Research map / Label-efficient and robust learning / Self-supervised and contrastive
Tracing Copied Pixels and Regularizing Patch Affinity in Copy Detection
Yichen Lu, Siwei Nie, Minlong Lu, Xudong Yang, Xiaobo Zhang, Peng Zhang
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 CopyNCE loss (ablation) | descriptor (ViT-S) · DISC21 dev II | muAP: 68.9 -> 70.5 | Improved |
| add CopyNCE loss without global hard negative mining (ablation) | descriptor (ViT-S) · DISC21 dev II | muAP: 57.7 -> 61.8 | Improved |
| add CopyNCE loss to ResNet-50 descriptor (ablation) | descriptor (R50, 448x448) · DISC21 dev II | muAP: 67.8 -> 68.8 | Improved |
| add CopyNCE loss on top of BCE (ablation) | matcher (ViT-S) · DISC21 dev II | muAP: 70.9 -> 83.5 | Improved |
| use NCE instead of cosine loss for patch affinity (ablation) | descriptor · DISC21 dev II | muAP: 68.6 -> 70.5 | Improved |
| use proportional-overlap prior (gamma=3) instead of uniform (gamma=0) (ablation) | descriptor · DISC21 dev II | muAP: 67.9 -> 70.5 | Improved |
| use proportional-overlap prior (gamma=1) instead of uniform (gamma=0) (ablation) | matcher · DISC21 dev II | muAP: 82.5 -> 83.5 | Improved |
| use PixTrace instead of FeatNN+LocNN matching with NCE (ablation) | matcher · DISC21 dev II | muAP: 80.8 -> 83.5 | Improved |
| use PixTrace instead of FeatNN+LocNN matching with cosine (ablation) | descriptor · DISC21 dev II | muAP: 68.5 -> 70.5 | Improved |
| use 4 fusion layers instead of 2 (enc-8-fus-4 vs enc-10-fus-2) (ablation) | matcher · DISC21 dev II | muAP: 79.4 -> 83.5 | Improved |
| add global hard negative mining (ablation) | descriptor baseline · DISC21 dev II | muAP: 57.7 -> 68.9 | Improved |
| add local crops ensembling | CopyNCE matcher ViT-S 224 · DISC21 dev II | muAP: 83.5 -> 87.4 | Improved |
Closest papers by meaning
- A Benchmark and Asymmetrical-Similarity Learning for Practical Image Copy Detection (AAAI 2023 · similarity 0.88)
- AnyPattern: Towards In-context Image Copy Detection (IJCV 2026 · similarity 0.86)
- Image Copy Detection for Diffusion Models (NeurIPS 2024 · similarity 0.86)
- PixPerfect: Seamless Latent Diffusion Local Editing with Discriminative Pixel-Space Refinement (NeurIPS 2025 · similarity 0.84)
- Transfer Learning of Real Image Features with Soft Contrastive Loss for Fake Image Detection (AAAI 2025 · similarity 0.84)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.83)
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.
