Bıyık Deseni Üretiminde Çekişmeli Üretici Ağların Performans Karşılaştırması

Bu çalışmada görüntüden görüntüye dönüşüm yapan çekişmeli üretici ağ mimarilerinin performans incelemesi yapılıp, sentetik görüntü üretimindeki başarımı değerlendirilmiştir. Bu modellerin kaliteli bir başarım değerlendirmesi için standartlaştırılmış veri kümeleri yerine gerçek iş alanından toplanılan denim2bıyık veri kümesi kullanılmıştır. Denim kumaşları üzerine çizilen bıyık desenleri lazer cihazıyla oluşturulmaktadır. Bu cihazın istenilen bıyık desenini oluşturabilmesi için uzmanlaşmış bir personel tarafından görsel düzenleme programları ile yaklaşık 2-3 saat süren bir çalışma yapması gerekir. Önerilen yaklaşımla otomatik bir bıyık üretim işlemi gerçekleşecek, manuel üretimdeki hatalar ve zamansal kayıplar elimine edilecektir. Yaptığımız literatür araştırması neticesinde denim ürün görsellerinin üretken ağlar ile üretilmesi ile ilgili farklı bir çalışma bulunmamaktadır. Bu durum yapılan çalışmanın akademik özgün değerini yükseltmektedir. Çalışmada kullanılan ÇÜA mimarileri Pix2Pix, CycleGAN, DiscoGAN ve AttentionGAN’dır. Her bir mimarinin denim2bıyık veri kümesindeki eğitim ve test verileri üzerinde bıyık deseni üretim başarım değerlendirmesi ve maliyet analizi yapılmıştır. Yapılan çalışmalar sonucunda, bıyık desen görseli üretim hızı bir saniyenin altına düşerken, üretim doğruluğu %86 seviyelerine çıktığı görülmektedir.

Performance Comparison of Generative Adversarial Networks in Mustache Pattern Production

In this study, performance analysis of generative adversarial network architectures that transform from image to image is made and its performance in synthetic image generation is evaluated. For a quality performance evaluation of these models, the denim2bıyık dataset collected from the real-world area was used instead of standardized datasets. Mustache patterns drawn on denim fabrics are created with a laser device. For this device to create the desired mustache pattern, it is necessary to work with visual editing programs for approximately 2-3 hours by specialized personnel. With the proposed approach, an automatic mustache production process will be realized, errors and time losses in manual production will be eliminated. As a result of our literature research, there is a no different study on the production of denim product images with productive networks. This situation increases the academic original value of the study. GAN architectures used in the study are Pix2Pix, CycleGAN, DiscoGAN, and AttentionGAN. Mustache pattern production performance evaluation and cost analysis were performed on the training and test data in the denim2bıyık dataset of each architecture. As a result of the studies, it is seen that the production speed of the mustache pattern image drops below one second, while the production accuracy reaches 86%.

PDF

___

[1] Das S., Dey A., Pal A., Roy N. 2015. Applications of artificial ıntelligence in machine learning: review and prospect. International Journal of Computer Applications, 115 (9), 31–41.
[2] LeCun Yann, et al. 1989. Backpropagation applied to handwritten zip code recognition. Neural Computation.
[3] Goodfellow I J., Pouget-Abadie J., Mirza M., Xu B., Warde-Farley D., Ozair S., Courville A., Bengio Y. 2014. Generative adversarial nets. Advances in Neural Information Processing Systems, 2672–2680.
[4] Radford A., Metz L., Chintala S. 2016. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. 4th International Conference on Learning Representations, ICLR 2016- Conference Track Proceedings, 1–16.
[5] Isola P., Zhu J Y., Zhou T., Efros A A. 2017. Image-to-Image Translation with Conditional Adversarial Networks. Proceedings- 30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, 5967–5976.
[6] Zhu J Y., Park T., Isola P., Efros A A. 2017. Unpaired Image-to-Image Translation Using CycleConsistent Adversarial Networks. Proceedings of the IEEE International Conference on Computer Vision, 2242–2251.
[7] Karras T., Aila T., Laine S., Lehtinen J. 2018. Progressive Growing of GANs for Improved Quality, Stability, and Variation. 6th International Conference on Learning Representations, ICLR 2018- Conference Track Proceedings, 1–25.
[8] Huang X., Belongie S. 2017. Arbitrary Style Transfer in Real-Time with Adaptive Instance Normalization. Proceedings of the IEEE International Conference on Computer Vision, 1510– 1519.
[9] Karras T., Laine S., Aila T. 2019. A Style-based Generator Architecture for Generative Adversarial Networks. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 4396–4405.
[10] Wang T C., Liu M Y., Zhu J Y., Tao A., Kautz J., Catanzaro B. 2018. High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 8798–8807.
[11] Park T., Liu M Y., Wang T C., Zhu J Y. 2019. Semantic Image Synthesis with Spatially-adaptive Normalization. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2332–2341.
[12] Dundar A., Sapra K., Liu G., Tao A., Catanzaro B. 2020. Panoptic-based Image Synthesis. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 8067–8076.
[13] Karras T., Aittala M., Laine S., Härkönen E., Hellsten J., Lehtinen J., Aila T. 2021. Alias-free generative adversarial networks. NeurIPS, 2106.12423.
[14] Wang X., Xie L., Dong C., Shan Y. 2021. Real-ESRGAN: Training Real-world Blind Superresolution with Pure Synthetic Data. Proceedings of the IEEE/CVF International Conference on Computer Vision, 1905-1914.
[15] Suvorov R., Logacheva E., Mashikhin A., Remizova A., Ashukha A., Silvestrov A., Kong N., Goka H., Park K., Lempitsky V. 2021. Resolution-robust large mask ınpainting with fourier convolutions. 2109.07161.
[16] Wang X., Li Y., Zhang H., Shan Y. 2021. Towards real-world blind face restoration with generative facial prior. 2101.04061.
[17] Huang H., Yu P S., Wang C. 2018. An Introduction to Image Synthesis with Generative Adversarial Nets. 1–17.
[18] Goodfellow I. 2016. Generative Adversarial Networks. NIPS 2016 Tutorial.
[19] Lazarou C. 2021. Generative Adversarial Networks. https://www.slideshare.net/ckmarkohchang/generative-adversarial-networks. (Erişim Tarihi: 20.04.2021)
[20] Ghosh A., Kumar H., Sastry P S. 2017. Robust Loss Functions under Label Noise for Deep Neural Networks. 31st AAAI Conference on Artificial Intelligence, AAAI 2017, 1919–1925.
[21] Ronneberger O., Fischer P., Brox T. 2015. U-Net: Convolutional Networks for Biomedical Image Segmentation. MICCAI.
[22] Mihelich M., Dognin C., Shu Y., Blot M. 2020. A Characterization of Mean Squared Error for Estimator with Bagging. ArXiv, abs/1908.02718.
[23] Kim T., Cha M., Kim H., Lee J. K., Kim J. 2017. Learning to Discover Cross-domain Relations with Generative Adversarial Networks. 34th International Conference on Machine Learning, ICML 2017, 4, 2941–2949.
[24] Mejjati Y A., Richardt C., Tompkin J., Cosker D. 2018. Unsupervised attention-guided ımage-toimage translation. NeurIPS 2018, 1–11. [25] Nilsson J., Akenine-Möller T. 2020. Understanding SSIM. ArXiv, abs/2006.13846.
[26] Fardo F A., Conforto V H., Oliveira F C., Rodrigues P. 2016. A formal evaluation of PSNR as quality measurement parameter for ımage segmentation algorithms. ArXiv, abs/1605.07116.