| Autor: |
Minarno, Agus Eko, Antoko, Toton Dwi, Azhar, Yufis |
| Předmět: |
|
| Zdroj: |
International Journal on Advanced Science, Engineering & Information Technology; 2023, Vol. 13 Issue 1, p348-356, 9p |
| Abstrakt: |
Every craftsman who draws Batik can also not necessarily draw various types of Batik. On the other hand, it takes a long time ranging from weeks to months, to make Batik. Image generation is regarded as an essential part of the field of computer vision. One of the popular methods includes the Generative Adversarial Network, commonly implemented to generate a new data set from an existing one. One model of the Generative Adversarial Network is BatikGAN SL generating batik images by inserting the two Batik patterns to produce a new Batik image. Currently, the generated Batik image does not maintain the input of the Batik pattern. Therefore, this study proposes a GAN model of BatikGAN SL, with the addition of a content loss function using hyperparameters to weight the content loss function. The content loss function is added from the Neural Transfer Style method. Previously, the style loss function in this method has been implemented in BatikGAN SL, and the dataset consists of Batik patches (326 images) and real Batik (163 images). This paper compares the BatikGAN SL model from previous studies with the BatikGAN SL model by implementing hyperparameters on the content loss function. The evaluation is conducted with FID, containing FID Local and FID Global. The results obtained in this study include a collection of Batik images, test evaluation value of 42 on FID Global and 16 on FID Local. These results are obtained by implementing the content loss function with a weight value of 1. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
