Á¤º¸°úÇÐȸ³í¹®Áö (Journal of KIISE)
ÇѱÛÁ¦¸ñ(Korean Title) |
¹°Ã¼ °ËÃâ ÄÁ¹ú·ç¼Ç ½Å°æ¸Á ¼³°è¸¦ À§ÇÑ È¿°úÀûÀÎ ³×Æ®¿öÅ© ÆĶó¹ÌÅÍ ÃßÃâ |
¿µ¹®Á¦¸ñ(English Title) |
Searching Effective Network Parameters to Construct Convolutional Neural Networks for Object Detection |
ÀúÀÚ(Author) |
±è´©¸®
À̵¿ÈÆ
¿À¼ºÈ¸
Nuri Kim
Donghoon Lee
Songhwai Oh
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 44 NO. 07 PP. 0668 ~ 0673 (2017. 07) |
Çѱ۳»¿ë (Korean Abstract) |
ÃÖ±Ù ¸î ³â°£ µö·¯´×(deep learning)Àº À½¼º ÀνÄ, ¿µ»ó ÀνÄ, ¹°Ã¼ °ËÃâÀ» ºñ·ÔÇÑ ´Ù¾çÇÑ ÆÐÅÏÀÎ½Ä ºÐ¾ß¿¡¼ Çõ½ÅÀûÀÎ ¼º´É ¹ßÀüÀ» °ÅµìÇØ¿Ô´Ù. ±×¿¡ ºñÇØ ³×Æ®¿öÅ©°¡ ¾î¶»°Ô ÀÛµ¿ÇÏ´ÂÁö¿¡ ´ëÇÑ ±íÀº ÀÌÇØ´Â Àß ÀÌ·ç¾îÁöÁö ¾Ê°í ÀÖ´Ù. º» ³í¹®Àº È¿°úÀûÀÎ ½Å°æ¸Á ³×Æ®¿öÅ©¸¦ ±¸¼ºÇϱâ À§ÇØ ³×Æ®¿öÅ© ÆĶó¹ÌÅ͵éÀÌ ½Å°æ¸Á ³»ºÎ¿¡¼ ¾î¶»°Ô ÀÛµ¿ÇÏ°í, ¾î¶² ¿ªÇÒÀ» ÇÏ°í ÀÖ´ÂÁö ºÐ¼®ÇÏ¿´´Ù. Faster R-CNN ³×Æ®¿öÅ©¸¦ ±â¹ÝÀ¸·Î ÇÏ¿© ½Å°æ¸ÁÀÇ °úÀûÇÕ(overfitting)À» ¸·´Â µå¶ø¾Æ¿ô(dropout) È®·ü°ú ¾ÞÄ¿ ¹Ú½º Å©±â, ±×¸®°í È°¼º ÇÔ¼ö¸¦ º¯È½ÃÄÑ ÇнÀÇÑ ÈÄ ±× °á°ú¸¦ ºÐ¼®ÇÏ¿´´Ù. ¶ÇÇÑ µå¶ø¾Æ¿ô°ú ¹èÄ¡ Á¤±ÔÈ(batch normalization) ¹æ½ÄÀ» ºñ±³Çغ¸¾Ò´Ù. µå¶ø¾Æ¿ô È®·üÀº 0.3ÀÏ ¶§ °¡Àå ÁÁÀº ¼º´ÉÀ» º¸¿´À¸¸ç ¾ÞÄ¿ ¹Ú½ºÀÇ Å©±â´Â ÃÖÁ¾ ¹°Ã¼ °ËÃâ ¼º´É°ú Å« °ü·ÃÀÌ ¾ø´Ù´Â °ÍÀ» ¾Ë ¼ö ÀÖ¾ú´Ù. µå¶ø¾Æ¿ô°ú ¹èÄ¡ Á¤±ÔÈ ¹æ½ÄÀº ¼·Î¸¦ ¿ÏÀüÈ÷ ´ëüÇÒ ¼ö´Â ¾ø´Â °ÍÀ» È®ÀÎÇÒ ¼ö ÀÖ¾ú´Ù. È°¼ºÈ ÇÔ¼ö´Â À½¼ö µµ¸ÞÀÎÀÇ ±â¿ï±â°¡ 0.02ÀÎ leaky ReLU°¡ ºñ±³Àû ÁÁÀº ¼º´ÉÀ» º¸¿´´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
Deep neural networks have shown remarkable performance in various fields of pattern recognition such as voice recognition, image recognition and object detection. However, underlying mechanisms of the network have not been fully revealed. In this paper, we focused on empirical analysis of the network parameters. The Faster R-CNN(region-based convolutional neural network) was used as a baseline network of our work and three important parameters were analyzed: the dropout ratio which prevents the overfitting of the neural network, the size of the anchor boxes and the activation function. We also compared the performance of dropout and batch normalization. The network performed favorably when the dropout ratio was 0.3 and the size of the anchor box had not shown notable relation to the performance of the network. The result showed that batch normalization can¡¯t entirely substitute the dropout method. The used leaky ReLU(rectified linear unit) with a negative domain slope of 0.02 showed comparably good performance.
|
Å°¿öµå(Keyword) |
µö·¯´×
¹°Ã¼°ËÃâ ³×Æ®¿öÅ©
ÄÁ¹ú·ç¼Ç ³×Æ®¿öÅ©
ÇÏÀÌÆÛÆĶó¹ÌÅÍ
deep learning
detection network
convolutional network
hyper parameter
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|