Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Fleet: Few Shots Lead Effective AI-generated Image Detection
Jiaan Wang, Sirui Liu, Yu Li, Kaiyuan Yang, Juan Cao, Sheng Tang
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 avoidance loss (ablation) | Fleet · Treasure (8-subset ablation) | mean accuracy: 89.25 -> 93.26 | Improved |
| add coverage loss (ablation) | Fleet · Treasure (8-subset ablation) | mean accuracy: 92.40 -> 93.26 | Improved |
| add all subspace losses (orth, cov, avoid) (ablation) | Fleet · Treasure (8-subset ablation) | mean accuracy: 89.22 -> 93.26 | Improved |
| use subspace routing instead of plain DINOv3+LoRA (ablation) | Fleet · Treasure (8-subset ablation) | mean accuracy: 91.35 -> 93.26 | Improved |
| use dynamic input-dependent addresses instead of static prototypes (ablation) | Fleet · Treasure (8-subset ablation) | mean accuracy: 92.21 -> 93.26 | Improved |
| structured MoE-style initialization (ablation) | Fleet · Treasure (8-subset ablation) | mean accuracy: 93.26 -> 93.53 | No clear effect |
| LoRA rank 8 vs 2 / 32 (ablation) | Fleet · Treasure (8-subset ablation) | mean accuracy: 91.25 (rank 2), 93.26 (rank 8), 91.46 (rank 32) | No clear effect |
| shrink replay set to 20 images (ablation) | Fleet · Treasure (8-subset ablation) | mean accuracy: -13.92% accuracy | Hurt |
| shrink replay set to 20 images (pretrain val) (ablation) | Fleet · Treasure (8-subset ablation) | pretrain val accuracy: above 99.9% | No clear effect |
| use 8 subspaces (ablation) | Fleet · Treasure (8-subset ablation) | mean accuracy | Improved |
| increase few-shot support from 1 to 10 shots | Fleet · AIGIBench-13 | AI accuracy: 92.12 -> 97.68 (AI acc) | Improved |
| replace FSD with Fleet (anti-forgetting replay) | Fleet · Treasure (64 generators) | pretrain val accuracy: 74.81 -> 99.96 (Pretrain Val, 10-shot) | Improved |
Closest papers by meaning
- Less is More: Fewer Tokens and Blocks Make AIGI Detection Faster and More Generalizable (NeurIPS 2026 · similarity 0.88)
- Is Artificial Intelligence Generated Image Detection a Solved Problem? (NeurIPS 2025 · similarity 0.86)
- Few-Shot Learner Generalizes Across AI-Generated Image Detection (ICML 2025 · similarity 0.85)
- Manifold Induced Biases for Zero-shot and Few-shot Detection of Generated Images (ICLR 2025 · similarity 0.84)
- Breaking Semantic Artifacts for Generalized AI-generated Image Detection (NeurIPS 2024 · similarity 0.84)
- Training-free Detection of AI-generated images via Cropping Robustness (NeurIPS 2025 · similarity 0.84)
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.
