adaptive_weighted_attribute/python_data_layer.py at master · qiexing/adaptive_weighted_attribute · 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
159
160
161
162
163
import numpy as np
import sys
import cv2
import caffe
import os
import random
import time
from pandas.io.parsers import read_csv
import matplotlib.pyplot as plt

datadir = 'data/CelebA/'
mean_file = 'model/resnet_50/ResNet_mean.binaryproto'
proto_data = open(mean_file, "rb").read()
a = caffe.io.caffe_pb2.BlobProto.FromString(proto_data)
mean = caffe.io.blobproto_to_array(a)[0]
selected_attr = np.zeros(40)
for attr_index in range(40):
    selected_attr[attr_index] = attr_index


def pre_process(color_img):
    resized_img = cv2.resize(color_img, (224, 224))
    return np.transpose(resized_img, (2, 0, 1)) - mean


def showImage(img,points=None, bbox=None):
    if points is not None:
        for i in range(0,points.shape[0]/2):
            cv2.circle(img,(int(round(points[i*2])),int(points[i*2+1])),1,(0,0,255),2)
    if bbox is not None:
        cv2.rectangle(img, (int(bbox[0]), int(bbox[2])), (int(bbox[1]), int(bbox[3])), (0,0,255), 2)
    plt.figure()
    plt.imshow(img)
    plt.show()
    print 'here'


class ValLayer(caffe.Layer):

    attri_num = 40

    """
    Compute the Euclidean Loss in the same manner as the C++ EuclideanLossLayer
    to demonstrate the class interface for developing layers in Python.
    """
    total_namelist = []
    attri_array =[]
    # landmark_array =[]
    img_num = 0

    target_height = 224
    target_witdh = 224
    # mean = []
    batch = 20
    imgset = 'align_val'


    def setup(self, bottom, top):
        filename = os.path.join(datadir, self.imgset, self.imgset + 'list.txt')
        attrfile = os.path.join(datadir, self.imgset, self.imgset + '_attr.txt')
        det_df = read_csv(filename,sep=' ',header=None)
        self.total_namelist = det_df[det_df.columns[0]].values
        self.img_num = len(self.total_namelist)
        self.attri_array = np.zeros((self.img_num, self.attri_num))
        attr_f = open(attrfile, 'r')
        attr_line = attr_f.readline().strip().split()
        i = 0
        while attr_line:
            for j in range(0, self.attri_num):
                value = int(float(attr_line[j + 1]))
                assert value == 1 or value == -1
                self.attri_array[i, j] = value
            attr_line = attr_f.readline().strip().split()
            i += 1
        top[0].reshape(self.batch,3,self.target_height, self.target_witdh)
        top[1].reshape(self.batch, len(selected_attr))

    def reshape(self, bottom, top):
        pass

    def forward(self, bottom, top):
        for i in range(self.batch):
            # print i
            idx = random.randint(0,self.img_num-1)
            im = cv2.imread(self.total_namelist[idx])
            if im is not None:
                top[0].data[i] = pre_process(im)
                for k in range(0, len(selected_attr)):
                    top[1].data[i][k] = self.attri_array[idx][int(selected_attr[k])]

    def backward(self, top, propagate_down, bottom):
        pass


class JointAttributeLayer(caffe.Layer):

    attri_num = 40
    point_num = 5

    """
    Compute the Euclidean Loss in the same manner as the C++ EuclideanLossLayer
    to demonstrate the class interface for developing layers in Python.
    """
    train_total_namelist = []
    train_attri_array = []
    train_img_num = 0

    val_total_namelist = []
    val_attri_array =[]
    val_img_num = 0
    target_height = 224
    target_witdh = 224
    batch = 10
    train_imgset = 'align_train'
    val_imgset = 'align_val'

    def do_setup(self, imgset, attri_num):
        filename = os.path.join(datadir, imgset, imgset + 'list.txt')
        attrfile = os.path.join(datadir, imgset, imgset + '_attr.txt')
        det_df = read_csv(filename,sep=' ', header=None)
        total_namelist = det_df[det_df.columns[0]].values
        img_num = len(total_namelist)
        attri_array = np.zeros((img_num, attri_num))
        attr_f = open(attrfile, 'r')
        attr_line = attr_f.readline().strip().split()
        i = 0
        while attr_line:
            for j in range(0, attri_num):
                value = int(float(attr_line[j + 1]))
                assert value == 1 or value == -1
                attri_array[i, j] = value
            attr_line = attr_f.readline().strip().split()
            i += 1
        return total_namelist, img_num, attri_array


    def setup(self, bottom, top):
        self.train_total_namelist, self.train_img_num, self.train_attri_array = self.do_setup(self.train_imgset, self.attri_num)
        self.val_total_namelist, self.val_img_num, self.val_attri_array = self.do_setup(self.val_imgset, self.attri_num)
        top[0].reshape(2 * self.batch,3,self.target_height,self.target_witdh)
        top[1].reshape(2 * self.batch, len(selected_attr))

    def reshape(self, bottom, top):
        pass

    def do_forward(self, top, begin_index, total_namelist, img_num, attri_array):
        for i in range(self.batch):
            # print i
            idx = random.randint(0, img_num-1)
            im = cv2.imread(total_namelist[idx])
            if im is not None:
                top[0].data[i + begin_index] = pre_process(im)
                for k in range(0, len(selected_attr)):
                    top[1].data[i + begin_index][k] = attri_array[idx][int(selected_attr[k])]

    def forward(self, bottom, top):
         self.do_forward(top, 0, self.train_total_namelist, self.train_img_num,
                         self.train_attri_array)
         self.do_forward(top, self.batch, self.val_total_namelist, self.val_img_num,
                         self.val_attri_array)

    def backward(self, top, propagate_down, bottom):
        pass