dongzhuoyao

Updated 4/11/2018

Here's my experience of installing the NVIDIA CUDA kit 9.0 on a fresh install of Ubuntu Desktop 16.04.4 LTS.

Table of Contents

• Install NVIDIA Graphics Driver via apt-get
• Install CUDA
• Install cuDNN

Table of contents generated with markdown-toc

dongzhuoyao

Felix Krull runs a PPA offering basically any version of Python (seriously, there is 2.3.7 build for vivid...) for many Ubuntu releases at

https://launchpad.net/~fkrull/+archive/ubuntu/deadsnakes

Do the usual:

sudo add-apt-repository ppa:fkrull/deadsnakes
sudo apt-get update
sudo apt-get install python3.5
It will not overwrite your existing python3.4 which is still symlinked as python3. If you want python3.5 to be the default python3, change the symlink

dongzhuoyao

name: "VGG_ILSVRC_16_layers"
layer {
  name: 'input-data'
  type: 'Python'
  top: 'data'
  top: 'im_info'
  top: 'gt_boxes'
  top: 'need_backprop'
  top: 'dc_label'
  python_param {

dongzhuoyao

name: "VGG_coco_SSD_300x300_train"
layer {
  name: "data"
  type: "AnnotatedData"
  top: "data"
  top: "label"
  include {
    phase: TRAIN
  }
  transform_param {

dongzhuoyao

layer {
  name: "first_img"
  type: "Input"
  top: "first_img"
  top: "second_img"
  top: "first_label"
  top: "second_label"
  input_param {
    shape {
      dim: 1

dongzhuoyao

使用规范

服务器配置: cuda 8.0, cudnn6.0.21

1.需要sudo权限的同学，简单操作我帮你做，复杂操作我暂时给你sudo权限，过后收回。
（系统崩了就得重装，重装系统虽简单，但是琐碎且耗时，因此为了管理方便，sudo权限采用集中管理）
2.大型数据存放在/data1,/data2,/data3,/data4,/data5,/data6这些硬盘上面。
3.不要私自建立账户，需要建立账户提前跟我联系。

dongzhuoyao

virturalenv 

sudo apt-get install python-dev 

dongzhuoyao

保存当前环境安装包
pip freeze > requirements.txt

安装大礼包
pip install -r requirements.txt

dongzhuoyao

list_tensors -n confc.*

fn+pagedown

dongzhuoyao

name: "SE-ResNet-50"
# mean_value: 104, 117, 123
layer { 
  name: "data"
  type: "Input"
  top: "data"
  input_param: {
    shape: {
      dim: 1 
      dim: 3