Benchmarking Multimodal CoT Reward Model Stepwise by Visual Program
Paper i proceeding, 2025
Recent advancements in reward signal usage for Large Language Models (LLMs) are remarkable. However, significant challenges exist when transitioning reward signal to the multimodal domain, including labor-intensive annotations, over-reliance on one-step rewards, and inadequate evaluation. To address these issues, we propose SVIP, a novel approach to train a step-level multi-dimensional Chain-of-Thought (CoT) reward model automatically. It generates code for solving visual tasks and transforms the analysis of code blocks into the evaluation of CoT step as training samples. Then, we train SVIP-Reward model using a multi-head attention mechanism called TriAtt-CoT. The advantages of SVIP-Reward are evident throughout the entire process of MLLM. We also introduce a benchmark for CoT reward model training and testing. Experimental results demonstrate that SVIP-Reward improves MLLM performance across training and inference-time scaling, yielding better results on benchmarks while reducing hallucinations and enhancing reasoning ability.
multimodal model
reward model
chain of thought
visual programming
Författare
Minghe Gao
Zhejiang University
Universiti Kebangsaan Singapura (NUS)
Xuqi Liu
Zhejiang University
Yue Nick Zhongqi
Nanyang Technological University
Data Science och AI 3
Göteborgs universitet
Yang Wu
Ant group
Shuang Chen
Zhejiang University
Juncheng Li
Zhejiang University
Siliang Tang
Zhejiang University
Fei Wu
Zhejiang University
Tat Seng Chua
Universiti Kebangsaan Singapura (NUS)
Yueting Zhuang
Zhejiang University
Proceedings of the IEEE International Conference on Computer Vision
15505499 (ISSN) 23807504 (eISSN)
1718-17289798331587758 (ISBN)
2025 IEEE/CVF International Conference on Computer Vision, ICCV 2025
Honolulu, USA,
Honolulu, USA,
Ämneskategorier (SSIF 2025)
Datavetenskap (datalogi)
DOI
10.1109/ICCV51701.2025.00168