Recently, Intelligent Information Sensing and Processing at Ocean University of China (OUC) has made a series of original achievements in artificial intelligence research, which have been published by top international AI conferences (CVPR; ICCV; ECCV; ACM MM) and journals (IJCV, International Journal of Computer Vision; IEEE; TPAMI, and Transactions on Pattern Analysis and Machine Intelligence). This indicates OUC’s growing impact in the field of AI. 

Over the years, high-quality synthesis of image to image transformation has been a great challenge that has drawn wide research interest. The research team proposed a block-based discriminant region proposal mechanism, and framed the Generative Adversarial Network (GANs) to obtain high-quality composite images with higher resolution and more accurate details.

Composite images with visual fidelity are important, especially in areas of entertainment, advertising and film production. The intelligent adjustment of composite images to make them realistic, or image harmonization, has become a challenging research hotspot. The team proposed innovative ideas and methods for intrinsic image harmonization, which generate composite images with consistent overall perception.

Recently, the team has been working on visual and graphic problems concerning image to image transformation such as image harmonization by effective use of Transformer with Self-Attention Network (SAN) for its remote context modeling capability. This is an important application of AI technology to resolve issues in image to image conversion, which is conducive to the development of general AI.