AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

DGS-Net: Distillation-Guided Gradient Surgery for CLIP Fine-Tuning in AI-Generated Image Detection

Jiazhen Yan, Ziqiang Li, Fan Wang, Boyu Wang, Ziwen He, Zhangjie Fu

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Generalizable fake image detection · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add Orthogonal Suppression to LoRA-tuned CLIP (ablation)CLIP-LoRA detector · AIGCDetectBenchmAcc: 79.9 -> 88.0Improved
add Prior Alignment to LoRA-tuned CLIP (ablation)CLIP-LoRA detector · AIGCDetectBenchmAcc: 79.9 -> 83.3Improved
add Orthogonal Suppression to LoRA-tuned CLIP (ablation)CLIP-LoRA detector · AIGIBenchaverage accuracy: 52.4 -> 61.7Improved
add Prior Alignment to LoRA-tuned CLIP (ablation)CLIP-LoRA detector · AIGIBenchaverage accuracy: 52.4 -> 58.0Improved
combine Orthogonal Suppression and Prior Alignment (ablation)CLIP-LoRA detector · AIGCDetectBenchmAcc: 79.9 -> 97.6Improved
combine Orthogonal Suppression and Prior Alignment (ablation)CLIP-LoRA detector · AIGIBenchaverage accuracy: 52.4 -> 81.6Improved
naive LoRA fine-tuning of CLIP vs frozen CLIP, linear probingCLIP image encoder · AIGIBench linear probingaccuracy: 71.5 -> 52.4Hurt
fine-tune with DGS-Net instead of naive LoRA, linear probingCLIP image encoder · AIGIBench linear probingaccuracy: 52.4 -> 76.7Improved
fine-tune with DGS-Net instead of naive LoRA, linear probingCLIP image encoder · AIGCDetectBench linear probingaccuracy: 82.4 -> 93.3Improved
DGS-Net vs retrained C2P-CLIP with same inferenceCLIP-based detector · AIGCDetectBenchmAcc: +15.2% mAccImproved
DGS-Net under JPEG QF=75DGS-Net · AIGCDetectBench robustnessaccuracy: 88.9 (QF=95) -> 82.4 (QF=75)Mixed
Gaussian blur sigma 0.5 to 1.5DGS-Net · AIGCDetectBench robustnessaccuracy: 92.7 -> 89.9No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy