Á¤º¸Ã³¸®ÇÐȸ ³í¹®Áö ¼ÒÇÁÆ®¿þ¾î ¹× µ¥ÀÌÅÍ °øÇÐ
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
¸ð¹ÙÀÏ µð¹ÙÀ̽º ȸéÀÇ Å¬¸¯ °¡´ÉÇÑ °´Ã¼ ŽÁö¸¦ À§ÇÑ ½Ì±Û ¼¦ µðÅØÅÍ |
¿µ¹®Á¦¸ñ(English Title) |
Single Shot Detector for Detecting Clickable Object in Mobile Device Screen |
ÀúÀÚ(Author) |
Á¶¹Î¼®
ÀüÇý¿ø
ÇѼº¼ö
Á¤Ã¢¼º
Min-Seok Jo
Hye-won Chun
Seong-Soo Han
Chang-Sung Jeong
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 11 NO. 01 PP. 0029 ~ 0034 (2022. 01) |
Çѱ۳»¿ë (Korean Abstract) |
¸ð¹ÙÀÏ µð¹ÙÀ̽º ȸé»óÀÇ Å¬¸¯ °¡´ÉÇÑ °´Ã¼¸¦ ÀÎÁöÇϱâ À§ÇÑ µ¥ÀÌÅͼÂÀ» ±¸ÃàÇÏ°í »õ·Î¿î ³×Æ®¿öÅ© ±¸Á¶¸¦ Á¦¾ÈÇÑ´Ù. ¸ð¹ÙÀÏ µð¹ÙÀ̽º ȸ鿡¼ Ŭ¸¯ °¡´ÉÇÑ °´Ã¼¸¦ ±âÁØÀ¸·Î ´Ù¾çÇÑ Çػ󵵸¦ °¡Áø µð¹ÙÀ̽º¿¡¼ ¿©·¯ ¾ÖÇø®ÄÉÀ̼ÇÀ» ´ë»óÀ¸·Î µ¥ÀÌÅ͸¦ ¼öÁýÇÏ¿´´Ù. ÃÑ 24,937°³ÀÇ annotation data¸¦ text, edit text, image, button, region, status bar, navigation barÀÇ 7°³ Ä«Å×°í¸®·Î ¼¼ºÐÈÇÏ¿´´Ù. ÇØ´ç µ¥ÀÌÅͼÂÀ» ÇнÀÇϱâ À§ÇÑ ¸ðµ¨ ±¸Á¶´Â Deconvolution Single Shot Detector¸¦ º£À̽º¶óÀÎÀ¸·Î, backbone network´Â ±âÁ¸ ResNet¿¡ Squeeze-and-Excitation blockÀ» Ãß°¡ÇÑ Squeeze-and-Excitation networks¸¦ »ç¿ëÇÏ°í, Single shot detector layers¿Í Deconvolution moduleÀ» Feature pyramid networks ÇüÅ·Π½×¾Æ ¿Ã·Á header¿Í ¿¬°áÇÑ´Ù. ¶ÇÇÑ, ±âÁ¸ input resolutionÀÇ 1:1 ºñÀ²¿¡¼ ¿À´Â Ư¡ÀÇ ¼Õ½ÇÀ» ÃÖ¼ÒÈÇϱâ À§ÇØ ¸ð¹ÙÀÏ µð¹ÙÀ̽º ȸé°ú À¯»çÇÑ 1:2 ºñÀ²·Î º¯°æÇÏ¿´´Ù. ÇØ´ç ¸ðµ¨À» ±¸ÃàÇÑ µ¥ÀÌÅͼ¿¡ ´ëÇÏ¿© ½ÇÇèÇÑ °á°ú º£À̽º¶óÀο¡ ´ëºñÇÏ¿© mean average precisionÀÌ ÃÖ´ë 101% °³¼±µÇ¾ú´Ù. |
¿µ¹®³»¿ë (English Abstract) |
We propose a novel network architecture and build dataset for recognizing clickable objects on mobile device screens. The data was collected based on clickable objects on the mobile device screen that have numerous resolution, and a total of 24,937 annotation data were subdivided into seven categories: text, edit text, image, button, region, status bar, and navigation bar. We use the Deconvolution Single Shot Detector as a baseline, the backbone network with Squeeze-and-Excitation blocks, the Single Shot Detector layer structure to derive inference results and the Feature pyramid networks structure. Also we efficiently extract features by changing the input resolution of the existing 1:1 ratio of the network to a 1:2 ratio similar to the mobile device screen. As a result of experimenting with the dataset we have built, the mean average precision was improved by up to 101% compared to baseline. |
Å°¿öµå(Keyword) |
Å×½ºÆ® ÀÚµ¿È
¾Èµå·ÎÀÌµå °´Ã¼ ŽÁö
¸ð¹ÙÀÏ È¸é ÀÎÁö
ÄÄÇ»ÅÍ ºñÀü
µö·¯´×
Test Automation
Android Object Detection
Test Automation
Android Object Detection
Computer Vision
Deep Learning
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|