SED-Net: Speedy Encoding-Decoding Network for Artistic Style Transfer
Haoqun Teng · 2024
To address the issues of high computational complexity and low adaptability in traditional image style transfer models, this paper proposes a Speedy Encoder-Decoder Network for Style Transfer (SED-Net). This network improves computational efficiency by introducing channel-wise convolution techniques. Additionally, the incorporation of attention mechanisms and Adaptive Instance Normalization between the encoder and decoder enhances the quality and adaptability of style transfer. Experimental results demonstrate that SED-Net achieves efficient real-time transfer and superior performance in terms of SSIM and PSNR compared to traditional Neural Style Transferand Fast Style Transfer. The generated images also exhibit more natural visual effects.