Code and Experimental Data for Interleaved Text-Image Generation Based on Multi-Agent Debate

It mainly includes the implementation code of the multi-agent debate image text interweaving generation method proposed in the paper, the text inference agent, the visual perception agent, and the global arbitration agent modules, the multi round debate and conflict resolution module, the prompt word template, the data processing and evaluation script, as well as the relevant output files and configuration instructions for reproducing experimental results.

The paper experiment is based on MRAMG Bench, and the original MRAMG Bench data is not duplicated in this dataset. Users need to obtain it separately according to the requirements of the original dataset publisher.