process.py

import matplotlib.pyplot as plt
from matplotlib.pyplot import cm
import math
import numpy as np

numberImgs = {}
# numberImgs["ship"]=[312,199,1223,1508,2725,5073,6181,7913,9018,9604]
# numberImgs["airplane"]=[1898, 2461, 3426, 5173, 180, 8563, 4408, 1573, 8105, 7684]

# numberImgs["dog"]=[239,1773,3000,7118,4917,9557,148,1548,640,1337]
# numberImgs["cat"]=[2804,760,1123,950,3479,4404,3799,4032,4718,3700]

# numberImgs["truck"]=[3344,970,349,2700,3841,7955]

numberImgs["automobile"] = [6, 9, 37, 66, 81, 82, 104, 105,
                            114, 122, 131, 134, 161, 193, 201, 204, 231, 240, 241, 246,
                            261, 283, 286, 290, 305, 308, 325, 330, 351, 363, 366, 369,
                            390, 407, 414, 439, 440, 462, 490, 493, 494, 513, 540, 546,
                            572, 594, 604, 619, 622, 629, 645, 656, 657, 659, 668, 677,
                            723, 726, 736, 738, 753, 759, 764, 771, 781, 796, 801, 830,
                            836, 844, 865, 869, 871, 887, 894, 895, 906, 915, 941, 947,
                            953, 961, 962, 968, 973, 978, 987, 990, 997, 1005, 1006,
                            1013, 1016, 1020, 1021, 1047, 1048, 1054, 1061, 1098, 1115,
                            1131, 1134, 1141, 1144, 1156, 1158, 1174, 1176, 1182, 1185,
                            1187, 1190, 1191, 1206, 1212, 1229, 1232, 1234, 1238, 1258,
                            1260, 1270, 1288, 1301, 1306, 1313, 1332, 1335, 1340, 1354,
                            1363, 1371, 1372, 1378, 1396, 1404, 1405, 1408, 1410, 1412,
                            1413, 1414, 1435, 1437, 1444, 1457, 1458, 1464, 1467]

# numberImgs["deer"]=[7730,9205,4047,4112,4402,835,32,223,2805,1842]
# numberImgs["horse"]=[216,934,1615,2145,2062,4092,4871,9102,9334,9850]

attackTarget = {}

# attackTarget["ship"]="airplane"
# attackTarget["airplane"]="ship"

# attackTarget["dog"]="cat"
# attackTarget["cat"]="dog"

# attackTarget["truck"]="automobile"
attackTarget["automobile"] = "truck"

# attackTarget["deer"]="horse"
# attackTarget["horse"]="deer"

distance = "L2"
bits = [2, 8, 16, 32, 64]
imagesSeries = {}
pathFiles = "/home/roki/GIT/QNNDeepGame/resultsCar2Truck16Agosto/"


def slope(x1, y1, x2, y2):
    m = (y2 - y1) / (x2 - x1)
    return m


plt.figure()
j = 0
label = "automobile"
results = {}
imagesSeries = {}
targetLabel = attackTarget[label]
delta = 0.11
delta2 = 0.005


def checkSlope(slope1, slope2):
    if slope1 * slope2 > 0:
        if abs(slope1 - slope2) < delta:
            return True
    else:
        if abs(slope2) + abs(slope1) < delta2:
            return True
    return False


for image in numberImgs[label]:
    imagesSeries[image] = []
    for i in bits:
        # print i
        # print pathFiles+"cifar10"+str(image)+"Wbits"+str(i)+"Abits"+str(i)+".txt"
        with open(pathFiles + "cifar10" + str(image) + "Wbits" + str(i) + "Abits" + str(i) + ".txt") as f:
            content = f.readlines()
        line0 = content[0]
        words0 = line0.replace("'", "").split()
        predictionCorrect = words0.count(label)
        line_1 = content[-1]
        words_1 = line_1.replace("'", "").split()
        # print "prediction count:"+str(predictionCorrect)
        if predictionCorrect == 2:
            if targetLabel in words_1:
                # it means prediction is ok
                # and target attack is satisfied
                for line in content:
                    if distance in line:
                        val = float(line.split()[-1])
                        imagesSeries[image].append(val)
            else:
                imagesSeries[image].append(float('inf'))
        else:
            imagesSeries[image].append(float('nan'))
listIS = list(imagesSeries.keys())
listISS = list(imagesSeries.keys())
for key in listIS:
    if key in imagesSeries.keys():
        valid = True
        for val in imagesSeries[key]:
            if math.isnan(val) or math.isinf(val):
                valid = False
        if valid:
            results[key] = []
            for j in range(0, len(imagesSeries[key])):
                slope2 = slope(0, imagesSeries[key][0], 10, imagesSeries[key][1])
                slope8 = slope(10, imagesSeries[key][1], 20, imagesSeries[key][2])
                slope16 = slope(20, imagesSeries[key][2], 30, imagesSeries[key][3])
                slope32 = slope(30, imagesSeries[key][3], 40, imagesSeries[key][4])
                for similarImage in listISS:
                    if similarImage in imagesSeries.keys():
                        if similarImage != key:
                            valid = True
                            for val in imagesSeries[similarImage]:
                                if math.isnan(val) or math.isinf(val):
                                    valid = False
                            if valid:
                                Simslope2 = slope(0, imagesSeries[similarImage][0], 10, imagesSeries[similarImage][1])
                                Simslope8 = slope(10, imagesSeries[similarImage][1], 20, imagesSeries[similarImage][2])
                                Simslope16 = slope(20, imagesSeries[similarImage][2], 30, imagesSeries[similarImage][3])
                                Simslope32 = slope(30, imagesSeries[similarImage][3], 40, imagesSeries[similarImage][4])
                                if checkSlope(slope2, Simslope2) and checkSlope(slope8, Simslope8) and checkSlope(
                                        slope16, Simslope16) and checkSlope(slope32, Simslope32):
                                    results[key].append(similarImage)
                                    del imagesSeries[similarImage]
print(results)