Derin Öğrenme Modellerinin Sinirsel Stil Aktarımı Performanslarının Karşılaştırılması

KARADAĞ B., ARI A., KARADAĞ M.

Journal of Polytechnic, cilt.24, ss.1611-1622, 2021 (ESCI) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 24
  • Basım Tarihi: 2021
  • Doi Numarası: 10.2339/politeknik.885838
  • Dergi Adı: Journal of Polytechnic
  • Derginin Tarandığı İndeksler: Emerging Sources Citation Index (ESCI), TR DİZİN (ULAKBİM)
  • Sayfa Sayıları: ss.1611-1622
  • Anahtar Kelimeler: Neural style transfer, convolutional neural network, deep convolutional neural network
  • İnönü Üniversitesi Adresli: Evet

Özet

Neural style transfer is one of the most studied topics in both academic and industrial fields. Quality and performance enhancement are among the most targeted goals in the studies. In this study, the performance of different CNN models in neural style transfer was investigated. Deep features were obtained using VGG16, VGG19 and ResNet50 models. Thanks to these attributes, a new target image is created by taking the content of the content image and the style of the style image. Adam, RMSprop and SGD optimization algorithms are used. In neural transfer studies, the best visual performance was obtained from VGG19 network model by using SGD optimization algorithm. The fastest neural style transfer in terms of time was obtained using the SGD optimization algorithm in the ResNet50 convolutional neural network model.