MindOmni: Unleashing Reasoning Generation in Vision Language Models with RGPO
This work addresses the problem of enhancing reasoning generation in vision language models for researchers and practitioners in AI, representing a novel method rather than an incremental improvement.
The paper tackles the limitations of text-to-image systems in handling multimodal inputs and complex reasoning tasks by introducing MindOmni, a unified multimodal large language model that incorporates reasoning generation through reinforcement learning, achieving impressive performance on understanding and generation benchmarks with advanced fine-grained reasoning capabilities.
Recent text-to-image systems face limitations in handling multimodal inputs and complex reasoning tasks. We introduce MindOmni, a unified multimodal large language model that addresses these challenges by incorporating reasoning generation through reinforcement learning. MindOmni leverages a three-phase training strategy: i) design of a unified vision language model with a decoder-only diffusion module, ii) supervised fine-tuning with Chain-of-Thought (CoT) instruction data, and iii) our proposed Reasoning Generation Policy Optimization (RGPO) algorithm, utilizing multimodal feedback to effectively guide policy updates. Experimental results demonstrate that MindOmni outperforms existing models, achieving impressive performance on both understanding and generation benchmarks, meanwhile showcasing advanced fine-grained reasoning generation capabilities, especially with mathematical reasoning instruction. All codes will be made public at https://github.com/TencentARC/MindOmni