Elamur Research⌕
☰
home / papers / 2026 / SSR-A: Spatial- and Semantic-Aware Instructions and Curriculum Reinforcement for Advertisement Compliant Rectification

SSR-A: Spatial- and Semantic-Aware Instructions and Curriculum Reinforcement for Advertisement Compliant Rectification

Paper · 2026 · Cao Te, Mengge Xue, Zhenyu Hu et al.

Summary

Ad platforms reject millions of non-compliant image ads every day, and fixing them by hand does not scale. Existing safety-oriented editing methods often over-edit and erase the advertiser's commercial message. SSR-A addresses this by leaving the diffusion editor frozen and training an MLLM (Qwen3-VL-8B) to convert a violation policy into a precise editing instruction.

To overcome scarce training data, the authors use Gemini3-Pro to synthesize two kinds of candidate instructions: spatially grounded ones with bounding boxes, and purely semantic ones. Candidates are filtered by MLLM quality scoring and the platform's violation detector, and the final choice is the one with the lowest LPIPS distance to the source image. The model is then refined with curriculum GRPO. It first optimizes a compliance-favoring exponential reward, then moves to a linear reward that restores visual fidelity.

SSR-A outperforms SFT and single-stage RL variants offline, is preferred in a user study, and improves adoption and compliance in a live A/B test. This makes it a practical tool for automated ad repair.

Key points

  • Ablation results (compliance rate / LPIPS / overall score):
    • SFT: 83.56% / 0.1911 / 74.5
    • Linear reward (PPR) only: 84.33% / 0.1753 / 76.3
    • Exponential reward (CPE) only: 86.89% / 0.1846 / 77.8
    • Two-stage curriculum: 88.00% / 0.1781 / 79.5
  • Online A/B test: run for 3 days on 20% of live traffic. Against the SFT baseline, SSR-A raised advertiser adoption by 8.6% and compliance by 6.4%.
  • User study: five professional advertisers made blind pairwise judgments on 100 rectifications. SSR-A won about 2.2× as often as SFT, with ties in 68.6% of cases.
  • Editor setup: the main experiments use a proprietary ad-optimized diffusion model as the frozen editor. Checks on open-source editors are reported in the appendix.
  • Released dataset: an evaluation set of AI-synthesized images only, intended for non-commercial use, available on Hugging Face as rosalie12345/AdRectification_AIGC.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

Cite

Cite as Cao Te, Mengge Xue, Zhenyu Hu, Yuan Chen, Liqun Liu, Peng Shu, Huan Yu and Jie Jiang. 2026. SSR-A: Spatial- and Semantic-Aware Instructions and Curriculum Reinforcement for Advertisement Compliant Rectification. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.23

@inproceedings{te2026ssr,
  title = {{SSR-A: Spatial- and Semantic-Aware Instructions and Curriculum Reinforcement for Advertisement Compliant Rectification}},
  author = {Cao Te, Mengge Xue, Zhenyu Hu, Yuan Chen, Liqun Liu, Peng Shu, Huan Yu and Jie Jiang},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.23},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.23}
}

References

35 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Barween Al Kurdi and Muhammad Turki Alshurideh. 2025. The effect of social media influencer traits on consumer purchasing decisions for keto products: examining the moderating influence of advertising repetition. Journal of Marketing Communications, 31(4):422–443. find
  2. Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhaohai Li, Mingsheng Li, and 45 others. 2025. Qwen3-vl technical report. arXiv prepri…
  3. Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. 2009. Curriculum learning. In Proceedings of the 26th annual international conference on machine learning, pages 41–48. ACM.
  4. Tim Brooks, Aleksander Holynski, and Alexei A Efros. 2023. Instructpix2pix: Learning to follow image editing instructions. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 18392–18402. find
  5. Colin Campbell, Sean Sands, Brent McFerran, and Alexis Mavrommatis. 2025. Diversity representation in advertising. Journal of the Academy of Marketing Science, 53(2):588–616. find
  6. SiXiang Chen, Jianyu Lai, Jialin Gao, Tian Ye, Haoyu Chen, Hengyu Shi, Shitong Shao, Yunlong Lin, Song Fei, Zhaohu Xing, and 1 others. 2025. Poster-craft: Rethinking high-quality aesthetic poster generation in a unified framework. arXiv preprint arXiv:2506.10741. — [arXiv:2506.10741](https://arxiv.org/abs/2506.10741) find
  7. Xi Chen, Lianghua Huang, Yu Liu, Yujun Shen, Deli Zhao, and Hengshuang Zhao. 2024. Anydoor: Zero-shot object-level image customization. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6593–6602. find
  8. Jiaxin Deng, Shiyao Wang, Kuo Cai, Lejian Ren, Qigen Hu, Weifeng Ding, Qiang Luo, and Guorui Zhou. 2025. Onerec: Unifying retrieve and rank with generative recommender and iterative preference alignment. arXiv preprint arXiv:2502.18965. — [arXiv:2502.18965](https://arxiv.org/abs/2502.18965) find
  9. Tsu-Jui Fu, Wenze Hu, Xianzhi Du, William Yang Wang, Yinfei Yang, and Zhe Gan. 2023. Guiding instruction-based image editing via multimodal large language models. arXiv preprint arXiv:2309.17102. — [arXiv:2309.17102](https://arxiv.org/abs/2309.17102) find
  10. Tsu-Jui Fu, Wenze Hu, Xianzhi Du, William Yang Wang, Yinfei Yang, and Zhe Gan. 2024. Guiding Instruction-based Image Editing via Multimodal Large Language Models. In International Conference on Learning Representations (ICLR). find
  11. Google. 2025. A new era of intelligence with gemini 3. https://blog.google/products/gemini/ gemini-3. ↗
  12. Shunan Guo, Zhuochen Jin, Fuling Sun, Jingwen Li, Zhaorui Li, Yang Shi, and Nan Cao. 2021. Vinci: An intelligent graphic design system for generating advertising posters. Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems. find
  13. Zipeng Guo, Lichen Ma, Xiaolong Fu, Gaojing Zhou, Lan Yang, Yuchen Zhou, Linkai Liu, Yu He, Xi-man Liu, Shiping Dong, and 1 others. 2025. Repainter: Empowering e-commerce object removal via spatial-matting reinforcement learning. arXiv preprint arXiv:2510.07721. — [arXiv:2510.07721](https://arxiv.org/abs/2510.07721) find
  14. Xiwei Hu, Haokun Chen, Zhongqi Qi, Hui Zhang, Dexiang Hong, Jie Shao, and Xinglong Wu. 2025. Dreamposter: A unified framework for image-conditioned generative poster design. arXiv preprint arXiv:2507.04218. — [arXiv:2507.04218](https://arxiv.org/abs/2507.04218) find
  15. Tianyuan Qu, Lei Ke, Xiaohang Zhan, Longxiang Tang, Yuqi Liu, Bohao Peng, Bei Yu, Dong Yu, and Jiaya Jia. 2025. Replan: Reasoning-guided region planning for complex instruction-based image editing. arXiv preprint arXiv:2512.16864. — [arXiv:2512.16864](https://arxiv.org/abs/2512.16864) find
  16. Deyi Ji, Yuekui Yang, Liqun Liu, Peng Shu, Haiyang Wu, Shaogang Tang, Xudong Chen, Shaoping Ma, Tianrun Chen, and Lanyun Zhu. 2025a. RAVEN++: Pinpointing fine-grained violations in advertisement videos with active reinforcement reasoning. In Proceedings of the 2025 Conference on Empirical Methods in… find
  17. Shelly Rathee and Tyler Milfeld. 2024. Sustainability advertising: literature review and framework for future research. International Journal of Advertising, 43(1):7–35. find
  18. Deyi Ji, Yuekui Yang, Haiyang Wu, Shaoping Ma, Tianrun Chen, and Lanyun Zhu. 2025b. RAVEN: Robust advertisement video violation temporal grounding via reinforcement reasoning. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track), pages… find
  19. Qifei Jia, Yu Liu, Yajie Chai, Xintong Yao, Qiming Lu, Yasen Zhang, Runyu Shi, Ying Huang, and Guoquan Zhang. 2025. Lego-edit: A general image editing framework with model-level bricks and mllm builder. arXiv preprint arXiv:2509.12883. — [arXiv:2509.12883](https://arxiv.org/abs/2509.12883) find
  20. Keonwoo Ko, Gu Jin, and Youngchul Sung. 2022. Curriculum offline reinforcement learning. In International Conference on Learning Representations. find
  21. Léo Laugier, John Pavlopoulos, Jeffrey Sorensen, and Lucas Dixon. 2021. Civil rephrases of toxic texts with self-supervised transformers. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 1442–1461. find
  22. Hongyu Li, Manyuan Zhang, Dian Zheng, Ziyu Guo, Yimeng Jia, Kaituo Feng, Hao Yu, Yexin Liu, Yan Feng, Peng Pei, and 1 others. 2025. Editthinker: Unlocking iterative reasoning for any image editor. arXiv preprint arXiv:2512.05965. — [arXiv:2512.05965](https://arxiv.org/abs/2512.05965) find
  23. Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. 2023. Visual instruction tuning.
  24. Leonardo Madio and Martin Quinn. 2025. Content moderation and advertising in social media platforms. Journal of Economics & Management Strategy, 34(2):342–369. find
  25. Minheng Ni, Chenfei Wu, Xiaodong Wang, Shengming Yin, Lijuan Wang, Zicheng Liu, and Nan Duan. 2023. Ores: Open-vocabulary responsible visual synthesis. arXiv preprint arXiv:2308.13785. — [arXiv:2308.13785](https://arxiv.org/abs/2308.13785) find
  26. OpenAI. 2023. Gpt-4 technical report. Technical report, OpenAI.
  27. OpenAI. 2025. GPT-5 technical report. Forthcoming. find
  28. Patrick Schramowski, Manuel Brack, Björn Deiseroth, and Kristian Kersting. 2023. Safe latent diffusion: Mitigating inappropriate degeneration in diffusion models. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). find
  29. Silpa Vadakkeeveetil Sreelatha, Sauradip Nag, Muhammad Awais, Serge Belongie, and Anjan Dutta. 2025. Respodiff: Dual-module bottleneck transformation for responsible faithful t2i generation. Preprint, arXiv:2509.15257. — [arXiv:2509.15257](https://arxiv.org/abs/2509.15257) find
  30. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. Llama: Open and efficient foundation language models. arXiv pr…
  31. Chenfei Wu, Jiahao Li, Jingren Zhou, Junyang Lin, Kaiyuan Gao, Kun Yan, Sheng ming Yin, Shuai Bai, Xiao Xu, Yilei Chen, Yuxiang Chen, Zecheng Tang, Zekai Zhang, Zhengyi Wang, An Yang, Bowen Yu, Chen Cheng, Dayiheng Liu, Deqing Li, and 20 others. 2025. Qwen-image technical report. Preprint, arXiv:250…
  32. Jun Zhang, Yi Li, Yue Liu, Changping Wang, Yuan Wang, Yuling Xiong, Xun Liu, Haiyang Wu, Qian Li, Enming Zhang, and 1 others. 2025a. Gpr: Towards a generative pre-trained one-model paradigm for large-scale advertising recommendation. arXiv preprint arXiv:2511.10138. — [arXiv:2511.10138](https://arxiv.org/abs/2511.10138) find
  33. Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. 2018. The unreasonable effectiveness of deep features as a perceptual metric. In CVPR. find
  34. Ruiyang Zhang, Jiahao Luo, Xiaoru Feng, Qiufan Pang, Yaodong Yang, and Juntao Dai. 2025b. Safeeditor: Unified mllm for efficient post-hoc t2i safety editing. arXiv preprint arXiv:2510.24820. — [arXiv:2510.24820](https://arxiv.org/abs/2510.24820) find
  35. Qihao Zhu Runxin Xu Junxiao Song Mingchuan Zhang Y.K. Li Y. Wu Daya Guo Zhihong Shao, Peiyi Wang. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/te-2026-ssr-spatial-semantic-aware-instructions/