Magic Mirror Stealth - Interactive Automatic Picture Editing System.

This paper builds an interactive automatic picture editing system based on the most advanced algorithms in the field of multi-modality. It removes a specific area or entity in the image according to the natural language expression, which is like a magic mirror that can make the target invisible. The system is composed of two core processing modules: a referring image segmentation module based on natural language expressions and an automatic image inpainting module. The two modules cooperate to precisely control the editing area and generate the editing results based on the image semantics, which can ensure that the edited images are natural, realistic, and meet the editing needs of users. The two modules of the system are trained on open-source datasets separately. Experimental results demonstrate that the system can help users quickly edit images by a natural language expression.