VISMA: A Machine Learning Approach to Image Manipulation

Abstract

Powerful tools for semantic picture modification have recently been developed thanks to recent improvements in image production. Existing methods, however, may need a large amount of supplementary data. They are unable to handle all editing actions, including the insertion, modification, or deletion of semantic attributes of the image. The proposed method, a new Image Semantic Manipulator pair for Adding, Moving, or Erasing Objects in Scenes during Semantic Editing, to overcome these restrictions. In our arrangement, the generator generates the relevant pixels and edits the image when the end user gives the semantic identifiers of the regions to be changed. The paper aims to present results for two jobs: (a) picture editing and (b) image production with semantic label constraint using General Adversarial Networks (GAN) machine learning approach. A pre-trained neural network can also adjust to images of various sizes. Making the image fit the needs of the user by altering the overall aesthetic and emotional tone also helps. When given appropriate training input/output pairings, our technique may be expanded to provide a range of outputs when label mappings are changed, enabling interactive image modification that enables users to interact with changing an object’s appearance in real time. Last but not least, our suggested approach will raise resolution while establishing a new standard for realistic photos.