Model training method, address localization methods, electronic device, storage medium and computer program product

Abstract

The embodiments of the present application relate to the technical fields of large models and image address localization, and provide a model training method, address localization methods, an electronic device, a storage medium and a computer program product. The model training method comprises: acquiring a training data set, the training data set comprising training scene images and a visual question answering data set of address information corresponding to the training scene images; using the training data set to perform cross-view alignment training on an initial multi-modal address localization model, so as to generate an intermediate multi-modal address localization model; and using the training data set to perform address localization training on the intermediate multi-modal address localization model, so as to generate a target multi-modal address localization model, the target multi-modal address localization model being used for performing address localization analysis on a target scene image to be processed and an address question, so as to obtain an address answer. The described method solves the problems of coarse localization granularity, poor localization accuracy, and poor flexibility of question-answering interaction of image address localization solutions in the prior art.

Images