Research map / Image generation / Autoregressive and latent generation
Diffusion Epistemic Uncertainty with Asymmetric Learning for Diffusion-Generated Image Detection
Yingsong Huang, Hui Guo, Jing Huang, Bing Bai, Qi Xiong
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 diffusion epistemic uncertainty (ablation) | baseline classifier · GenImage | ACC: 76.1 -> 90.3 | Improved |
| add diffusion epistemic uncertainty (ablation) | baseline classifier · GenImage | AP: 90.7 -> 98.7 | Improved |
| add asymmetric learning (ablation) | baseline classifier · GenImage | ACC: 76.1 -> 81.7 | Improved |
| add asymmetric learning (ablation) | baseline classifier · GenImage BigGAN subset | ACC: 50.8 -> 72.8 | Improved |
| add asymmetric learning to DEU (ablation) | DEU classifier · GenImage | ACC: 90.3 -> 91.5 | Improved |
| add DEU to asymmetric learning (ablation) | ASL classifier · GenImage | ACC: 81.7 -> 91.5 | Improved |
| combine DEU and ASL (ablation) | baseline classifier · GenImage | ACC: 76.1 -> 91.5 | Improved |
| add DRCT diffusion-reconstructed training images (ablation) | DEUA · GenImage | ACC: 85.6 -> 91.5 | Improved |
| add DRCT diffusion-reconstructed training images (ablation) | DEUA · DRCT-2M | ACC: 90.5 -> 98.7 | Improved |
| compare to LaRE2 under cross-dataset transfer (ablation) | LaRE2 · GenImage test trained on DRCT-2M SDv1.4 | ACC: 86.2 -> 53.7 | Hurt |
| vary reconstruction step t from 100 to 400 (ablation) | DEUA · GenImage | ACC | No clear effect |
| raise real-class margin m0 beyond 0.6 (ablation) | DEUA asymmetric loss · GenImage | ACC | Hurt |
Closest papers by meaning
- Epistemic Uncertainty for Generated Image Detection (NeurIPS 2025 · similarity 0.88)
- FIRE: Robust Detection of Diffusion-Generated Images via Frequency-Guided Reconstruction Error (CVPR 2025 · similarity 0.88)
- Consistency Purification: Effective and Efficient Diffusion Purification towards Certified Robustness (NeurIPS 2024 · similarity 0.88)
- BayesDiff: Estimating Pixel-wise Uncertainty in Diffusion via Bayesian Inference (ICLR 2024 · similarity 0.88)
- Estimating Epistemic and Aleatoric Uncertainty with a Single Model (NeurIPS 2024 · similarity 0.87)
- DRCT: Diffusion Reconstruction Contrastive Training towards Universal Detection of Diffusion Generated Images (ICML 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.
