Infrared and visible image fusion(IVIF) technology is designed to integrate complementary information from images captured by thermal radiation sensors and optical sensors of the same scene, generating a fused image that is more suitable for human understanding or computer analysis. With the advancement of deep learning, this technology has become increasingly important in fields such as military reconnaissance, autonomous driving, and security surveillance. Previous reviews only summarize relevant literature and do not provide detailed analysis from the perspective of network structure and loss function development, also lack the latest research progress and comparative experiments. In view of this, a comprehensive review and outlook on deep learning based IVIF methods are conducted. Firstly, the development history of deep learning-based IVIF methods is reviewed, introducing the evolution of network structures and loss functions. Secondly, common datasets and performance evaluation metrics in IVIF are summarized, and the desired characteristics of future datasets are discussed. Then, 18 representative methods are extensively tested on three public datasets, and their performance is analyzed from both subjective and objective perspectives. Finally, the current challenges faced by IVIF are summarized, and future research directions are envisioned.