-
Notifications
You must be signed in to change notification settings - Fork 0
/
HUDD.py
112 lines (111 loc) · 6.46 KB
/
HUDD.py
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
#
# Copyright (c) University of Luxembourg 2019-2020.
# Created by Hazem FAHMY, [email protected], SNT, 2019.
# Modified by Mojtaba Bagherzadeh, [email protected], University of Ottawa, 2019.
#
import Helper, RQ2, RQ1
from imports import basename, argparse, os, shutil, join, np, exists
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='DNN debugger')
parser.add_argument('-a', '--action', help='supported actions: test, heatmap, cluster, assign, retrain',
required=False)
parser.add_argument('-m', '--modelName', help='pretrainedWeights.pth Path', required=False)
parser.add_argument('-o', '--outputPathX', help='Output path for saving the result', required=True)
parser.add_argument('-sF', '--scratchFlag', help='Number of Classes', required=False)
parser.add_argument('-n', '--ClusterModeX', help='ICD - WICD - S', required=False)
parser.add_argument('-cF', '--clustF', help='clustering Flag', required=False)
parser.add_argument('-dcF', '--drawCF', help='Exporting images Flag', required=False)
parser.add_argument('-aF', '--assignF', help='Exporting images Flag', required=False)
parser.add_argument('-daF', '--drawAssignF', help='Exporting images Flag', required=False)
parser.add_argument('-err', '--errorMarginPixels', help='error Margin Pixels', required=False)
parser.add_argument('-sub', '--faceSubSet', help='Subset of the face', required=False)
parser.add_argument('-tl', '--transfer', help='scratch/pretrained', required=False)
parser.add_argument('-rF', '--retrainF', help='HUDD, BL1, BL2', required=False)
parser.add_argument('-mode', '--retrainMode', help='HUDD, BL1, BL2', required=False)
parser.add_argument('-app', '--approach', help='A, B', required=False)
parser.add_argument('-exp1', '--expNumber', help='Number of retrainings', required=False)
parser.add_argument('-exp2', '--expNumber2', help='Number of retrainings', required=False)
parser.add_argument('-ep', '--epoch', help='Number of epochs', required=False)
parser.add_argument('-ass', '--assignMode', help='ICD - Centroid - Closest - SSE', required=False)
parser.add_argument('-bs', '--BagSize', help='ICD - Centroid - Closest - SSE', required=False)
parser.add_argument('-mc', '--maxClust', help='ICD - Centroid - Closest - SSE', required=False)
parser.add_argument('-ow', '--ow', help='overwrite flag', required=False)
parser.add_argument('-sel', '--select', help='layer selection mode', required=False)
parser.add_argument('-fld', '--FLD', help='FLD selection mode', required=False)
parser.add_argument('-wc', '--workersCount', help='FLD selection mode', required=False)
parser.add_argument('-batchS', '--batchSize', help='FLD selection mode', required=False)
parser.add_argument('-cleanF', '--cleanFlag', help='FLD selection mode', required=False)
parser.add_argument('-rcc', '--rccSource', help='FLD selection mode', required=False)
parser.add_argument('-numR', '--numRuns', help='FLD selection mode', required=False)
parser.add_argument('-rA', '--retrieveAccuracy', help='FLD selection mode', required=False)
parser.add_argument('-rq', '--RQ1A', help='FLD selection mode', required=False)
parser.add_argument('-rS', '--retrainSet', help='FLD selection mode', required=False)
parser.add_argument('-HUDD', '--HUDDmode', help='FLD selection mode', required=False)
parser.add_argument('-iee', '--ieeVersion', default="1", help='iee_sim1, iee_sim2', required=False)
parser.add_argument('-cls', '--clsNum', default="1", help='iee_sim1, iee_sim2', required=False)
args = parser.parse_args()
components = ["noseridge", "nose", "mouth", "rightbrow", "righteye", "lefteye", "leftbrow"]
HUDD = Helper.Helper(outputPath=args.outputPathX, modelName=args.modelName, workersCount=args.workersCount,
batchSize=args.batchSize, metric="Euc", clustFlag=args.clustF, assignFlag=args.assignF,
retrainFlag=args.retrainF, retrainMode=args.retrainMode, retrainApproach=args.approach,
expNumber=args.expNumber, expNumber2=args.expNumber2, bagSize=args.BagSize,
clustMode=args.ClusterModeX, assMode=args.assignMode,
overWrite=args.ow, selectionMode=args.select, FLD=args.FLD, cleanFlag=args.cleanFlag,
RCC=args.rccSource, scratchFlag=args.scratchFlag, retrieveAccuracy=args.retrieveAccuracy,
RQ1A=False, retrainSet=args.retrainSet, drawClustFlag=args.drawCF, ieeVersion=args.ieeVersion, clustNum=args.clsNum)
#HUDD.updateCaseFile()
finalResultDict = {}
datasetName = basename(args.outputPathX)
if args.HUDDmode == "HUDD":
if datasetName == "FLD":
if args.faceSubSet is None:
maxSub = 0.0
else:
print(args.faceSubSet)
ResultDict, _ = HUDD.KPNet(args.faceSubSet)
else:
ResultDict, _ = HUDD.AlexNet()
if args.numRuns is None:
if datasetName == "FLD":
ResultDict, _ = HUDD.KPNet(components[0])
HUDD.faceSubSet = components[0]
# HUDD.saveResult()
HUDD.retrainDNN()
else:
for x in range(0, int(args.numRuns)):
HUDD.retrainDNN()
elif args.HUDDmode == "RQ2":
RQ2.run(HUDD.modelName, HUDD.outputPath, HUDD.datasetName, HUDD.modelPath, HUDD.numClass, HUDD.modelArch)
if datasetName == "FLD":
ResultDict, _ = HUDD.KPNet(args.faceSubSet)
else:
ResultDict, _ = HUDD.AlexNet()
elif args.HUDDmode == "xplain":
HUDD.explain()
elif args.HUDDmode == "params":
HUDD.faceSubset = args.faceSubSet
HUDD.getParams()
elif args.HUDDmode == "HUDD":
if datasetName == "FLD":
ResultDict, _ = HUDD.KPNet(None)
else:
ResultDict, _ = HUDD.AlexNet()
elif args.HUDDmode == "train":
HUDD.train()
elif args.HUDDmode == "RQ1":
RQ1.IEERQ1(HUDD.caseFile)
elif args.HUDDmode == "retrain":
HUDD.selectLayer()
HUDD.retrainDNN()
elif args.HUDDmode == "testModel":
HUDD.saveResult()
elif args.HUDDmode == "generateHeatmaps":
HUDD.generateHeatmaps()
elif args.HUDDmode == "generateHMDists":
HUDD.generateHMDistances()
elif args.HUDDmode == "generateClusters":
HUDD.generateClusters()
HUDD.selectLayer()
elif args.HUDDmode == "assignImages":
HUDD.selectLayer()
HUDD.assignImages()