A Unified Image-Dense Annotation Generation Model for Underwater Scenes

Figure 1 . We present TIDE, a unified underwater image-dense annotation generation model. Its core lies in the shared layout information and the natural complementarity between multimodal features. Our model, derived from the text-to-image model and finetuned with underwater data, enables the generation of highly consistent underwater image-dense annotations from solely text conditions.