2019³â Ãß°èÇмú´ëȸ
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
GANÀ» ÀÌ¿ëÇÑ À̹ÌÁö ºÐ·ù ³×Æ®¿öÅ©ÀÇ ÇнÀ µ¥ÀÌÅͼ ȮÀå |
¿µ¹®Á¦¸ñ(English Title) |
Expanding Training Datasets for Image Classification Network using GAN |
ÀúÀÚ(Author) |
Á¤ÁØÈ£
±èÁö¿ì
±èÇöÅÂ
Junho Jeong
Jiwoo Kim
Hyuntae Kim
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 23 NO. 02 PP. 0075 ~ 0076 (2019. 10) |
Çѱ۳»¿ë (Korean Abstract) |
º» ³í¹®¿¡¼´Â GAN°ú DCGANÀ» »ç¿ëÇÏ¿© MNIST¿Í CIFAR-10ÀÇ µ¥ÀÌÅͼÂÀ» È®ÀåÇÑ ÈÄ VGG-16 ³×Æ®¿öÅ©·Î ÇнÀÇÏ¿© ¿øº» µ¥ÀÌÅͼÂÀ¸·Î ÇнÀÇÏ¿´À» ¶§¿ÍÀÇ ¼º´ÉÀ» ºñ±³ÇÏ¿´´Ù. GANÀº MNIST¿Í CIFAR-10 µ¥ÀÌÅͼ ¸ðµÎ¿¡¼ ¼º´ÉÀÌ ´õ¿í Ç϶ôÇÏ¿´°í, DCGANÀº MNIST µ¥ÀÌÅͼ¿¡¼¸¸ ºÐ·ù¼º´ÉÀÌ Çâ»óµÇ¾ú°í, CIFAR-10 µ¥ÀÌÅͼ¿¡¼ ¼º´ÉÀÌ Ç϶ôÇÏ¿´´Ù. À̸¦ ÅëÇØ GANÀ» ÅëÇØ »ý¼ºÇÑ µ¥ÀÌÅͼÂÀÌ ³×Æ®¿öÅ©ÀÇ ºÐ·ù ¼º´É¿¡ ¿µÇâÀ» ¹ÌħÀ» º¸¿´Áö¸¸, ¼º´ÉÀÌ Ç×»ó Çâ»óµÇÁö´Â ¾ÊÀ½À» º¸¿´´Ù. |
¿µ¹®³»¿ë (English Abstract) |
In this paper, we trained VGG-16 networks with original/expanded MNIST and CIFAR-10 datasets. when we compared these network¡¯s performance, the dataset expanded with GAN shows reduced performance at all datasets. The MNIST dataset which expanded with DCGAN¡¯s result was improved, but CIFAR-10 shows reduced performance. It shows the expanded datasets with GAN affects to deep neural network¡¯s performance, but not means always improve it. |
Å°¿öµå(Keyword) |
deep neural network
gan
dcgan
dataset
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|