谷歌出品。移动端实时人脸检测，针对移动端GPU做优化，网络结构为mobilenet+ssd的变体，针对前置摄像头做人脸检测，在旗舰设备上检测速度为200-1000FPS，相当有吸引力。
网络结构，用5×5的卷积核替代3×3的卷积核，增大感受野，提出一种BlazeBlock结构
balzeblock 2019-07-15 上午10.12.28.png
anchor，最小featuremap为8×8，每个像素对应的anchor由2个增加到6个
anchor.png
后处理，不再使用nms，而是blending
network.png
speed1.png
speed2.png
谷歌出品。移动端实时人脸检测，针对移动端GPU做优化，网络结构为mobilenet+ssd的变体，针对前置摄像头做...
Paper Reading Note URL:论文pdf TL;DR Google出品。亚毫秒级的移动端人脸检测算...
The satellite will be used to prove that a hardened neura...
原文：《ImageNet Classification with Deep Convolutional Neura...
Learning both Weights and Connections for Efficient Neura...
转载 原文地址：Visualizing and understanding convolutional neura...
27组-ShuffleNet：An Extremely Efficient Convolutional Neura...
1 AUTOGRAD AUTOMATIC DIFFERENTIATION Central to all neura...
Deep Structured Semantic Model (DSSM) [1] is a deep neura...
Profiling based Out-of-core Hybrid Method for Large Neura...
本文标题：BlazeFace: Sub-millisecond Neura
本文链接：https://www.haomeiwen.com/subject/nkugkctx.html
网友评论