Efficient Re-parameterization Operations Search for Easy-to-Deploy Network Based on Directional Evolutionary Strategy

Traditional NAS methods improve performance by sacrificing the landing ability of architecture and re-parameterization technology is expected to solve this problem. However, most current Rep rely on prior knowledge select operations, which limits type operations knowledge. At same time, some hinder optimization network. To break these restrictions, in work, an improved search space designed, including more operations. Concretely, convolutional networks can be further enhanced space. An automatic enhancement strategy designed effectively explore based neural (NAS), excellent architecture. Then, we solved problem caused using enhance ResNet-style Besides, visualize output features analyze reasons for formation On public datasets, achieve better results. Under training conditions as ResNet, accuracy ResNet-50 1.82% ImageNet-1k.