シモセラ エドガー　Siameseãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ãƒ¢ãƒ‡ãƒ«ã‚’用ã„ãŸç”»åƒç‰¹å¾´é‡æŠ½å‡ºSiameseãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ãƒ¢ãƒ‡ãƒ«ã‚’効率的ã«å¦ç¿’ã•ã›ã‚‹ã“ã¨ã§ã€ ãƒãƒã‚¹ãƒˆãªç”»åƒç‰¹å¾´é‡ã‚’計算ã™ã‚‹æ‰‹æ³•ã‚’æ案ã™ã‚‹ã€‚ æ案手法ã§ã¯ã€ãƒ¢ãƒ‡ãƒ«ã«2ã¤ã®ç”»åƒãƒ‘ッãƒã‚’入力ã—ã€å‡ºåŠ›ã•ã‚ŒãŸç‰¹å¾´é‡ã®èª¤å·®ã«ã‚ˆã£ã¦ãƒ¢ãƒ‡ãƒ«ã‚’å¦ç¿’ã•ã›ã‚‹ã€‚ ã¾ãŸã€å…¥åŠ›ã™ã‚‹ãƒ‘ッãƒã‚’ãã®è˜åˆ¥ã®é›£ã—ã•ã«ã‚ˆã£ã¦åˆ†é¡žã—ã€è˜åˆ¥ãŒå›°é›£ãªãƒ‘ッãƒã‚’優先的ã«å¦ç¿’ã•ã›ã‚‹ã“ã¨ã§ã€SIFT特徴é‡ã‚ˆã‚Šã‚‚ãƒãƒã‚¹ãƒˆãªç‰¹å¾´é‡ã®æŠ½å‡ºã‚’実ç¾ã—ãŸã€‚ Our approach consists in training a Convolutional Neural Network (CNN) to build a feature representation of an image patch. We train by using two patches simultaneously that should either correspond to the same point
{{#tags}}- {{label}}
{{/tags}}