在Faster R-CNN 中DEMO 的CPU(i5)和GPU(GTX1060 )时间对比

原创 2016年08月31日 10:27:04

软硬件环境:

Ubuntu 14.04 64bit
NVIDIA GeForce GTX 1060 6GB
Intel® Core™ i5-6500 CPU @ 3.20GHz × 4
8GB memory
CUDA 8.0
cuDNN 5.0

zf net使用CPU的时间:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000456.jpg
Detection took 3.800s for 300 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000542.jpg
Detection took 3.013s for 135 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001150.jpg
Detection took 3.423s for 231 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001763.jpg
Detection took 3.215s for 200 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/004545.jpg
Detection took 3.788s for 300 object proposals

zf net使用GPU的时间:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000456.jpg
Detection took 0.069s for 300 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000542.jpg
Detection took 0.051s for 135 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001150.jpg
Detection took 0.058s for 231 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001763.jpg
Detection took 0.057s for 200 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/004545.jpg
Detection took 0.064s for 300 object proposals

VGG16 net 使用CPU的时间

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000456.jpg
Detection took 17.795s for 300 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000542.jpg
Detection took 15.609s for 161 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001150.jpg
Detection took 16.014s for 194 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001763.jpg
Detection took 15.853s for 196 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/004545.jpg
Detection took 16.921s for 300 object proposals

VGG16 net 使用GPU的时间

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000456.jpg
Detection took 0.162s for 300 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000542.jpg
Detection took 0.132s for 161 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001150.jpg
Detection took 0.148s for 194 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001763.jpg
Detection took 0.145s for 196 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/004545.jpg
Detection took 0.162s for 300 object proposals

结论:

VGG16 差距达到100倍!

Note: CPU只用了一个核心。

版权声明:本文为博主原创文章,未经博主允许不得转载。

相关文章推荐

cuDNN兼容性问题造成的caffe/mnist,py-faster-rcnn/demo运行结果错误

问题描述我有两台电脑,一台笔记本GTX965M显卡,台式机是GTX1060显卡 两台电脑上的软件环境都一样:ubuntu16+cuda8.0+cuDNN4,显卡驱动nvidia-378 在笔记本上...
  • 10km
  • 10km
  • 2017-03-16 13:30
  • 1322

faster-rcnn 在第一阶段训练fast出现的假死 step(1)

我用的是GTX1060,使用CUDA7.5会产生不兼容,测试是没有问题的 训练时发生在第一阶段的fast训练时 step(1)出错,GTX10系列的,需要CUDA8.0的支持。当然出这个错误不一定是两...

卷积神经网络(三):卷积神经网络CNN的简单实现(部分Python源码)

上周末利用python简单实现了一个卷积神经网络,只包含一个卷积层和一个maxpooling层,pooling层后面的多层神经网络采用了softmax形式的输出。实验输入仍然采用MNIST图像使用10...

Faster R-CNN 和最新的版本cuDNN V5.0不兼容问题

Faster R-CNN 和最新的版本cuDNN V5.0不兼容问题软件硬件环境: Ubuntu 14.04 64bit NVIDIA GTX1060 cuda 8.0 cuDNN 5.0...
  • RZJMPB
  • RZJMPB
  • 2016-08-30 21:09
  • 3502

Faster R-CNN安装笔记,只用CPU

官网下载 faster RCNN, https://github.com/rbgirshick/py-faster-rcnn 按教程操作,装pythonjie

Faster R-CNN CPU环境搭建(h 参考 成功)

操作系统: bigtop@bigtop-SdcOS-Hypervisor:~/py-faster-rcnn/tools$ cat /etc/issue Ubuntu 14.04.2 LTS \n ...

Ubuntu14.04下配置测试Faster R-CNN+Caffe(Matlab+CPU)目标检测

Faster R-CNN是当前目标检测领域内性能最好的算法之一,它将RPN(Region Proposal Network)网络和Fast R-CNN网络结合到了一起,实现了一个端到端的目标检测框架。...

Faster R-CNN安装笔记,只用CPU

转自:http://blog.sina.com.cn/s/blog_679f93560102wpyf.html 下载代码和数据 git clone --recursive https://...

cuda8+cuDNN Faster R-CNN安装塈运行demo

安装cudacuda8安装参见网上教程安装cudnnCaffe目前的代码貌似不支持cudnn5,如果安装cudnn5,在编译caffe时会报错: error: argument of type “...
  • 10km
  • 10km
  • 2017-03-14 16:23
  • 1164

Faster-R-CNN(Python).1:从配置到demo

本文主要讲解Faster-R-CNN(Python)配置过程中遇到的问题。
内容举报
返回顶部
收藏助手
不良信息举报
您举报文章:深度学习:神经网络中的前向传播和反向传播算法推导
举报原因:
原因补充:

(最多只允许输入30个字)