-
Notifications
You must be signed in to change notification settings - Fork 0
/
Food_Identification.m
212 lines (200 loc) · 8.21 KB
/
Food_Identification.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
% %% Fine Tuning A Deep Neural Network
% % This example shows how to fine tune a pre-trained deep convolutional
% % neural network (CNN) for a new recognition task.
% %% Load Image Data
% % Data is 5 different categories of food
% % Create an imageDataStore to read images
% location = 'C:\Users\michelle.de.bock\OneDrive\Documenten\Master Thesis\Afbeeldingen\Demo2_TransferLearning\Categories';
% imds = imageDatastore(location,'IncludeSubfolders',1,...
% 'LabelSource','foldernames');
% tbl = countEachLabel(imds);
%
%
% %% Load Pre-trained CNN
% % The CNN model is saved in MatConvNet's format [3]. Load the MatConvNet
% % network data into |convnet|, a |SeriesNetwork| object from Neural Network
% % Toolbox(TM), using the helper function |helperImportMatConvNet|. A
% % SeriesNetwork object can be used to inspect the network architecture,
% % classify new data, and extract network activations from specific layers.
%
% % Location of pre-trained "AlexNet"
% cnnURL = 'http://www.vlfeat.org/matconvnet/models/beta16/imagenet-caffe-alex.mat';
% % Store CNN model in a temporary folder
% cnnMatFile = fullfile(tempdir, 'imagenet-caffe-alex.mat');
%
% if ~exist(cnnMatFile, 'file') % download only once
% disp('Downloading pre-trained CNN model...');
% websave(cnnMatFile, cnnURL);
% end
%
% net = helperImportMatConvNet(cnnMatFile);
%
% %% Look at structure of pre-trained network
% % Notice the last layer performs 1000 object classification
% net.Layers
%
% %% Perform net surgery
% % The pre-trained layers at the end of the network are designed to classify
% % 1000 objects. But we need to classify different objects now. So the
% % first step in transfer learning is to replace the last 3 layers of the
% % pre-trained network with a set of layers that can classify 5 classes.
%
% % Get the layers from the network. the layers define the network
% % architecture and contain the learned weights. Here we only need to keep
% % everything except the last 3 layers.
% layers = net.Layers(1:end-3);
%
% % %optional, Add new fully connected layer for 2 categories.
% layers(end+1) = fullyConnectedLayer(64, 'Name', 'special_2');
% layers(end+1) = reluLayer;
%
% % introduce another layer : adding non-linearity
%
% % improving the network's ability to handle data
% % Not enough training 1.2 million (total)
%
% layers(end+1) = fullyConnectedLayer(height(tbl), 'Name', 'fc8_2')
%
% % Add the softmax layer and the classification layer which make up the
% % remaining portion of the networks classification layers.
% layers(end+1) = softmaxLayer
% layers(end+1) = classificationLayer()
%
% % Modify image layer to add randcrop data augmentation. This increases the
% % diversity of training images. The size of the input images is set to the
% % original networks input size.
% layers(1) = imageInputLayer([227 227 3]);
%
%
% %% Setup learning rates for fine-tuning
% % For fine-tuning, we want to changed the network ever so slightly. How
% % much a network is changed during training is controlled by the learning
% % rates. Here we do not modify the learning rates of the original layers,
% % i.e. the ones before the last 3. The rates for these layers are already
% % pretty small so they don't need to be lowered further. You could even
% % freeze the weights of these early layers by setting the rates to zero.
% %
% % Instead we boost the learning rates of the new layers we added, so that
% % they change faster than the rest of the network. This way earlier layers
% % don't change that much and we quickly learn the weights of the newer
% % layer.
%
% % fc 8 - bump up learning rate for last layers
% layers(end-2).WeightLearnRateFactor = 10;
% layers(end-2).WeightL2Factor = 1;
% layers(end-2).BiasLearnRateFactor = 20;
% layers(end-2).BiasL2Factor = 0;
%
% % we want the last layer to train faster than the other layers because
% % bias the training proces to quickly improve the last layer and keep the
% % other layers relatively unchanged
%
% % could have set the learning rates to 0, but this is faster
% % other layers are not getting updated as much
%
% %% Equalize number of images of each class in training set
% minSetCount = min(tbl{:,2}); % determine the smallest amount of images in a category
% %Use splitEachLabel method to trim the set.
% imds = splitEachLabel(imds, minSetCount);
%
% % Notice that each set now has exactly the same number of images.
% countEachLabel(imds)
% [trainingDS, testDS] = splitEachLabel(imds, 0.7,'randomize');
% % Convert labels to categoricals
% trainingDS.Labels = categorical(trainingDS.Labels);
% trainingDS.ReadFcn = @readFunctionTrain;
%
% % Setup test data for validation
% testDS.Labels = categorical(testDS.Labels);
% testDS.ReadFcn = @readFunctionValidation;
%
%
% %% Fine-tune the Network
%
% miniBatchSize = 128; % lower this if your GPU runs out of memory.
% numImages = numel(trainingDS.Files);
%
% % Run training for 5000 iterations. Convert 20000 iterations into the
% % number of epochs this will be.
% maxEpochs = 20; % one complete pass through the training data
% % batch size is the number of images it processes at once. Training
% % algorithm chunks into manageable sizes.
% lr = 0.0001
% opts = trainingOptions('sgdm', ...
% 'LearnRateSchedule', 'none',...
% 'InitialLearnRate', lr,...
% 'MaxEpochs', maxEpochs, ...
% 'MiniBatchSize', miniBatchSize);
%
% net = trainNetwork(trainingDS, layers, opts);
% save('trainedNet.mat','net')
% save('testDS.mat','testDS')
% % This could take over an hour to run, so lets stop and load a pre-traiend
% % version that used the same data
% return % the script will stop here if you run the entire file
%% Load in a previously saved network and test set
load('trainedNet.mat');
load('testDS.mat');
%% Test 5-class classifier on validation set
% Now run the network on the test data set to see how well it does:
[labels,err_test] = classify(net, testDS, 'MiniBatchSize', 64);
confMat = confusionmat(testDS.Labels, labels);
confMat = bsxfun(@rdivide,confMat,sum(confMat,2));
mean(diag(confMat))
%% Can we tell anything about the misses?
idx = find(testDS.Labels == 'Place');
misses_only = find(labels(idx) ~= testDS.Labels(idx));
misses_only = idx(misses_only);
%% Check for misses
for ii = 1: length(misses_only)
idx = misses_only(ii);
extra = ' ';
imshow(imread(testDS.Files{idx}));
if(err_test(idx) < .3)
extra = '?';
end
title(sprintf('%s %s',char(labels(idx)),extra)) ;
pause;
end
%% Choose a random image and visualize the results
randNum = randi(length(testDS.Files));
im = readAndPreprocessImage(testDS.Files{randNum}) ;
label = char(classify(net,im)); % classify with deep learning
imshow(im);
title(label);
%% We can also see how confident we are
% less than a certain score also shows the second most likely option
randNum = randi(length(testDS.Files));
im = readAndPreprocessImage(testDS.Files{randNum}) ;
[label,score] = classify(net,im); % classify with deep learning
imshow(im);
interesting_title = char(label);
%display(char(label));
[max_score,idx] = sort(score,'descend');
if(max_score < .66)
% display(max_score);
% display(idx);
interesting_title = sprintf('%s? or \n');
%strcat(interesting_title,'? or '), char(10);
second = find(score == max_score(2));
interesting_title = strcat(interesting_title,' ',char(tbl.Label(second)));
end
title(interesting_title);
%% Detect Laban Symbol category using CNN classification
videoReader = vision.VideoFileReader('video6.MOV');
videoPlayer = vision.DeployableVideoPlayer;
position = [35 160];
position2 = [35 600];
%% Video
while ~isDone(videoReader)
frame = step(videoReader);
im = im2uint8(imresize(frame, [227 227],'bilinear')) ;
[label,score] = classify(net,im); % classify with deep learning
%display(char(label));
%display(max(score));
RGB = insertText(frame,position,char(label),'FontSize',18,'TextColor','white','BoxColor','black');
RGB = insertText(RGB,position2,sprintf('%.2f',round(max(score),2)),'FontSize',18,'TextColor','white','BoxColor','green');
step(videoPlayer,RGB);
end
release(videoReader);
release(videoPlayer);