Method, device and equipment for training text graph model and medium

The invention provides a text graph model training method and device, equipment and a medium. The method comprises the following steps: acquiring a plurality of fine-grained labels of a to-be-processed image; inputting the to-be-processed image and the plurality of fine-grained labels into a first pre-training model to obtain an image description, which refers to REC (Registered Establishment) and RES (Registered Segmentation); inputting the plurality of fine-grained labels, the image description, the REC and the RES into a second pre-training model to obtain a description text; and generating a training sample according to the description text and the to-be-processed image, wherein the training sample is used for training a text graph model. By combining a plurality of fine-grained labels, image description, REC and RES, image-level, local and pixel-level understanding is provided for the model, and the understanding of the model on the spatial relationship between objects and the relationship between the ob