Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
ºñµð¿À ºÎºÐ º¹»ç °ËÃâÀ» À§ÇÑ ¼¼±×¸ÕÆ® ´ÜÀ§ Vision Transformer ±â¹Ý Ư¡ º¤ÅÍ À¶ÇÕ ¹æ¹ý |
¿µ¹®Á¦¸ñ(English Title) |
A Fusion Methods of Segment-level Vision Transformer based Feature Vector for Video Partial Copy Detection |
ÀúÀÚ(Author) |
°¹Î¿µ
°¼ö¿¬
³¶Á¾È£
Minyoung Kang
Sooyeon Kang
Jongho Nang
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 48 NO. 02 PP. 0491 ~ 0493 (2021. 12) |
Çѱ۳»¿ë (Korean Abstract) |
ÃÖ±Ù ¸ÖƼ¹Ìµð¾î Ç÷§ÆûÀÌ È°¼ºÈµÊ¿¡ µû¶ó ºñµð¿À ÄÁÅÙÃ÷ ½ÃÀåÀÇ ±Ô¸ð°¡ Áõ°¡ÇÏ°í ÀÖ´Ù. ÇÏÁö¸¸ ÀÌ¿¡ µû¶ó ¹«´Ü º¹Á¦³ª À¯Æ÷¿Í °°ÀÌ ÀúÀÛ±ÇÀ» ħÇØÇÏ´Â ¹®Á¦°¡ ¹ß»ýÇÏ°í ÀÖ´Ù. ÀÌ·± ¹®Á¦¸¦ ÇØ°áÇϱâ À§ÇÑ ¹æ¹ýµéÀÌ Á¦¾ÈµÇ¾ú´Âµ¥ ´Ù¾çÇÑ º¯ÇüÀÌ µîÀåÇÔ¿¡ µû¶ó ½ÇÁ¦ º¹»ç ºñµð¿À °ËÃâ¿¡ ½ÇÆÐÇÏ´Â °æ¿ì°¡ Áõ°¡ÇÏ¿´´Ù. µû¶ó¼, º» ³í¹®¿¡¼´Â ´Ù¾çÇÑ º¯Çü¿¡ °°ÇÇÑ ºñµð¿À º¹»ç °ËÃ⠽ýºÅÛÀ» ¼³°èÇÏ°íÀÚ Vision Transformer [1]¸ðµ¨À» »ç¿ëÇÏ¿© ºñµð¿ÀÀÇ °ø°£ Á¤º¸¸¦ À¶ÇÕÇÏ°í, Frame Stitching°ú Max Pooling ¹æ¹ýÀ» »ç¿ëÇÏ¿© ºñµð¿ÀÀÇ ½Ã°£ Á¤º¸¸¦ À¶ÇÕÇÏ´Â ¹æ¹ýÀ» Á¦¾ÈÇÑ´Ù. À̸¦ ±âÁ¸ÀÇ CNN Local Ư¡ º¤Å͸¦ ÃßÃâÇÏ¿© Bag-of-local Feature(BoF)·Î À¶ÇÕÀ» ÇÏ¿© ºñµð¿À º¹»ç °ËÃâÀ» ¼öÇàÇÏ´Â ¹æ¹ý[2]°ú ºñ±³ÇÏ¿© °¢ À¶ÇÕ ¹æ¹ý º° ¼º´ÉÀ» ºñ±³ÇÑ´Ù. ½ÇÇèÀ» ÅëÇØ º» ³í¹®¿¡¼ Á¦¾ÈÇÏ´Â Vision Transformer ±â¹ÝÀÇ Max PoolingÀ» »ç¿ëÇÑ Æ¯Â¡ º¤ÅÍ À¶ÇÕ ¹æ¹ýÀÌ Æò±Õ µî¼ö 953µîÀ¸·Î BoF ¹æ¹ý[2]ÀÇ µî¼öº¸´Ù 30% ¼º´ÉÀÌ Áõ°¡ÇÏ¿´´Ù. |
¿µ¹®³»¿ë (English Abstract) |
|
Å°¿öµå(Keyword) |
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|