• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

Çмú´ëȸ ÇÁ·Î½Ãµù

Ȩ Ȩ > ¿¬±¸¹®Çå > Çмú´ëȸ ÇÁ·Î½Ãµù > Çѱ¹Á¤º¸Åë½ÅÇÐȸ Çмú´ëȸ > 2019³â Ãß°èÇмú´ëȸ

2019³â Ãß°èÇмú´ëȸ

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) VGG ³×Æ®¿öÅ© ±â¹Ý µö·¯´× ¾Ë°í¸®Áò¿¡¼­ ÁÖ °´Ã¼ À§Ä¡ °ËÃâ ¹æ¹ý¿¡ °üÇÑ ¿¬±¸
¿µ¹®Á¦¸ñ(English Title) A Study on the Detection of Main Object Location in the Deep Learning Algorithm Based on VGG
ÀúÀÚ(Author) ±è¼±Áø   ÀÌÁ¾±Ù   ¾ÈÀçÇü   Seon-jin Kim   Jong-keun Lee   Jae-hyung Ahn  
¿ø¹®¼ö·Ïó(Citation) VOL 23 NO. 02 PP. 0641 ~ 0643 (2019. 10)
Çѱ۳»¿ë
(Korean Abstract)
º» ³í¹®¿¡¼­´Â VGG19¸¦ ±â¹ÝÀ¸·Î ÇÑ ³×Æ®¿öÅ© ±¸Á¶¿¡¼­ ¾àÇÑ ÁöµµÇнÀÀ» ÀÌ¿ëÇÑ ÁÖ °´Ã¼ÀÇ À§Ä¡ °ËÃâ ¹æ¹ýÀ» Á¦¾ÈÇÑ´Ù. »ç¿ëÇÑ ³×Æ®¿öÅ©´Â VGG19¿¡ Ãß°¡ÀûÀÎ Ç®¸µ °èÃþ°ú ÇÕ¼º°ö °èÃþÀ» Ãß°¡ÇÏ¿© ±¸¼ºÇÏ¿´´Ù. ¿µ»ó ³» Á¸ÀçÇÏ´Â ´Ù¼öÀÇ °´Ã¼ Áß ÁÖ °´Ã¼ÀÇ À§Ä¡ °ËÃâÀ» À§ÇØ Grad-CAM ±â¹ýÀ» Àû¿ëÇÏ¿´´Ù. ½ÇÇè ¼º´ÉÀº À̹ÌÁö ºÐ·ù¿Í ÁÖ °´Ã¼ÀÇ À§Ä¡ °ËÃâ È¿À²À» ³ª´©¾î ÃøÁ¤ÇÏ¿´´Ù. ÁÖ °´Ã¼ÀÇ À§Ä¡ °ËÃâ ¼º´ÉÀº Top-1 Localization err¿¡¼­ Pascal-voc2012ÀÇ °æ¿ì 37.68%, Dogs and catsÀÇ °æ¿ì 8.926%ÀÇ ¼º´ÉÀ» º¸¿´´Ù.
¿µ¹®³»¿ë
(English Abstract)
In this paper, we propose a method of detection of main objects location using weakly supervised learning in network structure based on VGG19. The network structure was constructed by adding an additional pooling layer and convolution layer to the VGG19 network. We applied the Grad-CAM for main object detection among multiple objects in a image. The performance of the experiment is measured by dividing image classification and efficiency of location detection of main object. The image classification performance is similar to that of the VGG19 network, and the location detection performance of the main object is 37.68%, 8.926% in term of the Top-1 loc err for a each dataset.
Å°¿öµå(Keyword) Weakly-Supervised learning   Deep learning   Object localization   VGG19   ILSVRC  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå