Xinyue Li, Zhiming Xu, Zhichao Zhang, Zhaolin Cai, Sijing Wu, Xiongkuo Min, Yitong Chen, Guangtao Zhai
ELIQ is a label-free framework designed to assess the quality of AI-generated images, adapting to evolving generative models without the need for human annotations.
AI-generated images are rapidly improving, making it difficult to keep up with the quality standards using traditional methods that rely on human labels. ELIQ is a new framework that evaluates the quality of these images without needing human input. It uses a combination of positive and negative example pairs to learn about image quality and alignment with prompts, adapting a pre-trained model for this purpose. ELIQ has been shown to outperform other methods and can work with both AI-generated and user-generated content.
