Text generation image model training method and system and text generation image method and system

• Main Authors: XUE ZHIHANG, LANG CONGYAN, WANG ZHENXUE, LI YIDONG
• Format: Patent
• Language: Chinese; English
• Published: 14.11.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a text generation image model training method and system and a text generation image method and system, and belongs to the technical field of computer vision and artificial intelligence. Text description is input into a pre-trained text encoder, and global sentence features and a word feature matrix are obtained through extraction; randomly sampled noise features are connected with sentence features and input into a generative network, and low-scale visual text features are extracted from a first generation module; and the image feature and visual text feature cross-scale channel activation module obtains enhanced image features, and the enhanced image features are respectively transmitted into a generator in sequence to obtain a generated image. According to the invention, the generation network based on the cross-scale channel activation module fuses low-scale visual text features containing rich global semantic information with image features, so that the final image generation quali