Flow style-aware network for arbitrary style transfer

Researchers have recently proposed arbitrary style transfer methods based on various model frameworks. Although all of them have achieved good results, they still face the problems of insufficient stylization, artifacts and inadequate retention of content structure. In order to solve these problems, we propose a flow style-aware network (FSANet) for arbitrary style transfer, which combines a VGG network and a flow network. FSANet consists of a flow style transfer module (FSTM), a dynamic regulation attention module (DRAM), and a style feature interaction module (SFIM). The flow style transfer module uses the reversible residue block features of the flow network to create a sample feature containing the target content and style. To adapt the FSTM to VGG networks, we design the dynamic regulation attention module and exploit the sample features both at the channel and pixel levels. The style feature interaction module computes a style tensor that optimizes the fused features. Extensive qualitative and quantitative experiments demonstrate that our proposed FSANet can effectively avoid artifacts and enhance the preservation of content details while migrating style features. • We designed a flow style transfer module (FSTM) based on the flow network. It uses the characteristics of a flow network to transfer style and content features, generating a sample feature containing content and style. It can prevent images from generating repetitive patterns. • We propose a dynamic regulation attention mechanism (DRAM), which can effectively transfer the style information in the sample features to the content features. DRAM is able to utilize the output of the FSTM. Avoiding the direct use of FSTM leads to residual shadows in the generated image. • We introduce a style feature interaction module (SFIM), which can extract the style tensor of the four-layer output of the VGG network, thereby optimizing the final generated image. We use high and low frequency division to reduce SFIM calculation costs.