A COMPARATIVE EVALUATION OF CNN ARCHITECTURES FOR SINGLE-IMAGE GANS

Abstract

Generative Adversarial Networks (GANs) have achieved remarkable success in image synthesis, enabling the generation of photorealistic and diverse visual content. While most generative models depend on large datasets to capture visual variability, single-image GANs such as SinGAN demonstrate that rich generative behavior can emerge from the internal patch statistics of a single natural image. However, the effect of convolutional neural network (CNN) backbones on single-image generative performance remains underexplored. This study presents a comparative analysis of five CNN architectures—Inception, ResNet, DenseNet, CBAM, and MobileNet—integrated into the SinGAN framework to investigate their influence on image quality, diversity, and computational efficiency. Each architecture was trained under identical multi-scale SinGAN settings using 15 natural images, and evaluated with Single-Image Fréchet Inception Distance (SIFID), Multi-Scale Structural Similarity (MS-SSIM), and Learned Perceptual Image Patch Similarity (LPIPS), complemented by qualitative visual assessment. The results reveal consistent trade-offs among backbones: ResNet best preserves global structural coherence; DenseNet maximizes fine-detail diversity through dense feature reuse; CBAM enhances perceptual realism via attention module; Inception provides balanced multi-scale feature representation; and MobileNet achieves strong diversity and quality with favorable computational efficiency. These findings demonstrate that architectural design fundamentally governs generative behavior in single-image GANs. The study provides empirical insights and practical guidelines for selecting CNN backbones based on trade-offs between quality, diversity, and efficiency-supporting the design of data-efficient generative models for real-world and resource-constrained applications.

Keywords

• Single-image GAN
• SinGAN
• CNN architectures
• image synthesis
• SIFID
• MS-SSIM
• LPIPS

TEK GÖRÜNTÜ TABANLI GAN’LAR İÇİN CNN MİMARİLERİNİN KARŞILAŞTIRMALI DEĞERLENDİRMESİ

Abstract

Üretken Çekişmeli Ağlar (Generative Adversarial Networks, GANs), görüntü sentezinde olağanüstü bir başarı elde ederek fotogerçekçi ve çeşitli görsel içeriklerin üretilmesini mümkün kılmıştır. Çoğu üretken model, görsel çeşitliliği yakabilmek için büyük veri kümelerine ihtiyaç duyarken, SinGAN gibi tek görüntü tabanlı GAN'lar, zengin üretken davranışların yalnızca tek bir doğal görüntünün iç yama istatistiklerinden öğrenilebileceğini göstermiştir. Ancak, evrişimli sinir ağı (Convolutional Neural Network, CNN) omurgaların tek görüntü tabanlı üretken performans üzerindeki etkisi yeterince araştırılmamıştır. Bu çalışma, SinGAN çerçevesine entegre edilen beş farklı CNN mimarisinin (Inception, ResNet, DenseNet, CBAM ve MobileNet) görüntü kalitesi, çeşitlilik ve hesaplama verimliliği üzerindeki etkilerini inceleyen sistematik bir karşılaştırmalı analiz sunmaktadır. Her mimari, 15 doğal görüntü kullanılarak aynı çok ölçekli SinGAN ayarları altında eğitilmiş ve Tek Görüntü Fréchet Inception Mesafesi (SIFID), Çok Ölçekli Yapısal Benzerlik (MS-SSIM) ve Öğrenilmiş Algısal Görüntü Yama Benzerliği (LPIPS) ile değerlendirilmiş, niteliksel görsel değerlendirme ile desteklenmiştir. Elde edilen sonuçlar, mimariler arasında tutarlı ödünleşmeler olduğunu göstermektedir: ResNet, küresel yapısal bütünlüğü en iyi koruyan mimaridir; DenseNet, yoğun özellik yeniden kullanımıyla ince ayrıntı çeşitliliğini en üst düzeye çıkarır; CBAM, dikkat mekanizması aracılığıyla algısal gerçekçiliği artırır; Inception, çok ölçekli özellik temsili açısından dengeli performans sunar; MobileNet ise güçlü çeşitlilik ve yüksek kaliteyi üstün hesaplama verimliliği ile birleştirir. Bu bulgular, mimari tasarımın tek görüntü tabanlı GAN'larda üretken davranışı temelden şekillendirdiğini doğrulamaktadır. Çalışma, görüntü kalitesi, çeşitlilik ve verimlilik arasındaki ödünleşmelere dayalı olarak uygun CNN omurga seçimi için ampirik içgörüler ve pratik rehberler sunmakta, gerçek dünya ve kaynak kısıtlı uygulamalar için veri verimli üretken modellerin geliştirilmesine katkı sağlamaktadır.

Keywords

• Tek görüntü GAN
• SinGAN
• CNN mimarileri
• görüntü sentezi
• SIFID
• MS-SSIM
• LPIPS

References

1. L. Ruthotto, E. Haber, An introduction to deep generative modeling, GAMM‐Mitteilungen 44 (2) (2021) e202100008.
2. K. Wang, G. Zhang, Y. Leng, H. Leung, Synthetic aperture radar image generation with deep generative models, IEEE Geoscience and Remote Sensing Letters 16 (6) (2018) 912-916.
3. L. Regenwetter, A.H. Nobari, F. Ahmed, Deep generative models in engineering design: A review, Journal of Mechanical Design 144 (7) (2022) 071704.
4. N. Killoran, L.J. Lee, A. Delong, D. Duvenaud, B.J. Frey, Generating and designing DNA with deep generative models, arXiv preprint arXiv:1712.06148 (2017).
5. M. Suzuki, K. Nakayama, Y. Matsuo, Joint multimodal learning with deep generative models, arXiv preprint arXiv:1611.01891 (2016).
6. C. Doersch, Tutorial on variational autoencoders, arXiv preprint arXiv:1606.05908 (2016).
7. J. Xiong, G. Liu, L. Huang, C. Wu, T. Wu, Y. Mu, ..., N. Wong, Autoregressive models in vision: A survey. arXiv preprint arXiv:2411.05902 (2024).
8. I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, ..., Y. Bengio, Generative adversarial networks, Communications of the ACM 63 (11) (2020) 139-144.

9. H. Huang, P.S. Yu, C. Wang, An introduction to image synthesis with generative adversarial nets, arXiv preprint arXiv:1803.04469 (2018).
10. H. Zhang, T. Xu, H. Li, S. Zhang, X. Wang, X. Huang, D.N. Metaxas, Stackgan: Text to photo-realistic image synthesis with stacked generative adversarial networks, in: Proceedings of the IEEE International Conference on Computer Vision, (2017), pp. 5907-5915.
11. D. Nie, R. Trullo, J. Lian, C. Petitjean, S. Ruan, Q. Wang, D. Shen, Medical image synthesis with context-aware generative adversarial networks, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, (2017), pp. 417-425.
12. M. Liu, P. Maiti, S. Thomopoulos, A. Zhu, Y. Chai, H. Kim, N. Jahanshad, Style transfer using generative adversarial networks for multi-site MRI harmonization, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, (2021), pp. 313-322.
13. Z. Qin, Z. Liu, P. Zhu, W. Ling, Style transfer in conditional GANs for cross-modality synthesis of brain magnetic resonance images, Computers in Biology and Medicine 148 (2022) 105928.
14. J.S. Ubhi, A.K. Aggarwal, Neural style transfer for image within images and conditional GANs for destylization, Journal of Visual Communication and Image Representation 85 (2022) 103483.
15. P.L. Vidal, J. de Moura, J. Novo, M.G. Penedo, M. Ortega, Image-to-image translation with generative adversarial networks via retinal masks for realistic optical coherence tomography imaging of diabetic macular edema disorders, Biomedical Signal Processing and Control 79 (2023) 104098.
16. A. Alotaibi, Deep generative adversarial networks for image-to-image translation: A review. Symmetry 12 (10) (2020) 1705.
17. K. Armanious, C. Jiang, M. Fischer, T. Küstner, T. Hepp, K. Nikolaou, ..., B. Yang, MedGAN: Medical image translation using GANs, Computerized Medical Imaging and Graphics 79 (2020) 101684.
18. C. Ledig, L. Theis, F. Huszár, J. Caballero, A. Cunningham, A. Acosta, ..., W. Shi, Photo-realistic single image super-resolution using a generative adversarial network, in: Proceedings of The IEEE Conference on Computer Vision and Pattern Recognition, (2017), pp. 4681-4690.
19. H. Xiao, X. Wang, J. Wang, J.Y. Cai, J.H. Deng, J.K. Yan, Y.D. Tang, Single image super-resolution with denoising diffusion GANS, Scientific Reports 14 (1) (2024) 4272.
20. L. Zhao, H. Bai, J. Liang, B. Zeng, A. Wang, Y. Zhao, Simultaneously color-depth super-resolution with conditional generative adversarial network, arXiv preprint arXiv:1708.09105 (2017).
21. T.R. Shaham, T. Dekel, T. Michaeli, SinGAN: Learning a generative model from a single natural image. arXiv e-prints, arXiv-1905 (2019).
22. E. Yildiz, M.E. Yuksel, S. Sevgen, A single-image GAN model using self-attention mechanism and DenseNets, Neurocomputing 596 (2024) 127873.
23. Z. Zhang, C. Han, T. Guo, Exsingan: Learning an explainable generative model from a single image, arXiv preprint arXiv:2105.07350 (2021).
24. V. Sushko, J. Gall, A. Khoreva, One-Shot Gan: Learning to generate samples from single images and videos, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, (2021), pp. 2596-2600.
25. X. Chen, H. Zhao, D. Yang, Y. Li, Q. Kang, H. Lu, SA-SinGAN: Self-attention for single-image generation adversarial networks, Machine Vision and Applications 32 (4) (2021) 104.
26. P. Wang, Y. Li, K.K. Singh, J. Lu, N. Vasconcelos, IMAGINE: Image synthesis by image-guided model inversion, in: Proceedings of The IEEE Conference on Computer Vision and Pattern Recognition, (2021), pp. 3681-3690.
27. C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, Z. Wojna, Rethinking the inception architecture for computer vision, in: Proceedings of The IEEE Conference on Computer Vision and Pattern Recognition, (2016), pp. 2818-2826.
28. K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in: Proceedings of The IEEE Conference on Computer Vision and Pattern Recognition, (2016), pp. 770-778.
29. G. Huang, Z. Liu, L. Van Der Maaten, K.Q. Weinberger, Densely connected convolutional networks, in: Proceedings of The IEEE Conference on Computer Vision and Pattern Recognition, (2017), pp. 4700-4708.
30. S. Woo, J. Park, J.Y. Lee, I.S. Kweon, CBAM: Convolutional block attention module, in: Proceedings of The European Conference on Computer Vision, (2018), pp. 3-19.
31. A.G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, ..., H. Adam, MobileNets: Efficient convolutional neural networks for mobile vision applications, arXiv preprint arXiv:1704.04861 (2017).
32. Z. Wang, A.C. Bovik, H.R. Sheikh, E.P. Simoncelli, Image quality assessment: From error visibility to structural similarity, IEEE Transactions on Image Processing 13 (4) (2004) 600-612.
33. R. Zhang, P. Isola, A.A. Efros, E. Shechtman, O. Wang, The unreasonable effectiveness of deep features as a perceptual metric, in: Proceedings of The IEEE Conference on Computer Vision and Pattern Recognition, (2018), pp. 586-595.
34. M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, S. Hochreiter, GANs trained by a two time-scale update rule converge to a local nash equilibrium, Advances in Neural Information Processing Systems 30 (2017).