Research map / Image generation / Autoregressive and latent generation
MCID: Multi-aspect Copyright Infringement Detection for Generated Images
Chuanwei Huang, Zexi Jia, Hongyan Fei, Yeshuang Zhu, Zhiqiang Yuan, Ying Deng, Jiapei Zhang, Xiaoyue Duan, Jinchao Zhang, Jie Zhou
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 multi-aspect similarity scores to the VLM input (ablation) | HIDM (GPT-4o) · MCID testset | accuracy (%): 23.27 -> 72.46 | Improved |
| add similarity scores to the VLM input (ablation) | HIDM (GPT-4V) · MCID testset | accuracy (%): 22.03 -> 69.49 | Improved |
| add similarity scores to the VLM input (ablation) | HIDM (GPT-4o mini) · MCID testset | accuracy (%): 18.64 -> 64.41 | Improved |
| use GPT-4o mini instead of GPT-4o as the VLM (ablation) | HIDM · MCID testset | accuracy (%): 72.46 -> 64.41 | Hurt |
| decouple features by attribute with three separate encoders instead of one aggregated network (ablation) | HIDM · MCID testset | accuracy (%) | Improved |
| replace VLM decision-making with an FFN classifier (ablation) | HIDM · MCID testset | accuracy (%) | Hurt |
| aggregate content, style, structure encoders instead of a single encoder (ablation) | HIDM encoder · MCID retrieval | µAP: 0.502 -> 0.671 | Improved |
| use HIDM instead of Gowthami [42] for infringement classification | HIDM · MCID testset | average accuracy (%): 42.80 -> 72.46 | Improved |
| use HIDM instead of Gowthami [42] on structural infringement | HIDM · MCID testset | accuracy (%): 18.64 -> 64.41 | Improved |
| use HIDM instead of Gowthami [42] on CPDM | HIDM · CPDM | accuracy (%): 48.32 -> 64.17 | Improved |
| use HIDM VLM-based pipeline instead of Wang [49] for IP infringement | HIDM · MCID testset IP | accuracy (%): 62.71 -> 59.32 | No clear effect |
| add multi-aspect LoRA encoders on a shared backbone | HIDM · inference cost | GFLOPs: 22.30 -> 22.39 | No clear effect |
Closest papers by meaning
- Bridging the Copyright Gap: Do Large Vision-Language Models Recognize and Respect Copyrighted Content? (AAAI 2026 · similarity 0.84)
- CGI-DM: Digital Copyright Authentication for Diffusion Models via Contrasting Gradient Inversion (CVPR 2024 · similarity 0.84)
- The Stronger the Diffusion Model, the Easier the Backdoor: Data Poisoning to Induce Copyright BreachesWithout Adjusting Finetuning Pipeline (ICML 2024 · similarity 0.84)
- How to Evaluate and Mitigate IP Infringement in Visual Generative AI? (ICML 2025 · similarity 0.84)
- Tracking the Copyright of Large Vision-Language Models through Parameter Learning Adversarial Images (ICLR 2025 · similarity 0.83)
- Disguised Copyright Infringement of Latent Diffusion Models (ICML 2024 · 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.
