Á¤º¸°úÇÐȸ³í¹®Áö (Journal of KIISE)
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
ÀúÇØ»óµµ À̹ÌÁö ºÐ·ù¸¦ À§ÇÑ °íÇØ»óµµ À̹ÌÁö·ÎºÎÅÍÀÇ Self-Attention Á¤º¸ ÃßÃâ ³×Æ®¿öÅ© |
¿µ¹®Á¦¸ñ(English Title) |
Low-Resolution Image Classification Using Knowledge Distillation From High-Resolution Image Via Self-Attention Map |
ÀúÀÚ(Author) |
½Å¼ºÈ£
ÀÌÁÖ¼ø
ÀÌÁؼ®
ÃÖ½ÂÁØ
À̱Ժó
Sungho Shin
Joosoon Lee
Junseok Lee
Seungjun Choi
Kyoobin Lee
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 47 NO. 11 PP. 1027 ~ 1031 (2020. 11) |
Çѱ۳»¿ë (Korean Abstract) |
±âÁ¸ÀÇ µö·¯´× ¸ðµ¨µéÀº °íÈÁúÀÇ À̹ÌÁöµéÀ» È°¿ëÇÏ¿© ¿¬±¸ °³¹ßµÇ¾úÀ¸¸ç, ÈÁúÀÌ ³·¾ÆÁú¼ö·Ï ±Þ°ÝÈ÷ ¼º´ÉÀÌ ³·¾ÆÁø´Ù. º» ¿¬±¸´Â ÀúÈÁú À̹ÌÁö¿¡µµ È¿°úÀûÀ¸·Î ´ëÀÀÇÒ ¼ö ÀÖ´Â µö·¯´× ¸ðµ¨À» °³¹ßÇÏ°íÀÚ, °íÈÁúÀÇ À̹ÌÁö·ÎºÎÅÍ È¿°úÀûÀ¸·Î ºÐ·ù¸¦ ÇÒ ¼ö ÀÖ´Â Á¤º¸¸¦ Attention MapÀÇ ÇüÅ·ΠÃßÃâÇß´Ù. ÀÌÈÄ Knowledge Distillation ±â¹ýÀ» È°¿ëÇÏ¿© °íÈÁú À̹ÌÁö»ó¿¡¼ ÃßÃâÇÑ Attention MapÀ» ÀúÇØ»óµµ À̹ÌÁö ¸ðµ¨¿¡ Àü´ÞÇÏ´Â ³×Æ®¿öÅ©¸¦ Á¦¾ÈÇßÀ¸¸ç, 16¡¿16ÀÇ ÀúÇØ»óµµ CIFAR100¡¡À̹ÌÁö¸¦ ºÐ·ùÇßÀ» ¶§ ¿¡·¯À²À» 2.94% ³·Ãâ ¼ö ÀÖ¾ú´Ù. ÀÌ´Â 32¡¿32¿¡¼ 16¡¿16À¸·Î À̹ÌÁö Çػ󵵸¦ ³·ÃèÀ» ¶§ ¿¡·¯ °¨¼ÒÀ²ÀÇ 38.43%¿¡ ÇØ´çÇÏ´Â ¼öÄ¡·Î, º» ³×Æ®¿öÅ©ÀÇ ¿ì¼ö¼ºÀ» ÀÔÁõÇÒ ¼ö ÀÖ¾ú´Ù. |
¿µ¹®³»¿ë (English Abstract) |
Traditional deep-learning models have been developed using high-quality images. However, when the low resolution images are rendered, the performances of the model drop drastically. To develop a deep-learning model that can respond effectively to low-resolution images, we extracted the information from the model, which uses high-resolution images as input, in the form of the Attention Map. Using the knowledge distillation technique, the information delivering Attention Map, extracted from the high-resolution images to low-resolution image models, could reduce the error rate by 2.94%, when classifying the low-resolution CIFAR images of 16¡¿16 resolution. This was at 38.43% of the error reduction rate when the image resolution was lowered from 32¡¿32 to 16¡¿16, which could demonstrate excellence in this network. |
Å°¿öµå(Keyword) |
ÀúÇØ»óµµ À̹ÌÁö
À̹ÌÁö ºÐ·ù
Á¤º¸ Àü´Þ ±â¹ý
low resolution image
image classification
knowledge distillation
self-attention map
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|