Benchmarking and Evolving Reason-Reflect-Rectify for Reflective Visual Generation

Junjie Wang, Xinghua Lou, Xiangtai Li, Ye Tian, Keyu Chen, Yulin Li, Bin Kang, Guangcan Mai, Yanwei Li, Zhuotao Tian, Liqiang Nie
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:128684-128713, 2026.

Abstract

Text-to-Image (T2I) models and Unified Multimodal Models (UMMs) have achieved remarkable progress in visual generation. However, their reliance on a single-pass generation paradigm limits their ability to handle complex prompts requiring iterative refinement. To enable multi-round Reflective Visual Generation (RVG), we formalize the Reason–Reflect–Rectify (R$^3$) loop as a core framework and introduce R$^3$-Bench, a benchmark of over 600 expert-annotated instances that quantifies iterative reasoning and rectification capabilities. Evaluation on R$^3$-Bench reveals a critical gap: while state-of-the-art models can identify generation errors, they fail to generate actionable rectification instructions. To bridge this gap, we propose R$^3$-Refiner, a dual-stage framework leveraging Group Relative Policy Optimization (GRPO) and a Hierarchical Reward Mechanism (HRM) to better align rectification with reflective reasoning. Experiments show that R$^3$-Refiner achieves significant improvements on R$^3$-Bench (+12.0% in Reflective Verdict Score, +9.0% in Rectification Score), and can be seamlessly integrated with various MLLMs to enhance the generation quality of different T2I models on GenEval++ and T2I-CompBench. Code is available at https://github.com/xiaomoguhz/R3-Bench.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-wang26fe, title = {Benchmarking and Evolving Reason-Reflect-Rectify for Reflective Visual Generation}, author = {Wang, Junjie and Lou, Xinghua and Li, Xiangtai and Tian, Ye and Chen, Keyu and Li, Yulin and Kang, Bin and Mai, Guangcan and Li, Yanwei and Tian, Zhuotao and Nie, Liqiang}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {128684--128713}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/wang26fe/wang26fe.pdf}, url = {https://proceedings.mlr.press/v306/wang26fe.html}, abstract = {Text-to-Image (T2I) models and Unified Multimodal Models (UMMs) have achieved remarkable progress in visual generation. However, their reliance on a single-pass generation paradigm limits their ability to handle complex prompts requiring iterative refinement. To enable multi-round Reflective Visual Generation (RVG), we formalize the Reason–Reflect–Rectify (R$^3$) loop as a core framework and introduce R$^3$-Bench, a benchmark of over 600 expert-annotated instances that quantifies iterative reasoning and rectification capabilities. Evaluation on R$^3$-Bench reveals a critical gap: while state-of-the-art models can identify generation errors, they fail to generate actionable rectification instructions. To bridge this gap, we propose R$^3$-Refiner, a dual-stage framework leveraging Group Relative Policy Optimization (GRPO) and a Hierarchical Reward Mechanism (HRM) to better align rectification with reflective reasoning. Experiments show that R$^3$-Refiner achieves significant improvements on R$^3$-Bench (+12.0% in Reflective Verdict Score, +9.0% in Rectification Score), and can be seamlessly integrated with various MLLMs to enhance the generation quality of different T2I models on GenEval++ and T2I-CompBench. Code is available at https://github.com/xiaomoguhz/R3-Bench.} }
Endnote
%0 Conference Paper %T Benchmarking and Evolving Reason-Reflect-Rectify for Reflective Visual Generation %A Junjie Wang %A Xinghua Lou %A Xiangtai Li %A Ye Tian %A Keyu Chen %A Yulin Li %A Bin Kang %A Guangcan Mai %A Yanwei Li %A Zhuotao Tian %A Liqiang Nie %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-wang26fe %I PMLR %P 128684--128713 %U https://proceedings.mlr.press/v306/wang26fe.html %V 306 %X Text-to-Image (T2I) models and Unified Multimodal Models (UMMs) have achieved remarkable progress in visual generation. However, their reliance on a single-pass generation paradigm limits their ability to handle complex prompts requiring iterative refinement. To enable multi-round Reflective Visual Generation (RVG), we formalize the Reason–Reflect–Rectify (R$^3$) loop as a core framework and introduce R$^3$-Bench, a benchmark of over 600 expert-annotated instances that quantifies iterative reasoning and rectification capabilities. Evaluation on R$^3$-Bench reveals a critical gap: while state-of-the-art models can identify generation errors, they fail to generate actionable rectification instructions. To bridge this gap, we propose R$^3$-Refiner, a dual-stage framework leveraging Group Relative Policy Optimization (GRPO) and a Hierarchical Reward Mechanism (HRM) to better align rectification with reflective reasoning. Experiments show that R$^3$-Refiner achieves significant improvements on R$^3$-Bench (+12.0% in Reflective Verdict Score, +9.0% in Rectification Score), and can be seamlessly integrated with various MLLMs to enhance the generation quality of different T2I models on GenEval++ and T2I-CompBench. Code is available at https://github.com/xiaomoguhz/R3-Bench.
APA
Wang, J., Lou, X., Li, X., Tian, Y., Chen, K., Li, Y., Kang, B., Mai, G., Li, Y., Tian, Z. & Nie, L.. (2026). Benchmarking and Evolving Reason-Reflect-Rectify for Reflective Visual Generation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:128684-128713 Available from https://proceedings.mlr.press/v306/wang26fe.html.

Related Material