-
Notifications
You must be signed in to change notification settings - Fork 16
/
Copy pathmain.m
197 lines (180 loc) · 8.85 KB
/
main.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
function main(input)
clc;
close all;
warning ('off','all');
global netCaffe;
if(nargin == 0)
input = InputParameters;
end
% Load CNN model if specified
% > Comment "if" statement if features/CNN/caffe_.mexw64 not compatible
% if(~isempty(strfind(input.typeDescriptor,'CNN')))
% desc = input.typeDescriptor(5:end);
% model = [input.PATH_CNN input.cnnName '\' input.cnnName '-deploy-' desc '.prototxt'];
% weights = [input.PATH_CNN input.cnnName '\' input.cnnModel '.caffemodel'];
% caffe.set_mode_gpu();
% caffe.set_device(0);
% netCaffe = caffe.Net(model, weights, 'test');
% fprintf('==========\n');
% fprintf('CNN Net: %s\nWeights: %s\nOutput layer: %s\n', input.cnnName, input.cnnModel, desc);
% fprintf('==========\n');
% end
if(strcmpi(input.typePipeline,'class'))
fprintf('\nMulti-Object/Viewpoint Classification:\n');
elseif(strcmpi(input.typePipeline,'det'))
fprintf('\nObject Detection:\n');
elseif(strcmpi(input.typePipeline,'classDet'))
fprintf('\n2-Step Object Detection and Meta-Data estimation:\n');
end
fprintf('Class(es): ');
for i = 1:length(input.sourceDataset.classes)
fprintf('%s ', input.sourceDataset.classes{i});
end
fprintf('\n');
% Prepare discrete list of azimuth viewpoint angles (if not regression)
if(isprop(input.sourceDataset, 'azimuth') && ~isempty(input.sourceDataset.azimuth))
input.sourceDataset.azimuth = getDiscreteAzimuths(input.sourceDataset.azimuth, input.is4ViewSupervised);
else
input.is4ViewSupervised = false; % No viewpoints to coarse them
end
if(isprop(input.sourceDataset,'azimuth') && isprop(input.targetDataset, 'azimuth')); % && ~isempty(input.targetDataset.azimuth))
input.targetDataset.azimuth = input.sourceDataset.azimuth;
if(input.is4ViewSupervised) % Cannot be 4view:true sup:false!
input.isClassSupervised = true;
end
else
input.is4ViewSupervised = false; % No viewpoints to coarse them
end
% -> Get source dataset (train)
fprintf('Source dataset: %s', class(input.sourceDataset));
if(isprop(input.sourceDataset,'source'))
fprintf(' [%s] ', input.sourceDataset.source);
end
fprintf('\n');
[input, srcData, srcFeatures] = getData(input, 'source');
% -> Set Up Bounding boxes and features for each class & view
srcData = setupData(input.sourceDataset, srcData);
% -> Get target dataset (train + test)
fprintf('Target dataset: %s', class(input.targetDataset));
if(isprop(input.targetDataset,'target'))
fprintf(' [%s] ', input.targetDataset.target);
end
fprintf('\n');
[input, tgtData, tgtFeatures, testData, testFeatures] = getData(input, 'target');
fprintf('\n');
if(input.isOpenset && (strcmpi(class(input.sourceDataset),'CrossDataset') || strcmpi(class(input.sourceDataset),'Video') || ...
strcmpi(class(input.sourceDataset),'Visda17')))
[input, srcData, srcFeatures, tgtData, tgtFeatures, testData, testFeatures] = ...
intersectionCrossDataset(input, srcData, srcFeatures, tgtData, tgtFeatures);
input.numSrcClusters = length(input.sourceDataset.classes);
if(strcmpi(class(input.sourceDataset),'Visda17'))
% Sort samples in order of getFeature loading
% - Source data
auxClasses = [];
auxPaths = [];
for i = 1:length(input.sourceDataset.classes)
isClass = ismember(srcData.annotations.classes,input.sourceDataset.classes(i));
auxClasses = [auxClasses; srcData.annotations.classes(isClass)];
auxPaths = [auxPaths; srcData.imgPaths(isClass)];
end
srcData.annotations.classes = auxClasses;
srcData.imgPaths = auxPaths;
end
end
% Check error in compatible viewpoints
if(strcmpi(input.typePipeline, 'det') && strcmpi(input.trainDomain,'both') && ...
(~isprop(input.sourceDataset,'azimuth') && isprop(input.targetDataset,'azimuth')) && ...
(isprop(input.sourceDataset,'azimuth') && ~isprop(input.targetDataset,'azimuth')))
error('[[Caught ERROR: One domain lacks viewpoints: %s]]', phase, class(dataset));
elseif(strcmpi(input.typePipeline, 'det') && strcmpi(input.trainDomain,'both') && ...
isprop(input.sourceDataset,'azimuth') && isprop(input.targetDataset,'azimuth') && ...
length(input.sourceDataset.azimuth) ~= length(input.targetDataset.azimuth))
error('[[Caught ERROR: Domains have different viewpoint granularity: %s]]', phase, class(dataset));
end
% -> Prepare output folders
mDir = [getResultsPath(input) '/'];
removeDir(mDir);
createDir(mDir);
% -> Gathering data features
input = setupFeatures(input, srcData);
% Load source
fprintf('Loading source features\n');
if(isempty(srcFeatures) && isfield(srcData,'annotations'))
srcFeatures = getFeatures(input, input.sourceDataset, srcData, input.numSrcTrain);
srcData.annotations.imgId = srcData.annotations.imgId(1:size(srcFeatures));
if(isfield(srcData.annotations,'BB'))
srcData.annotations.BB = srcData.annotations.BB(1:size(srcFeatures),:);
end
srcData.annotations.classes = srcData.annotations.classes(1:size(srcFeatures));
if(isfield(srcData.annotations,'vp'))
srcData.annotations.vp.azimuth = srcData.annotations.vp.azimuth(1:size(srcFeatures));
end
end
% Load target features (train)
fprintf('Loading target features (train)\n');
if(isempty(tgtFeatures) && isfield(tgtData,'annotations'))
tgtFeatures = getFeatures(input, input.targetDataset, tgtData, input.numTgtTrain);
end
fprintf('Loading target features (test)\n');
if(isempty(testFeatures) && isfield(testData,'annotations'))
testFeatures = getFeatures(input, input.targetDataset, testData, +Inf, '_test');
end
% -> Classification (Label Transfer)
vpClassifiers = [];
[~, ~, metadata] = getIdLabels(input.sourceDataset, srcData.annotations);
if(strfind(lower(input.typePipeline),'class'))
[transferLabels, srcFeatures, vpClassifiers] = step_Classification(input, srcData, srcFeatures, tgtData, tgtFeatures, testData, testFeatures);
if(strfind(lower(input.typePipeline),'det'))
for i = 1:length(metadata)
values = transferLabels(:,i);
if(strcmpi(metadata{i},'azimuth'))
values = cell2mat(cellfun(@str2num, values, 'UniformOutput', false));
input.targetDataset.azimuth = input.sourceDataset.azimuth;
end
tgtData.annotations.(metadata{i}) = values;
end
end
end
% -> Step 2: Object Detection + Viewpoint/Meta-data Estimation
if(strfind(lower(input.typePipeline),'det'))
% Load training data
trainData = [];
fprintf('Loading training features\n');
if(strcmpi(input.trainDomain,'src'))
trainData = srcData;
trainFeatures = srcFeatures;
trainInfo = input.sourceDataset;
elseif(strcmpi(input.trainDomain,'tgt'))
if(strcmpi(input.sourceDataset, input.targetDataset))
trainData = srcData;
trainFeatures = srcFeatures;
trainInfo = input.sourceDataset;
else
trainData = tgtData;
trainFeatures = tgtFeatures;
trainInfo = input.targetDataset;
end
elseif(strcmpi(input.trainDomain,'tgt_gt'))
trainData = tgtData;
trainFeatures = tgtFeatures;
trainInfo = input.targetDataset;
elseif(strcmpi(input.trainDomain,'both'))
trainData.imgPaths = [srcData.imgPaths; tgtData.imgPaths];
trainData.annotations.imgId = [srcData.annotations.imgId; tgtData.annotations.imgId + length(srcData.imgPaths)];
trainData.annotations.BB = [srcData.annotations.BB; tgtData.annotations.BB];
[~, ~, tgtMetadata] = getIdLabels(input.targetDataset, tgtData.annotations, metadata);
for i = 1:length(tgtMetadata)
trainData.annotations.(tgtMetadata{i}) = [srcData.annotations.(tgtMetadata{i}); tgtData.annotations.(tgtMetadata{i})];
end
trainInfo = input.sourceDataset;
trainInfo.classes = unique([trainInfo.classes, input.targetDataset.classes]);
trainFeatures = [srcFeatures; tgtFeatures];
end
trainData = setupData(trainInfo, trainData);
step2_ObjectDetection(input, trainInfo, trainData, trainFeatures, testData, vpClassifiers);
end
if(~isempty(strfind(input.typeDescriptor,'CNN')))
clearvars -global netCaffe;
caffe.reset_all();
end
end