NVIDIADIGITS
I've forgotten where I've read a few similar articles, and in an effort to make more of my friends aware of this great tool, I'm also acting as a qualified mover today, so I hope you'll try to spread positive energy and make progress with us all! Thank you!
NVIDIA's preference for deep learning in order to sell more graphics cards is really lighting it up. Along with CUDNN, comes out DIGITS, really wish elementary school kids could learn deep learning and then go buy their cards. NVIDIA DIGITS is a web application tool for graphical manipulation and visualization of Caffe on the web, which is very helpful for Caffe beginners. But unfortunately, according to NVIDIA's official documentation, the best supported system for DIGITS is Ubuntu 14.04. It is not known how well other systems work.
I. Installation of DIGITS 4.0
DIGITS is running on cuda and caffe, so there is no doubt that you have to configure cuda+caffe first.
Open a terminal and run the following commands in sequence.
cd sudo -s
Go to the current user root and switch to superuser (the symbol changes from $ to #, no need to type sudo in every sentence)
CUDA_REPO_PKG=cuda-repo-ubuntu1404_7.5-18_amd64.deb &&wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/$CUDA_REPO_PKG &&sudo dpkg -i $CUDA_REPO_PKG
And then..,
ML_REPO_PKG=nvidia-machine-learning-repo_4.0-2_amd64.deb &&wget http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1404/x86_64/$ML_REPO_PKG &&sudo dpkg -i $ML_REPO_PKG
apt-get update
apt-get install digits
That's fine, then it's just a testament to how fast your internet is, hahaha, take your time and wait 。。。。
II. Running digits
By default, digits is installed in the following directory.
/usr/share/digits
Once the installation is complete, open your browser and type http://localhost/That'll do it. into the address bar, it's that simple.
Even more powerful: on other machines in the local area, you can also access it with a browser, only the localhost becomes the host ip address. Many people like to remote connect to linux on a windows system to execute caffe. Well, now there's no need for a remote connection, just access to a website!
III. Running the mnist (handwritten digital dataset) example
Now to run an example.mnist
The raw data needs to be an image, but the mnist data available online is not in image format, so we need to convert it to an image in order to run it.
digits provides a script file for downloading mnist,cifar10 and cifar100 data and converting it to png format images. The file path is.
/usr/share/digits/tools/download_data/main.py
Let's start by creating a new mnist folder in the root directory of the current user to save mnist images.
cd your file directory mkdir mnist
Then execute the script
/usr/share/digits/tools/download_data/main.py mnist ~/mnist
main.py with two parameters, the first is the dataset name (can be set to mnist,cifar10 or cifar100) and the second is the output path (~/mnist)
After successful execution, two folders (train folder and test folder) are generated under the mnist folder, and under each folder are the images we need (10 categories in 10 subfolders), as well as the corresponding in image list files train.txt and test.txt.
next， Running on a browserdigits, Click on the leftDataset modular"Image" Button Selection“classification", Create adataset。
On the left side of this page， You can set whether the image is a color image or a grayscale image， If the original image provided is not the same size， also availableResize Transformation Functionality converted to consistent size。 As you can see from the middle of the page， The system defaults the training image to the25% Fetch it as a validation set。 If you want to put the image used for testing， also generateslmdb, standardize“separate test image folder" Check this option.。 Once all set up， clicking"create" buttons， Start generatinglmdb data。
Note the Job Directory in the upper left corner, the generated lmdb files are placed under this directory, you'd better open this directory to see what files are generated and to understand the principle of operation. In this interface, we can also visualize the images of the training and testing, as follows.
train.txt It holds a list of all the training images， The bar chart clearly shows the10 Number of respective samples in class。 clicking"Explorer the db” Instantly view pictures。 lastly， Click on the top left corner“DIGITS" Link back to the root of the site。
The document is available at: Link: http://pan.baidu.com/s/1qYfJ9ve Password: rtqg
Recommended>>
1、AutoMagic Design Ideas Introduction and New Custom Keywords Example
2、Linux system and kernel parameter optimization under high concurrency
3、How good is Japanese AI Youll know after seeing these 3 use cases
4、Worried about personal information being compromised when you see a doctor online Come hear experts explain the key points of Internet healthcare law
5、Friends gathering no fear of paying the bill with this