Research map / Image generation / Autoregressive and latent generation
SimLBR: Learning to Detect Fake Images by Learning to Detect Real Images
Aayush Dhakal, Subash Khanal, Srikumar Sastry, Jacob Arndt, Philipe Dias, Dalton Lunga, Nathan Jacobs
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 latent blending regularization in DINOv3 space (ablation) | SimLBR (DINOv3 MLP) · Chameleon (train ProGAN) | accuracy (%): 59.65 -> 84.33 | Improved |
| add latent blending regularization in DINOv3 space (ablation) | SimLBR (DINOv3 MLP) · Mean accuracy over five test settings | mean accuracy (%): 78.05 -> 88.26 | Improved |
| add latent blending regularization in DINOv3 space (ablation) | SimLBR (DINOv3 MLP) · Chameleon (train ProGAN), fake accuracy | fake accuracy (%): 7.24 -> 75.80 | Improved |
| add latent blending regularization in DINOv3 space (ablation) | SimLBR (DINOv3 MLP) · Chameleon (train SD v1.4) | accuracy (%): 87.79 -> 85.57 | Mixed |
| add latent blending regularization in DINOv3 space (ablation) | SimLBR (DINOv3 MLP) · GenImage (train SD v1.4) | accuracy (%): 92.19 -> 94.54 | Improved |
| add latent blending regularization in DINOv3 space (ablation) | SimLBR (DINOv3 MLP) · RSFake | accuracy (%): 73.90 -> 88.88 | Improved |
| add latent blending regularization in DINOv2 space (ablation) | DINOv2 MLP · Mean accuracy over test settings (DINOv2) | mean accuracy (%): 68.92 -> 68.51 | No clear effect |
| use DINOv3 instead of DINOv2 features with LBR (ablation) | SimLBR · Chameleon (train ProGAN) | accuracy (%): 55.77 -> 84.33 | Improved |
| sample blend weight from Uniform(0.5, B) with B in 0.7-0.9 instead of only low or only high values (ablation) | SimLBR · AIGC (train ProGAN) | accuracy | Improved |
| increase MLP depth beyond 4 layers (ablation) | SimLBR · AIGC (train ProGAN) | accuracy | Hurt |
| use SimLBR instead of AIDE on AIGC mean accuracy | SimLBR vs AIDE · AIGC (train ProGAN, mean over 16 test sets) | mean accuracy (%): 92.77 -> 88.40 | Hurt |
| use SimLBR instead of AIDE on Chameleon | SimLBR vs AIDE · Chameleon (train SD v1.4) | accuracy (%): 62.60 -> 85.57 | Improved |
Closest papers by meaning
- Few-Shot Learner Generalizes Across AI-Generated Image Detection (ICML 2025 · similarity 0.89)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.88)
- Stay-Positive: A Case for Ignoring Real Image Features in Fake Image Detection (ICML 2025 · similarity 0.87)
- A Bias-Free Training Paradigm for More General AI-generated Image Detection (CVPR 2025 · similarity 0.87)
- Spot the Fake: Large Multimodal Model-Based Synthetic Image Detection with Artifact Explanation (NeurIPS 2025 · similarity 0.87)
- WildFake: A Large-Scale and Hierarchical Dataset for AI-Generated Images Detection (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.
