Composition Vision-Language Understanding via Segment and Depth Anything Model

This integration signifies a We introduce a pioneering unified library that leverages significant advancement in the field, facilitating a deeper depth anything, segment anything models to augment neural understanding of images through language models and improving comprehension in language-vision model zero-shot understanding. the efficacy of multi-modal tasks. This library synergizes the capabilities of the In recent works on text-image multi-modal tasks [1, 6, Depth Anything Model (DAM), Segment Anything Model 7, 9], the primary focus has been on training specific models (SAM), and GPT-4V, enhancing multimodal tasks such as to enhance the similarity between text-image pairs and vision-question-answering (VQA) and composition reasoning.