Network Structure: Backpropagation applied to handwritten zip code recognition

Network description

  • H1: 768(8*8*12) units, 1068 parameters(768+25*12)(bias associated with units, not kernel)

Comments

Post a Comment

Popular posts from this blog

Uknown words

Unknown words tsunami : 海啸 hyperbolic : 曲线的；夸张的 supersede : 取代；超越 meteoric : 大气的；流星的；极速的 reign : 统治；支配 unyielding : 不服的；坚强的 redeem : 救赎；挽回 primer : 初级读本；原始物 wimpy : 懦弱的；无用的 nemesis : 复仇女神；报应 folly : 愚蠢；荒唐 ostensibly : 表面；外表 synapse : 突触 irk : 惹恼；厌烦 SNARC hype : 大肆宣传；皮下兴奋 thaw : 溶解；变暖和 oval : 椭圆 folly : 愚蠢；荒唐 ostensibly : 表面；外表 synapse : 突触 irk : 惹恼；厌烦 escort :护送 escort :远征 探险 lousy :讨厌的
Read more

Thesis writing note

Chapter 1 Description for projected light sensor 3D sensing Jabil time-of-flight sensing sloa Overview projected light sensor https://medium.com/advanced-computer-vision/3d-sensors-what-is-out-there-5e65fd3e8afe
Read more

Object detection

Difference of regress losses in Both fast rcnn and faster rcnn fast rcnn. The bounding box regression is based on the region proposal (obtained from selective search) faster rcnn. The bounding box regression is based on anchor boxes (all the ground truth is also transformed based on anchor boxes)
Read more