pytorch-layoutnet/torch2pytorch_data.py at master · sunset1995/pytorch-layoutnet · 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
import os
import glob
from torch.utils.serialization import load_lua
import numpy as np
from scipy.io import loadmat
from scipy.ndimage import convolve
from pano import find_N_peaks

from PIL import Image

ROOT_DIR = os.path.dirname(os.path.realpath(__file__))
DATA_DIR = os.path.join(ROOT_DIR, 'data')
ORGIN_DATA_DIR = os.path.join(DATA_DIR, 'origin', 'data')
ORGIN_GT_DIR = os.path.join(DATA_DIR, 'origin', 'gt')


# Variables for train/val/test split
cat_list = ['img', 'line', 'edge', 'cor']
train_pats = [
    'panoContext_%s_train.t7',
    'stanford2d-3d_%s_area_1.t7', 'stanford2d-3d_%s_area_2.t7',
    'stanford2d-3d_%s_area_4.t7', 'stanford2d-3d_%s_area_6.t7']
valid_pats = ['panoContext_%s_val.t7', 'stanford2d-3d_%s_area_3.t7']
test_pats = ['panoContext_%s_test.t7', 'stanford2d-3d_%s_area_5.t7']

train_pano_map = os.path.join(DATA_DIR, 'panoContext_trainmap.txt')
valid_pano_map = os.path.join(DATA_DIR, 'panoContext_valmap.txt')
test_pano_map = os.path.join(DATA_DIR, 'panoContext_testmap.txt')


def cvt2png(target_dir, patterns, pano_map_path):
    os.makedirs(target_dir, exist_ok=True)
    for cat in cat_list:
        for pat in patterns:
            # Define source file paths
            th_path = os.path.join(ORGIN_DATA_DIR, pat % cat)
            assert os.path.isfile(th_path), '%s not found !!!' % th_path

            if pat.startswith('stanford'):
                gt_path = os.path.join(
                    ORGIN_GT_DIR, 'pano_id_%s.txt' % pat[-9:-3])
            else:
                gt_path = os.path.join(
                    ORGIN_GT_DIR, 'panoContext_%s.txt' % pat.split('_')[-1].split('.')[0])
            assert os.path.isfile(gt_path), '%s not found !!!' % gt_path

            # Parse file names from gt list
            with open(gt_path) as f:
                fnames = [line.strip() for line in f]
            print('%-30s: %3d examples' % (pat % cat, len(fnames)))

            # Remapping panoContext filenames
            if pat.startswith('pano'):
                fnames_cnt = dict([(v, 0) for v in fnames])
                with open(pano_map_path) as f:
                    for line in f:
                        v, k, _ = line.split()
                        k = int(k)
                        fnames[k] = v
                        fnames_cnt[v] += 1
                for v in fnames_cnt.values():
                    assert v == 1

            # Parse th file
            imgs = load_lua(th_path).numpy()
            assert imgs.shape[0] == len(fnames), 'number of data and gt mismatched !!!'

            # Dump each images to target direcotry
            target_cat_dir = os.path.join(target_dir, cat)
            os.makedirs(target_cat_dir, exist_ok=True)
            for img, fname in zip(imgs, fnames):
                target_path = os.path.join(target_cat_dir, fname)
                if img.shape[0] == 3:
                    # RGB
                    Image.fromarray(
                        (img.transpose([1, 2, 0]) * 255).astype(np.uint8)).save(target_path)
                else:
                    # Gray
                    Image.fromarray(
                        (img[0] * 255).astype(np.uint8)).save(target_path)


cvt2png(os.path.join(DATA_DIR, 'train'), train_pats, train_pano_map)
cvt2png(os.path.join(DATA_DIR, 'valid'), valid_pats, valid_pano_map)
cvt2png(os.path.join(DATA_DIR, 'test'), test_pats, test_pano_map)

# Convert ground truth corner map to corner index
for split in ['test', 'valid', 'train']:
    target_dir = os.path.join(DATA_DIR, split, 'label_cor')
    os.makedirs(target_dir, exist_ok=True)

    # Kernel used to extract corner position from gt corner map
    kernel = np.array([
        [0, 1, 0],
        [1, 1, 1],
        [0, 1, 0],
    ])

    for path in glob.glob('data/%s/cor/*' % split):
        img_cor = np.array(Image.open(path))
        signal = (img_cor == 255).astype(np.int)
        signal = convolve(signal, kernel, mode='constant', cval=0.0)
        if (signal == 5).sum() != 8:
            cor_id = []
            X_loc = find_N_peaks(signal.sum(0), prominence=None,
                                 distance=20, N=4)[0]
            for x in X_loc:
                x = int(np.round(x))
                V_signal = signal[:, max(0, x-15):x+15].sum(1)
                y1, y2 = find_N_peaks(V_signal, prominence=None,
                                      distance=20, N=2)[0]
                cor_id.append((x, y1))
                cor_id.append((x, y2))
            cor_id = np.array(cor_id)
        else:
            cor_id = np.array([(x, y) for x, y in zip(*np.where((signal.T == 5)))])

        # Arange corner in order if needed
        for i in range(1, len(cor_id), 2):
            if cor_id[i, 1] < cor_id[i-1, 1]:
                cor_id[[i-1, i]] = cor_id[[i, i-1]]
            cor_id[[i-1, i], 0] = cor_id[[i-1, i], 0].mean()

        basename = path.split('/')[-1].split('.')[0]
        with open(os.path.join(target_dir, '%s.txt' % basename), 'w') as f:
            for x, y in cor_id:
                f.write('%d %d\n' % (x, y))