-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathdenoise_1.m
178 lines (132 loc) · 6.56 KB
/
denoise_1.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
clear, clc, close all
addpath('code')
addpath(['edncnn' filesep 'code'])
addpath(['edncnn' filesep 'camera'])
edncnnDir = '/media/wescomp/DDD17/6_features/'
epmDir = '/media/wescomp/WesDataDrive/edncnn_output_linear/'; %Data processed with EDnCNN is here
toreDir = '/home/wescomp/data/denoise/tore_randomSample/'; %output directory
if ~exist(toreDir,'dir')
mkdir(toreDir)
end
%For each dataset from EDnCNN - read in representations/labels and make Tore to match
files = dir([epmDir '*epm.mat'])
numSamplesPerFile = 10e3;
k = 7;
for fLoop = 1:numel(files)
if exist([toreDir files(fLoop).name(1:end-4) '_tore.mat'],'file')
disp('file already processed')
continue
end
load([epmDir files(fLoop).name])
%convert to doubles
aedat.data.polarity.timeStamp = double(aedat.data.polarity.timeStamp);
%Ensure events are sorted by time
if ~issorted(aedat.data.polarity.timeStamp)
[aedat.data.polarity.timeStamp,idx] = sort(aedat.data.polarity.timeStamp);
aedat.data.polarity.y = aedat.data.polarity.y(idx);
aedat.data.polarity.x = aedat.data.polarity.x(idx);
aedat.data.polarity.polarity = aedat.data.polarity.polarity(idx);
aedat.data.polarity.closestFrame = aedat.data.polarity.closestFrame(idx);
aedat.data.polarity.frameTimeDelta = aedat.data.polarity.frameTimeDelta(idx);
aedat.data.polarity.duringAPS = aedat.data.polarity.duringAPS(idx);
aedat.data.polarity.apsIntensity = aedat.data.polarity.apsIntensity(idx);
aedat.data.polarity.apsIntGood = aedat.data.polarity.apsIntGood(idx);
aedat.data.polarity.Jt = aedat.data.polarity.Jt(idx);
aedat.data.polarity.Prob = aedat.data.polarity.Prob(idx);
end
numRows = double(aedat.data.frame.size(1));
numCols = double(aedat.data.frame.size(2));
numEvents = aedat.data.polarity.numEvents;
inputVar.neighborhood = 4; %9x9 chip centered on event of interest
%Random sample (DO not use the EDnCNN sample strategy of balances EPM scores)
%don't even balance pos/neg events since that isn't normalized in the EPM score
%Do not sample the beginning and ending of the files
timeQuantiles = quantile(aedat.data.polarity.timeStamp,[0.15 0.85]);
qFilter = aedat.data.polarity.timeStamp >= timeQuantiles(1) & ...
aedat.data.polarity.timeStamp <= timeQuantiles(2);
%Do not to sample near the edge
nearEdgeIdx = ((aedat.data.polarity.y-inputVar.neighborhood) < 1) | ...
((aedat.data.polarity.x-inputVar.neighborhood) < 1) | ...
((aedat.data.polarity.y+inputVar.neighborhood) > numRows) | ...
((aedat.data.polarity.x+inputVar.neighborhood) > numCols);
%Sample where EPM is most accurate (i.e. during APS frame and avoid saturated
%and underexposed areas of APS)
sampleIdx = ~nearEdgeIdx & qFilter & (aedat.data.polarity.duringAPS>0) & aedat.data.polarity.apsIntGood;
samples = find(sampleIdx);
sampleList = randsample(numel(samples), numSamplesPerFile);
sampleList = samples(sampleList);
samples = false(numEvents,1);
samples(sampleList) = true;
Xtore = events2ToreChip(...
aedat.data.polarity.x, aedat.data.polarity.y, aedat.data.polarity.timeStamp, aedat.data.polarity.polarity, ...
aedat.data.polarity.y(samples), aedat.data.polarity.x(samples), aedat.data.polarity.timeStamp(samples), aedat.data.polarity.polarity(samples), ...
k, inputVar.neighborhood);
Ytore = categorical(aedat.data.polarity.Prob(samples)>0.5,[true false],{'valid' 'noise'});
samples_tore = samples;
save([toreDir files(fLoop).name(1:end-4) '_tore.mat'], 'Xtore', 'Ytore', 'samples_tore')
end
%Combine data from each dataset into one train/test dataset
buildTrainTestData_tore(toreDir)
% load X/Y from the all_label.mat file that contains 10k samples per dataset
Y = load([toreDir 'all_labels.mat'],'Y');
Y = Y.Y;
grpLabel = load([toreDir 'all_labels.mat'],'grpLabel');
grpLabel = grpLabel.grpLabel;
testIdx = ismember(grpLabel,[2 3 10]); %test on bench, bigChecker, labFast
testY = Y(testIdx);
trainY = Y(~testIdx);
clear Y
X = load([toreDir 'all_labels.mat'],'Xtore');
X = X.Xtore;
testX = X(:,:,:,testIdx);
X(:,:,:,testIdx) = [];
trainX = X;
clear X
results = trainEDnCNN_tore(toreDir, trainX, trainY, testX, testY); %9x9x16 (MP layers off)(Dropout ON) (77.13 acc) ***
%% Use network to predict data labels (real/noise)
testSet = [4:9 28:30]; %test on bench, bigChecker, labFast
for fLoop = 1:numel(testSet)
file = [epmDir files(testSet(fLoop)).name]
[fp,fn,fe] = fileparts(file);
if exist([toreDir fn '_pred_MPF.mat'],'file')
disp('file already processed')
continue
end
load(file, 'aedat', 'inputVar')
%Ensure events are sorted by time
aedat.data.polarity.timeStamp = double(aedat.data.polarity.timeStamp);
if ~issorted(aedat.data.polarity.timeStamp)
[aedat.data.polarity.timeStamp,idx] = sort(aedat.data.polarity.timeStamp);
aedat.data.polarity.y = aedat.data.polarity.y(idx);
aedat.data.polarity.x = aedat.data.polarity.x(idx);
aedat.data.polarity.polarity = aedat.data.polarity.polarity(idx);
aedat.data.polarity.closestFrame = aedat.data.polarity.closestFrame(idx);
aedat.data.polarity.frameTimeDelta = aedat.data.polarity.frameTimeDelta(idx);
aedat.data.polarity.duringAPS = aedat.data.polarity.duringAPS(idx);
aedat.data.polarity.apsIntensity = aedat.data.polarity.apsIntensity(idx);
aedat.data.polarity.apsIntGood = aedat.data.polarity.apsIntGood(idx);
aedat.data.polarity.Jt = aedat.data.polarity.Jt(idx);
aedat.data.polarity.Prob = aedat.data.polarity.Prob(idx);
end
%Smaller spatial / deeper temporal
inputVar.neighborhood = 4; %9x9 chip centered on event of interest
inputVar.depth = k;
load('pretrainedNetworks/denoise/denoise_acc_0p77134_size_9__9_14_tore.mat', 'net')
YPred = makeLabeledAnimations_4Tore(aedat, inputVar, net);
save([toreDir fn '_pred_Tore_9x9x14_0p77134.mat'],'YPred','-v7.3')
YPred_multiColumn = YPred;
YPred = YPred(:,1);
YPred_s = nan(numel(YPred),1);
YPred_s(YPred>0.5) = 1;
YPred_s(YPred<=0.5) = 0;
[noisyScore(fLoop), denoiseScoreTore(fLoop)] = scoreDenoise(aedat, YPred_s);
YPred = categorical(YPred_s,[1 0],{'valid' 'noise'});
% writeOutGifExamples
end
%Average results for each scene and plot
figure
bar(cat(1,mean(reshape(noisyScore,3,[]),1),mean(reshape(denoiseScoreTore,3,[]),1))')
legend('Noisy','DenoisedTore')
xlabel('Scene')
ylabel('RPMD')
grid on