conditional-image-generation/preprocess.py at master · patrickmesana/conditional-image-generation · 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
import os, sys
import glob
import cPickle as pkl
import numpy as np
import PIL.Image as Image
import ntpath
import readsample as rs

def resize_mscoco():
    '''
    function used to create the dataset,
    Resize original MS_COCO Image into 64x64 images
    '''

    ### PATH need to be fixed
    data_path="/datasets/coco/coco/images/train2014"
    save_dir = "/Tmp/64_64/train2014/"

    if not os.path.exists(save_dir):
        os.makedirs(save_dir)

    preserve_ratio = True
    image_size = (64, 64)
    #crop_size = (32, 32)

    imgs = glob.glob(data_path+"/*.jpg")


    for i, img_path in enumerate(imgs):
        img = Image.open(img_path)
        print i, len(imgs), img_path

        if img.size[0] != image_size[0] or img.size[1] != image_size[1] :
            if not preserve_ratio:
                img = img.resize((image_size), Image.ANTIALIAS)
            else:
                ### Resize based on the smallest dimension
                scale = image_size[0] / float(np.min(img.size))
                new_size = (int(np.floor(scale * img.size[0]))+1, int(np.floor(scale * img.size[1])+1))
                img = img.resize((new_size), Image.ANTIALIAS)

                ### Crop the 64/64 center
                tocrop = np.array(img)
                center = (int(np.floor(tocrop.shape[0] / 2.)), int(np.floor(tocrop.shape[1] / 2.)))
                print tocrop.shape, center, (center[0]-32,center[0]+32), (center[1]-32,center[1]+32)
                if len(tocrop.shape) == 3:
                    tocrop = tocrop[center[0]-32:center[0]+32, center[1] - 32:center[1]+32, :]
                else:
                    tocrop = tocrop[center[0]-32:center[0]+32, center[1] - 32:center[1]+32]
                img = Image.fromarray(tocrop)

        img.save(save_dir + os.path.basename(img_path))


def createInputAndTarget(
                  mscoco="inpainting/", split="train2014"):
    '''
    Show an example of how to read the dataset
    '''

    data_path = os.path.join(mscoco, split)
    imgs = glob.glob(data_path + "/*.jpg")

    for i, img_path in enumerate(imgs):
        print(img_path)
        img = Image.open(img_path)
        img_array = np.array(img)

        ### Get input/target from the images
        center = (int(np.floor(img_array.shape[0] / 2.)), int(np.floor(img_array.shape[1] / 2.)))
        if len(img_array.shape) == 3:
            input = np.copy(img_array)
            input[center[0]-16:center[0]+16, center[1]-16:center[1]+16, :] = 0
            target = img_array[center[0] - 16:center[0] + 16, center[1] - 16:center[1] + 16, :]
        else:
            input = np.copy(img_array)
            input[center[0]-16:center[0]+16, center[1]-16:center[1]+16] = 0
            target = img_array[center[0] - 16:center[0] + 16, center[1] - 16:center[1] + 16]


        sample_name = 'training'
        if split == 'val2014':
            sample_name = 'validation'

        head, tail = ntpath.split(img_path)
        img_name = tail or ntpath.basename(head)

        Image.fromarray(img_array).save('./{sn}/target-full/{n}'.format(sn=sample_name, n=img_name))
        Image.fromarray(input).save('./{sn}/input/{n}'.format(sn=sample_name, n=img_name))
        Image.fromarray(target).save('./{sn}/target/{n}'.format(sn=sample_name, n=img_name))


def create_full_with_middle(imgs):
    n = len(imgs)
    np_images = []
    for i in range(0, n):
        img_array = imgs[i]
        black_img = np.zeros([64, 64, 3], dtype=np.uint8)
        center = (int(np.floor(black_img.shape[0] / 2.)), int(np.floor(black_img.shape[1] / 2.)))
        black_img[center[0] - 16:center[0] + 16, center[1] - 16:center[1] + 16, :] = img_array
        np_images.append(img_array)
    return np_images

def persist_samples_on_disk():
    training_input_images = rs.read_sample_as_tensor(base_path="training/", split="input")
    training_target_full_images = rs.read_sample_as_tensor(base_path="training/", split="target-full")
    training_target_images = rs.read_sample_as_tensor(base_path="training/", split="target")
    validation_input_images = rs.read_sample_as_tensor(base_path="validation/", split="input")
    validation_target_full_images = rs.read_sample_as_tensor(base_path="validation/", split="target-full")
    validation_target_images = rs.read_sample_as_tensor(base_path="validation/", split="target")
    rs.write_images_to_pkl(training_input_images, 'training_input.pkl')
    rs.write_images_to_pkl(training_target_full_images, 'training_target_full.pkl')
    rs.write_images_to_pkl(training_target_images, 'training_target.pkl')
    rs.write_images_to_pkl(validation_input_images, 'validation_input.pkl')
    rs.write_images_to_pkl(validation_target_full_images, 'validation_target_full.pkl')
    rs.write_images_to_pkl(validation_target_images, 'validation_target.pkl')

# resize_mscoco()
# show_examples(5, 5, mscoco="/Users/patrickmesana/Dev/inpainting")
createInputAndTarget(split="train2014")
createInputAndTarget(split="val2014")
persist_samples_on_disk()

# rs.write_images_to_pkl(create_full_with_middle(x_train_target), 'training_target2.pkl')
# saved_images1 = rs.read_images_from_pkl('training_target2.pkl')
# rs.write_images_to_pkl(create_full_with_middle(x_test_target), 'validation_target2.pkl')
# saved_images2 = rs.read_images_from_pkl('validation_target2.pkl')