Research map / Vision-language and multimodal / Human and embodied modeling
How Do Data Owners Say No? A Case Study of Data Consent Mechanisms in Web-Scraped Vision-Language AI Training Datasets
Chung Peng Lee, Rachel Hong, Harry H. Jiang, Aster Plotnik, William Agnew, Jamie Heather Morgenstern
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 watermark-finetuned YoloV8 detector evaluated on in-the-wild CommonPool images vs standard wm-eval | watermark detectors · CommonPool | watermark detection F1: 96.66 -> 46.80 | Mixed |
| evaluate Finetuned MobileViTv2 on CommonPool images instead of wm-eval | Finetuned MobileViTv2 · CommonPool | watermark detection accuracy: 89.25 -> 30.37 | Hurt |
| evaluate Rolm-OCR on CommonPool images instead of wm-eval | Rolm-OCR · CommonPool | watermark detection precision: 99.15 -> 50.80 | Hurt |
| evaluate Gemma-3-12b-it on CommonPool images instead of wm-eval | Gemma-3-12b-it · CommonPool | watermark detection F1: 89.71 -> 52.70 | Hurt |
| use LAION-5B released watermark score at threshold 0.5 to find watermarks | LAION-5B watermark score · LAION-5B | precision / recall: precision 34.09%, recall 51.13%, AUC 0.74 | Hurt |
| scale audit from small-en to medium-en | copyright notice union · CommonPool | fraction of samples with copyright indication: 2.56% -> 2.55% | No clear effect |
| add EXIF metadata search to caption and OCR search for copyright notice | copyright notice detection · CommonPool | samples found: Caption 10,585 (0.22%) -> Caption U OCR U EXIF 123,096 (2.56%) | Improved |
| compare AI-purpose agent GPTBot to wildcard agent in robots.txt | robots.txt analysis · CommonPool | All Disallowed rate: 93.1% vs 3.9% | Mixed |
| scale robots.txt audit from small-en to medium-en | robots.txt analysis GPTBot · CommonPool | All Disallowed rate: 93.1% -> 93.0% | No clear effect |
Closest papers by meaning
- Consent in Crisis: The Rapid Decline of the AI Data Commons (NeurIPS 2024 · similarity 0.91)
- The Stronger the Diffusion Model, the Easier the Backdoor: Data Poisoning to Induce Copyright BreachesWithout Adjusting Finetuning Pipeline (ICML 2024 · similarity 0.86)
- Bridging the Data Provenance Gap Across Text, Speech, and Video (ICLR 2025 · similarity 0.86)
- CDI: Copyrighted Data Identification in Diffusion Models (CVPR 2025 · similarity 0.86)
- Blameless Users in a Clean Room: Defining Copyright Protection for Generative Models (NeurIPS 2025 · similarity 0.85)
- Position: Data Authenticity, Consent, & Provenance for AI are all broken: what will it take to fix them? (ICML 2024 · similarity 0.85)
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.
