This model consists of several key modules, including: a large language model, visual encoder, segmentation decoder, visual text mapper, classification layer, and positioning structure. The training ...