Research map / Image generation / Text-to-image diffusion
AnyPattern: Towards In-context Image Copy Detection
Wenhao Wang, Yifan Sun, Zhentao Tan, Yi Yang
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 |
|---|---|---|---|
| train on 90-pattern AnyPattern instead of SmallPattern | Baseline · novel patterns | uAP: 16.18 -> 42.84 | Improved |
| add in-context ImageStacker prompting | Baseline · novel patterns | uAP: 42.84 -> 56.65 | Improved |
| add in-context ImageStacker prompting | Baseline · base patterns | uAP: 79.37 -> 83.56 | Improved |
| add ImageStacker when training on SmallPattern with ImageNet | Baseline · novel patterns | uAP: 16.18 -> 15.28 | No clear effect |
| train ImageStacker on AnyPattern without ImageNet pretraining (ablation) | ImageStacker · novel patterns | uAP: 40.52 -> 53.53 | Improved |
| use in-context learning instead of fine-tuning on 2,400 images | ImageStacker · novel patterns | uAP: 41.15 -> 56.65 | Improved |
| fine-tune on 200,000 novel-pattern images instead of in-context | ImageStacker · base patterns | uAP: 83.56 -> 20.06 | Hurt |
| stack pair on channel dimension instead of concatenating (ImageCombiner) (ablation) | ImageStacker · novel patterns | uAP: 48.47 -> 56.65 | Improved |
| stack pair on channel dimension instead of adding to replica (ImageAdder) (ablation) | ImageStacker · novel patterns | uAP: 25.07 -> 56.65 | Improved |
| feature-level integration of pair instead of image-level stacking (ablation) | ImageStacker · novel patterns | uAP: 37.42 -> 56.65 | Improved |
| retrieve prompt by pattern retrieval instead of random (ablation) | ImageStacker · novel patterns | uAP: 50.42 -> 56.65 | Improved |
| use ground-truth pattern pair instead of retrieved pair (ablation) | ImageStacker · novel patterns | uAP: 56.65 -> 56.99 | No clear effect |
Closest papers by meaning
- Image Copy Detection for Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Pattern Analogies: Learning to Perform Programmatic Image Edits by Analogy (CVPR 2025 · similarity 0.86)
- Common Inpainted Objects In-N-Out of Context (CVPR 2026 · similarity 0.85)
- Zero-shot Image Editing with Reference Imitation (NeurIPS 2024 · similarity 0.85)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.84)
- Context-I2W: Mapping Images to Context-Dependent Words for Accurate Zero-Shot Composed Image Retrieval (AAAI 2024 · 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.
