Image-based-recommendations-on-styles-and-substitutes/extractImgs.py at master · Ashish-Gupta03/Image-based-recommendations-on-styles-and-substitutes · GitHub

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
import json
import cv2
from sklearn.metrics.pairwise import euclidean_distances
from sklearn.neighbors import NearestNeighbors
import numpy as np
import os
from sklearn.cluster import KMeans
# from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.decomposition import PCA
from scipy.spatial import distance
from sklearn.metrics import accuracy_score
import matplotlib.pyplot as plt
from sklearn.externals import joblib
from scipy.spatial.distance import mahalanobis

# plt.figure(figsize = (11,10))
# gs1 = gridspec.GridSpec(11, 10)
# gs1.update(wspace=0, hspace=1)
sift = cv2.xfeatures2d.SIFT_create()
# 	g = gzip.open(path,'r')
# 	for l in g:
# 		yield eval(l)
# sift = cv2.xfeatures2d.SIFT_create()
# 'tshirts','jeans','sweat_shirts','shoes','googles','ties','watches','shirts','all'
trainPath = ['tshirts','jeans','sweat_shirts','shoes','googles','ties','watches','shirts']

path = os.path.join(os.getcwd(),'static')
path = os.path.join(path,'images')
path = os.path.join(path,'test')

for k in range(len(trainPath)):
	train_path = os.path.join(path,trainPath[k])

	print ('trainpath ',train_path)
	def load_images():
		count = 0
		imagesList = []
		for i in os.listdir(train_path):
			im1 = cv2.imread(os.path.join(train_path,i))
			imagesList.append(im1)
		return imagesList

	trainData = load_images()
	# print ('tri ',trainData)
	num_images = len(trainData)
	# ==================================   Bag of Visual Words  =================================================================
	desc_list = []
	for img in trainData:
		kp,des = sift.detectAndCompute(img,None)
		desc_list.append(des)

	desc_vStack = np.concatenate(desc_list,axis=0)

	n_components = 69
	pca = PCA(n_components=n_components)
	desc_vStack = pca.fit_transform(desc_vStack)
	# ================================	perform kmeans  =======================================
	n_clusters = 20
	kmeans = KMeans(n_clusters=n_clusters,random_state=0)
	desc_predict = kmeans.fit_predict(desc_vStack)
	# ==================================== BOVW end =============================

	# Generate histogram corresponding to frequency of each cluster(visual word)
	# like there are 50 pixels with 250 gray levels histogram is orderless(distribution)
	def gen_hist(n_clusters,num_images,desc_list,desc_predict):
		hist = np.array([np.zeros(n_clusters) for i in range(num_images)])
		init = 0
		for i in range(num_images):
			for j in range(len(desc_list[i])):
				idx = desc_predict[init+j]
				hist[i][idx] += 1
			init += 1

		return hist

	hist = gen_hist(n_clusters,num_images,desc_list,desc_predict)
	hist = hist.tolist()

	##############################################################################################
	distances = euclidean_distances(hist, hist)
	distances = distances.tolist()
	# print ('distances ',distances)

	#save distances
	joblib.dump(distances, 'distances'+str(k+1)+'.pkl')

	cluster_assignment = []
	for i in range(len(distances)):
		cluster_assignment.append(np.argsort(distances[i])[1:11])
	# print ('cluster_assignment ',cluster_assignment)

	#save cluster_assignment
	joblib.dump(cluster_assignment, 'cluster_assignment'+str(k+1)+'.pkl')

	# plt.imshow(cv2.cvtColor(trainData[5],cv2.COLOR_BGR2RGB))
	# plt.show()
	# k = 1
	# for i in cluster_assignment[5]:
	# 	plt.subplot(2,5,k)
	# 	plt.imshow(cv2.cvtColor(trainData[i], cv2.COLOR_BGR2RGB))
	# 	# plt.title(kmeans.labels_[i])
	# 	# plt.tight_layout()
	# 	plt.xticks([])
	# 	plt.yticks([])
	# 	k += 1
	# plt.show()

###############################################################################################
# cov = np.cov(hist, rowvar=False)
# nn = NearestNeighbors(algorithm='brute',n_neighbors=10, metric='mahalanobis',metric_params=dict(V=cov))
# distances,indices = nn.fit(hist).kneighbors(hist)

# cluster_assignment = []
# for i in range(len(distances)):
# 	cluster_assignment.append(np.argsort(distances[i])[1:11])
# # print ('cluster_assignment ',cluster_assignment)

# plt.imshow(cv2.cvtColor(trainData[50], cv2.COLOR_BGR2RGB))
# plt.show()
# k = 1
# for i in cluster_assignment[50]:
# 	plt.subplot(2,5,k)
# 	plt.imshow(cv2.cvtColor(trainData[i], cv2.COLOR_BGR2RGB))
# 	plt.xticks([])
# 	plt.yticks([])
# 	k += 1
# plt.show()
################################################################################################


# hist = np.asarray(hist)

# covar = np.cov(hist, rowvar=0)
# if(hist.shape[1:2]==(1,)):
#     invcovar = np.linalg.pinv(covar.reshape(1,1))
# else:
#     invcovar = np.linalg.pinv(covar)

# dis = []
# finDis = []
# for i in hist:
# 	dis = []
# 	for j in hist:
# 		if np.array_equal(i,j) == False:
# 			dis.append(mahalanobis(i,j,invcovar))
# 	finDis.append(dis)

# joblib.dump(finDis, 'finDis'+str(k)+'.pkl')
	# cov = np.cov(hist, rowvar=False)
	# nn = NearestNeighbors(algorithm='brute',n_neighbors=10, metric='mahalanobis',metric_params=dict(V=cov))
	# distances,indices = nn.fit(hist).kneighbors(hist)

# cluster_assignmentN = []
# for i in range(len(finDis)):
# 	cluster_assignmentN.append(np.argsort(finDis[i])[1:11])

# joblib.dump(cluster_assignmentN, 'cluster_assignmentN'+str(k)+'.pkl')