Video synthesis using artificial intelligence neural networks provides superior image quality compared to traditional synthesis methods, but it requires a significant amount of computation. While Graphics Processing Units (GPUs) are primarily used for training and inference of artificial intelligence neural networks, they are often unsuitable for real-time applications due to performance limitations or high costs. Therefore, we have designed a lightweight Convolutional Neural Network (CNN) using the MobileNet structure and bit optimization techniques. Additionally, we have developed a code that can be executed on a Field-Programmable Gate Array (FPGA) board.
