4 instead of ProGAN only | RA-Det · 16 generators average | accuracy (%): 93.47 / 92.70 / 93.33 | No clear effect |
| use RA-Det learned discrepancy modeling instead of RIGID threshold scoring | RA-Det vs RIGID · 16 generators average | accuracy (%): 77.17 -> 93.47 | 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.
- FerretNet — FerretNet: Efficient Synthetic Image Detection via Local Pixel Dependencies (NeurIPS 2025) · better on 5 of 5 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
- Dual-Branch Asymmetric Discrepancy Learning Based on Fake Image Pattern-Coexistence for AI-Generated Image Detection (AAAI 2026 · similarity 0.89)
- Stay-Positive: A Case for Ignoring Real Image Features in Fake Image Detection (ICML 2025 · similarity 0.86)
- BIAS-ID: A Framework for Analyzing Transformation Biases in AI-Generated Image Detectors (NeurIPS 2026 · similarity 0.86)
- A Bias-Free Training Paradigm for More General AI-generated Image Detection (CVPR 2025 · similarity 0.85)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · similarity 0.85)
- Semi-Truths: A Large-Scale Dataset of AI-Augmented Images for Evaluating Robustness of AI-Generated Image detectors (NeurIPS 2024 · 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.
