Style-Guided and Disentangled Representation for Robust Image-to-Image Translation

Recently, various image-to-image translation (I2I) methods have improved mode diversity and visual quality in terms of neural networks or regularization terms. However, conventional I2I relies on a static decision boundary the encoded representations those are entangled with each other, so they often face ‘mode collapse’ phenomenon. To mitigate collapse, 1) we design so-called style-guided discriminator that guides an input image to target style based strategy flexible boundary. 2) Also, make include independent domain attributes. Based two ideas, this paper proposes Style-Guided Disentangled Representation for Robust Image-to-Image Translation (SRIT). SRIT showed outstanding FID by 8%, 22.8%, 10.1% CelebA-HQ, AFHQ, Yosemite datasets, respectively. The translated images reflect styles successfully. This indicates shows better than previous works.