Frequency Domain Image Translation: More Photo-realistic, Better Identity-preserving

Cai, Mu; Zhang, Hong; Huang, Huijuan; Geng, Qichuan; Li, Yixuan; Huang, Gao

doi:10.48550/arxiv.2011.13611

Cited by 2 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Dzanic et al [14] observe that discrepancy exists between the images generated by the GAN [16] and the real images through the analysis of high-frequency Fourier modes. In addition, attempts to utilize the frequency domain information in the various fields, including image forensics [13,15,48], image generation [5,9,20], and domain adaptation [44,45] are gradually increasing. However, image denoising methods combining the frequency domain analysis with DNN remain much less explored.…”

Section: Frequency Domain In Cnnsmentioning

confidence: 99%

“…Frequency Reconstruction Loss Cai et al [5] demonstrate the existence of a gap between the real and generated image in the frequency domain, which leads to artifacts in the spatial domain. Motivated by this observation, we propose to use frequency reconstruction loss with cycle consistency loss to ameliorate the quality of denoised images while reducing the gap.…”

Section: Frequency Domain Constraintsmentioning

confidence: 99%

See 1 more Smart Citation

Unsupervised Image Denoising with Frequency Domain Knowledge

Kim¹,

Jang²,

Lee³

et al. 2021

Preprint

View full text Add to dashboard Cite

Supervised learning-based methods yield robust denoising results, yet they are inherently limited by the need for large-scale clean/noisy paired datasets. The use of unsupervised denoisers, on the other hand, necessitates a more detailed understanding of the underlying image statistics. In particular, it is well known that apparent differences between clean and noisy images are most prominent on high-frequency bands, justifying the use of low-pass filters as part of conventional image preprocessing steps. However, most learning-based denoising methods utilize only one-sided information from the spatial domain without considering frequency domain information. To address this limitation, in this study we propose a frequency-sensitive unsupervised denoising method. To this end, a generative adversarial network (GAN) is used as a base structure. Subsequently, we include spectral discriminator and frequency reconstruction loss to transfer frequency knowledge into the generator. Results using natural and synthetic datasets indicate that our unsupervised learning method augmented with frequency information achieves stateof-the-art denoising performance, suggesting that frequency domain information could be a viable factor in improving the overall performance of unsupervised learning-based methods.

show abstract

Section: Frequency Domain In Cnnsmentioning

confidence: 99%

Section: Frequency Domain Constraintsmentioning

confidence: 99%

Unsupervised Image Denoising with Frequency Domain Knowledge

Kim¹,

Jang²,

Lee³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Di↵erent from [66,67], the proposed focal frequency loss in Chapter 5 dynamically focuses the model on hard frequencies by down-weighting the easy ones and ameliorates image quality through the frequency domain directly. Some concurrent works include [115][116][117].…”

Section: Frequency Domain Analysismentioning

confidence: 99%

Image and video generation via deep learning

Jiang

View full text Add to dashboard Cite

The FID (lower is better) and IS (higher is better) scores (256⇥256) of our method compared to previous techniques for regularizing GANs on StyleGAN2 trained with FFHQ-5k [5]. . . . . . . . 89 6.5 Ablation studies on variants of APA on FFHQ-5k [5] (256 ⇥ 256). We study three key elements of APA, i.e., the overfitting heuristic , the deception strength p, and the deception strategy.The "main" denotes the main version used in our previous experiments (i.e., = r , p is adaptively adjusted, and the deception strategy is analogous to one-sided label flipping

show abstract

Frequency Domain Image Translation: More Photo-realistic, Better Identity-preserving

Cited by 2 publications

References 0 publications

Unsupervised Image Denoising with Frequency Domain Knowledge

Unsupervised Image Denoising with Frequency Domain Knowledge

Image and video generation via deep learning

Contact Info

Product

Resources

About