Caffe weight visualization, feature visualization, network model visualization

--------------------------------------------------------------------------------

Visualization of weight values

After training, the network weights can be visualized to judge the model and whether it owes (too) fit. Well-trained network weights usually appear to be aesthetically pleasing, smooth, whereas the opposite is a noisy image, or the pattern correlation is too high (very regular dots and stripes), or lack of structural or more ' dead ' areas.

zz@zz-inspiron-5520:~$ CD Caffe

zz@zz-inspiron-5520:~/caffe$ CD python

zz@zz-inspiron-5520:~/caffe/python$ ls
Caffe CMakeLists.txt detect.py Lenet_iter_10000.caffemodel mnist_deploy.prototxt requirements.txt t_extract_weights.py~
classify.py conv2.jpg draw_net.py lenet_train_test.prototxt mnist.jpg test_extract_weights.py

zz@zz-inspiron-5520:~/caffe/python$ python test_extract_weights.py

Results: [(' Conv1 ', (1, 5, 5)), (' Conv2 ', (M, 5, 5)), (' Ip1 ', (+)), (' Ip2 ', (10, 500))]

20 5*5 convolution kernel, with 1-D as input.

50 5*5 convolution kernel, with 20-D as input.
------------------------------------------------------------------------------

Feature Visualization

zz@zz-inspiron-5520:~$ CD Caffe

zz@zz-inspiron-5520:~/caffe$ CD python

zz@zz-inspiron-5520:~/caffe/python$ python test_extract_data.py

------------------------------------------------------------------------------

Visualization of network models

Method 1: In code: caffe/python/draw_net.py

To enter the Caffe/python to execute the python script

zz@zz-inspiron-5520:~$ CD Caffe

zz@zz-inspiron-5520:~/caffe$ CD python

zz@zz-inspiron-5520:~/caffe/python$ ls

Caffe classify.py CMakeLists.txt detect.py draw_net.py requirements.txt

zz@zz-inspiron-5520:~/caffe/python$./draw_net.py
usage:draw_net.py [-h] [--rankdir rankdir] [--phase phase]
Input_net_proto_file Output_image_file
Draw_net.py:error:too few arguments/no need to ignore the error

zz@zz-inspiron-5520:~/caffe/python$./draw_net.py--rankdir TB./lenet_train_test.prototxt mnist.jpg

Drawing NET to Mnist.jpg

Tb=top to bottom; network model: Lenet_train_test.prototxt; save to Mnist.jpg

20 5*5 convolution kernel, sliding with 1 steps.

Method 2. Online visualization tool Http://ethereon.github.io/netscope/#/editor

Shift key + ENTER, you can paint.

Put the original delete, put the network model: Lenet_train_test.prototxt code to copy the past

-----------------------------------------------------------------------------

Visualization of Accuracy/loss curves

Step1: Copy this file to Caffe/python and change the name to :p lot_training_log.py

Step2:save Training phase Print to screen information

$ sh train_lenet.sh >& 1.log