[caffe with docker][01] train demo, the mnist dataset

Hi, every one, here will be a series for record my learning process of caffe with docker. The reason why using docker, not with a bare metal like linux or windows, is just cuz the installation is too hard for me, a noob.

1. basic requirements

• linux with docker, if you don’t know how to install docker on your linux, check my previous blog here.
• skilled in docker
• a little knowledges about shell

2. get and run caffe image

official installation weblink is http://caffe.berkeleyvision.org/installation.html, here is the shell command.

docker pull bvlc/caffe:cpu
docker run -it -d bvlc/caffe:cpu /bin/bash

then we can go into the container

docker exec -it ba1 /bin/bash

the default WORKDIR is /workspace, we should cd /opt/caffe, the rectangled zone is important in the next.

3. install vim in docker

we need to edit the files in it, so we need to install vim with apt-get install vim, and then edit the file ~/.vimrc, add one line syntax on in it is enough.

4. download the mnist dataset

./data/mnist/get_mnist.sh

5. create database

./examples/mnist/create_mnist.sh

6. train it

./examples/mnist/train_lenet.sh

saddly, we got en error, cuz we didn’t change the mode.

change the mode in file vi examples/mnist/lenet_solver.prototxt

then we rerun the command ./examples/mnist/train_lenet.sh, we can see, the training is started! and the process will be printed every 100 loops.

over, here we just need to wait for it, and then a caffemodel will be generated. we’ll learn how to use the model in next article.

---

after go out for about half an hour running, the training ended. some models generated.