Frequency Domain Image Translation: More Photo-realistic, Better Identity-preserving
This work provides an incremental improvement in image-to-image translation quality, specifically for researchers and practitioners working on GAN-based image generation tasks where identity preservation is crucial.
This paper addresses the problem of identity preservation in GAN-based image-to-image translation, where existing methods often lose source identity and suffer from suboptimal visual quality. The proposed Frequency Domain Image Translation (FDIT) framework decomposes images into low- and high-frequency components to better preserve identity, achieving a 5.6% reduction in average FID score compared to the previous state-of-the-art.
Image-to-image translation has been revolutionized with GAN-based methods. However, existing methods lack the ability to preserve the identity of the source domain. As a result, synthesized images can often over-adapt to the reference domain, losing important structural characteristics and suffering from suboptimal visual quality. To solve these challenges, we propose a novel frequency domain image translation (FDIT) framework, exploiting frequency information for enhancing the image generation process. Our key idea is to decompose the image into low-frequency and high-frequency components, where the high-frequency feature captures object structure akin to the identity. Our training objective facilitates the preservation of frequency information in both pixel space and Fourier spectral space. We broadly evaluate FDIT across five large-scale datasets and multiple tasks including image translation and GAN inversion. Extensive experiments and ablations show that FDIT effectively preserves the identity of the source image, and produces photo-realistic images. FDIT establishes state-of-the-art performance, reducing the average FID score by 5.6% compared to the previous best method.