%% raw history of commands from seminar
clear
clc
net = alexnet;
plot(net)
analyzeNetwork(net)
ls
cd custom_images
ls
cd test
ls
%% reading image into matlab
im = imread('img2.JPG');
image(im)
imshow(im)
net.classify(im)
net
net.Layers
net.Layers(1)
net.Layers(end)
net.Layers(end).Classes
net.Layers(2)
net.Layers(2).Weights
net.Layers(3)
net.Layers(4)
net.Layers(5)
net.Layers(6)
%% how to classify many images
for i = 1:8
im(:,:,:,i) = imread(sprintf('img%d.JPG',i));
end
size(im)
net.classify(im)
%% the better way how to read and deal with images
imgds = imageDatastore("D:\dokumenti\HPC\neural_network_seminar\neural_network_training_sets\custom_images\test\")
% how to read an image from datastore
im1 = read(imgds);
image(im1)
im1 = read(imgds);
image(im1)
im1 = readimage(imgds,3);
image(im1)
% how to see all or part of images that is in datastore
montage(imgds)
montage(imgds,'Size',[2 2])
net.Layers(1)
cd ../test2
%% if image size does not fit to neural network input
imgds2 = imageDatastore('D:\dokumenti\HPC\neural_network_seminar\neural_network_training_sets\custom_images\test2\');
net.classify(imgds2)
% to resize one image:
im1 = read(imgds2);
imres = imresize(im1,[227 227]);
image(imres)
% how to resize a whole datastore
augs = augmentedImageDatastore([227 227],imgds2)
classify([227 227],augs)
net.classify([227 227],augs)
classify(net,augs)
montage(augs)
montage(imgds2)
imgds = imageDatastore("D:\dokumenti\HPC\neural_network_seminar\neural_network_training_sets\custom_images\test3");
augs = augmentedImageDatastore([227 227],imgds,'ColorPreprocessing','gray2rgb');
classify(net,augs)
ls('D:\dokumenti\HPC\neural_network_seminar\neural_network_training_sets\custom_images\test3\')
digitDatasetPath = fullfile(matlabroot,'toolbox','nnet', ...
'nndemos','nndatasets','DigitDataset');
imds = imageDatastore(digitDatasetPath, ...
'IncludeSubfolders',true, ...
'LabelSource','foldernames');
figure
numImages = 10000;
perm = randperm(numImages,20);
for i = 1:20
subplot(4,5,i);
imshow(imds.Files{perm(i)});
drawnow;
end
imds
im = read(imds);
size(im)
options = trainingOptions('sgdm')
my_network
options = trainingOptions('sgdm');
options
my_network
ValidationData: []
ValidationFrequency: 50
%% next example
net = alexnet;
deepNetworkDesigner
layer_1
layers_1
layers_1(end-2)
layers_1(20)
save alexnet_with_weights_modified_for_5_classes.mat layers_1
help alexnet
layers = alexnet('Weights', 'none');
layers(3)
layers(2)
layers = alexnet('Weights', 'none');
deepNetworkDesigner
whos
save alexnet_without_learnables.mat layers_3
ls *.mat
clear
load fine_tuning_example
whos
net = flowernet;
directory = "./neural_network_training_sets/Linnaeus 5 256X256/"
imds = imageDatastore(directory+"test","IncludeSubfolders",...
true,'LabelSource','foldernames');
augsTest = augmentedImageDatastore([227 227],imds,'ColorPreprocessing','gray2rgb');
load alexnet_with_weights_modified_for_5_classes.mat
load fine_tuning_example2.mat
imds = imageDatastore(directory+"test","IncludeSubfolders",...
true,'LabelSource','foldernames');
directory = "./neural_network_training_sets/Linnaeus 5 256X256/"
imds = imageDatastore(directory+"test","IncludeSubfolders",...
true,'LabelSource','foldernames');
realClass = imds.Labels;
load alexnet_with_weights_modified_for_5_classes.mat
load fine_tuning_example2.mat
[predClass,scores] = flowernet.classify(augsTest);
confusionchart(realClass,predClass);
prob = nnz(realClass==predClass)/numel(realClass);
realClass(1:3)
size(realClass)
nnz(realClass==predClass)
nnz(realClass==predClass)/numel(realClass)
%% how to look on wrong classes
imgds = imageDatastore("./neural_network_training_sets/custom_images");
imgds = imageDatastore("./test/*.jpg");
auds = augmentedImageDatastore([227 227],imgds),
[preds,variety]=classify(net,auds)
net.classify(auds)
indexes_of_wrong_images = (realClass!=predClass);
indexes_of_wrong_images = (realClass~=predClass);
find(indexes_of_wrong_images)
imds = imageDatastore(directory+"test","IncludeSubfolders",...
true,'LabelSource','foldernames');
directory = "./neural_network_training_sets/Linnaeus 5 256X256/"
imds = imageDatastore(directory+"test","IncludeSubfolders",...
true,'LabelSource','foldernames');
augsTest = augmentedImageDatastore([227 227],imds,'ColorPreprocessing','gray2rgb');
image(readimage(imds,39))
[predClass,scores] = flowernet.classify(augsTest);
im = (readimage(imds,39));
[predClass,scores] = flowernet.classify(imresize(im,[227 227]));
predClass
bar(scores)
figure,image(im)
75/23
75/12
65*2.54
%-- 07.03.2023 10:45 --%
doc trainingOptions
%-- 07.03.2023 14:14 --%
trainNetwork
clc
my_network
trainingOptions
options
my_network
clc
my_network
edit my_network
help splitEachLabel
options
my_network
doc trainingOptions
ls
nano my_network
edit my_network
%-- 13.03.2023 08:20 --%