AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

WMVLM: Evaluating Diffusion Model Image Watermarking via Vision-Language Models

Zijin Yang, Yu Sun, Kejiang Chen, Jiawei Zhao, Jun Jiang, Weiming Zhang, Nenghai Yu

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Robust diffusion watermarking · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
train Qwen3-VL-8B with WMVLM three-stage pipeline instead of using zero-shot VLMsVLM watermark evaluator · Residual watermarks, SD v2.1quality PLCC / SRCC, security accuracy: RobustWide-like zero-shot VINE 0.067/0.052/0.500 (GPT-5) -> 0.983/0.986/1.000Improved
add explainability cold start (S2) after pre-training (S1) (ablation)WMVLM · RobustWide / VINE / RingIDPLCC and accuracy: RobustWide PLCC 0.967 -> 0.869Hurt
add GRPO (S3) after S1+S2 (ablation)WMVLM · RivaGAN and HiDDeN residualPLCC / SRCC: RivaGAN PLCC 0.185 -> 0.885Improved
add GRPO (S3) after S1+S2 (ablation)WMVLM · Semantic Losslessquality/security accuracy: 0.844 -> 0.910Improved
add pre-training stage S1 before cold start and GRPO (ablation)WMVLM · Semantic watermarksaccuracy: Lossless 0.000/0.000 -> 0.910/0.910Improved
add cold start S2 before GRPO (instead of GRPO alone) (ablation)WMVLM · All methodsformat reward / accuracyImproved
evaluate on MS-COCO prompts instead of SD prompts (cross-dataset)WMVLM · RingID semanticaccuracy: 0.852/0.852 -> 0.746/0.746Hurt
evaluate on SD v1.4 instead of SD v2.1 (cross-model)WMVLM · Stable Signature residualPLCC: 0.981 -> 0.368Hurt
evaluate on unseen watermarking methods (cross-method)WMVLM · GaussMarker, T2SMark, Oriaccuracy: 0.956/0.956, 0.930/0.930, 0.874/0.874No clear effect
scale zero-shot open-source VLM size (Qwen3-VL 4B to 32B)Qwen3-VL · Residual quality scoringPLCCNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy