Fkdiff: An Efficient Diffusion Model for Image Super-Resolution With Fourier Frequency Domain Transformation and Knowledge Distillation

Yu Chen

doi:10.62051/ijcsit.v4n3.25

Authors

Yu Chen

DOI:

https://doi.org/10.62051/ijcsit.v4n3.25

Keywords:

Image super-resolution, Diffusion Models, Knowledge Distillation, Computational efficiency

Abstract

Image super-resolution (SR) techniques play a crucial role in various applications such as image restoration, medical imaging, surveillance, and remote sensing. Traditional methods often employ interpolation algorithms to upscale images, resulting in artifacts and reduced perceptual quality. Recent advancements in diffusion models (DM) have shown promising results in image generation tasks but are hindered by computational complexity, particularly in resource-constrained environments. By leveraging low-resolution images as prior information and operating in the frequency domain, FKdiff achieves enhanced computational efficiency and preserves high-frequency details effectively. The proposed method integrates a progressive hexagonal knowledge distillation (PHexKD) approach, ensuring lightweight model deployment without compromising performance. Experimental results demonstrate that FKdiff outperforms existing methods in terms of efficiency and effectiveness while a small amount of image generation quality is lost.

Downloads

Download data is not yet available.

References

[1] J. Ho, A. Jain, and P.J.A.i.n.i.p.s. Abbeel, Denoising diffusion probabilistic models, 33 (2020), 6840-6851.

[2] C. Buciluǎ, R. Caruana, and A. Niculescu-Mizil, Model compression, in: Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, 2006, pp. 535-541.

[3] R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, High-resolution image synthesis with latent diffusion models, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 10684-10695.

[4] B.B. Moser, S. Frolov, F. Raue, S. Palacio, and A. Dengel, Dwa: Differential wavelet amplifier for image super-resolution, in: International Conference on Artificial Neural Networks, Springer, 2023, pp. 232-243.

[5] F. Guth, S. Coste, V. De Bortoli, and S.J.A.i.N.I.P.S. Mallat, Wavelet score-based generative modeling, 35 (2022), 478-491.

[6] S. Shang, Z. Shan, G. Liu, and J.J.a.p.a. Zhang, Resdiff: Combining cnn and diffusion model for image super-resolution, (2023).

[7] D. Podell, Z. English, K. Lacey, A. Blattmann, T. Dockhorn, J. Müller, J. Penna, and R.J.a.p.a. Rombach, Sdxl: Improving latent diffusion models for high-resolution image synthesis, (2023).

[8] B. Xia, Y. Zhang, S. Wang, Y. Wang, X. Wu, Y. Tian, W. Yang, and L. Van Gool, Diffir: Efficient diffusion model for image restoration, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 13095-13105.

[9] Y. Wang, W. Yang, X. Chen, Y. Wang, L. Guo, L.-P. Chau, Z. Liu, Y. Qiao, A.C. Kot, and B. Wen, SinSR: diffusion-based image super-resolution in a single step, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 25796-25805.

[10] S. Zagoruyko and N.J.a.p.a. Komodakis, Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer, (2016).