multi-view_feature_aggregation/pre_data2.py at master · wohaiyo/multi-view_feature_aggregation · 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
164
165
166
167
168
169
170
171
172
173
174
from __future__ import print_function
import numpy as np
import config as cfg


IMAGE_HEIGHT = cfg.IMAGE_HEIGHT
IMAGE_WIDTH = cfg.IMAGE_WIDTH

gr_shape = cfg.IMAGE_HEIGHT


# For sequnce input, output single frame
def data_crop_test_output(session, gr_data, logits, image, mean, std, mean_rgb, crop_size_h, crop_size_w, stride):
    image_h = image.shape[0]
    image_w = image.shape[1]
    pad_h = 0
    pad_w = 0
    if image_h >= crop_size_h and image_w >= crop_size_w:
        image_pad = image
    else:
        if image_h < crop_size_h:
            pad_h = crop_size_h - image_h
        if image_w < crop_size_w:
            pad_w = crop_size_w - image_w
        image_pad = np.pad(image, ((0, pad_h), (0, pad_w), (0, 0)), 'constant', constant_values=0)
    image_pad = np.asarray(image_pad, dtype='float32')

    mean_rgb_sub = mean_rgb
    for i in range(1, cfg.seq_num):
        mean_rgb_sub = np.concatenate([mean_rgb_sub, mean_rgb], axis=0)
    image_pad = image_pad - mean_rgb_sub  # sub rgb mean

    image_crop_batch = []
    x_start = [x for x in range(0, image_pad.shape[0] - crop_size_h + 1, stride)]
    y_start = [y for y in range(0, image_pad.shape[1] - crop_size_w + 1, stride)]
    if (image_pad.shape[0] - crop_size_h) % stride != 0:
        x_start.append(image_pad.shape[0] - crop_size_h)
    if (image_pad.shape[1] - crop_size_w) % stride != 0:
        y_start.append(image_pad.shape[1] - crop_size_w)
    for x in x_start:
        for y in y_start:
            image_crop_batch.append(image_pad[x:x + crop_size_h, y:y + crop_size_w])

    logit = []
    for crop_batch in image_crop_batch:
        lo = session.run(logits, feed_dict={gr_data: [crop_batch]})
        logit.append(lo[0])

    num_class = cfg.NUM_OF_CLASSESS
    score_map = np.zeros([image_pad.shape[0], image_pad.shape[1], num_class], dtype='float32')
    count = np.zeros([image_pad.shape[0], image_pad.shape[1], num_class], dtype='float32')
    crop_index = 0
    for x in x_start:
        for y in y_start:
            crop_logits = logit[crop_index]
            score_map[x:x + crop_logits.shape[0], y:y + crop_logits.shape[1], :] += crop_logits
            count[x:x + crop_logits.shape[0], y:y + crop_logits.shape[1], :] += 1
            crop_index += 1

    score_map = score_map[:image_h, :image_w] / count[:image_h, :image_w]
    return score_map

# For pose, depth and intrinsics to input
def data_crop_test_output_pose(session, image_data, depth_data, pose_data, ins_data, logits, image, depth, pose, ins,
                               mean, std, mean_rgb, crop_size_h, crop_size_w, stride):
    image_h = image.shape[0]
    image_w = image.shape[1]
    pad_h = 0
    pad_w = 0
    if image_h >= crop_size_h and image_w >= crop_size_w:
        image_pad = image
        depth_pad = depth
    else:
        if image_h < crop_size_h:
            pad_h = crop_size_h - image_h
        if image_w < crop_size_w:
            pad_w = crop_size_w - image_w
        image_pad = np.pad(image, ((0, pad_h), (0, pad_w), (0, 0)), 'constant', constant_values=0)
        depth_pad = np.pad(depth, ((0, pad_h), (0, pad_w), (0, 0)), 'constant', constant_values=0)
    image_pad = np.asarray(image_pad, dtype='float32')
    depth_pad = np.asarray(depth_pad, dtype='float32')

    mean_rgb_sub = mean_rgb
    for i in range(1, cfg.seq_num):
        mean_rgb_sub = np.concatenate([mean_rgb_sub, mean_rgb], axis=0)
    image_pad = image_pad - mean_rgb_sub  # sub rgb mean

    image_crop_batch = []
    depth_crop_batch = []
    x_start = [x for x in range(0, image_pad.shape[0] - crop_size_h + 1, stride)]
    y_start = [y for y in range(0, image_pad.shape[1] - crop_size_w + 1, stride)]
    if (image_pad.shape[0] - crop_size_h) % stride != 0:
        x_start.append(image_pad.shape[0] - crop_size_h)
    if (image_pad.shape[1] - crop_size_w) % stride != 0:
        y_start.append(image_pad.shape[1] - crop_size_w)
    for x in x_start:
        for y in y_start:
            image_crop_batch.append(image_pad[x:x + crop_size_h, y:y + crop_size_w])
            depth_crop_batch.append(depth_pad[x:x + crop_size_h, y:y + crop_size_w])

    logit = []
    for i, crop_batch in enumerate(image_crop_batch):
        lo = session.run(logits, feed_dict={image_data: [crop_batch], depth_data: [depth_crop_batch[i]],
                                            pose_data: [pose], ins_data: [ins]})
        logit.append(lo[0])

    num_class = cfg.NUM_OF_CLASSESS
    score_map = np.zeros([image_pad.shape[0], image_pad.shape[1], num_class], dtype='float32')
    count = np.zeros([image_pad.shape[0], image_pad.shape[1], num_class], dtype='float32')
    crop_index = 0
    for x in x_start:
        for y in y_start:
            crop_logits = logit[crop_index]
            score_map[x:x + crop_logits.shape[0], y:y + crop_logits.shape[1], :] += crop_logits
            count[x:x + crop_logits.shape[0], y:y + crop_logits.shape[1], :] += 1
            crop_index += 1

    score_map = score_map[:image_h, :image_w] / count[:image_h, :image_w]
    return score_map

# For multiple frames to output
def data_crop_test_output_seq(session, gr_data, logits, image, mean, std, mean_rgb, crop_size_h, crop_size_w, stride):
    image_h = image.shape[0]
    image_w = image.shape[1]
    pad_h = 0
    pad_w = 0
    if image_h >= crop_size_h and image_w >= crop_size_w:
        image_pad = image
    else:
        if image_h < crop_size_h:
            pad_h = crop_size_h - image_h
        if image_w < crop_size_w:
            pad_w = crop_size_w - image_w
        image_pad = np.pad(image, ((0, pad_h), (0, pad_w), (0, 0)), 'constant', constant_values=0)
    image_pad = np.asarray(image_pad, dtype='float32')

    mean_rgb_sub = mean_rgb
    for i in range(1, cfg.seq_num):
        mean_rgb_sub = np.concatenate([mean_rgb_sub, mean_rgb], axis=0)
    image_pad = image_pad - mean_rgb_sub  # sub rgb mean

    image_crop_batch = []
    x_start = [x for x in range(0, image_pad.shape[0] - crop_size_h + 1, stride)]
    y_start = [y for y in range(0, image_pad.shape[1] - crop_size_w + 1, stride)]
    if (image_pad.shape[0] - crop_size_h) % stride != 0:
        x_start.append(image_pad.shape[0] - crop_size_h)
    if (image_pad.shape[1] - crop_size_w) % stride != 0:
        y_start.append(image_pad.shape[1] - crop_size_w)
    for x in x_start:
        for y in y_start:
            image_crop_batch.append(image_pad[x:x + crop_size_h, y:y + crop_size_w])

    logit = []
    for crop_batch in image_crop_batch:
        lo = session.run(logits, feed_dict={gr_data: [crop_batch]})
        logit_img = lo[0][0]
        for i in range(1, cfg.seq_num):
            logit_img = np.concatenate([logit_img, lo[i][0]], axis=2)
        logit.append(logit_img)


    num_class = int(cfg.NUM_OF_CLASSESS * cfg.seq_num)
    score_map = np.zeros([image_pad.shape[0], image_pad.shape[1], num_class], dtype='float32')
    count = np.zeros([image_pad.shape[0], image_pad.shape[1], num_class], dtype='float32')
    crop_index = 0
    for x in x_start:
        for y in y_start:
            crop_logits = logit[crop_index]
            score_map[x:x + crop_logits.shape[0], y:y + crop_logits.shape[1], :] += crop_logits
            count[x:x + crop_logits.shape[0], y:y + crop_logits.shape[1], :] += 1
            crop_index += 1

    score_map = score_map[:image_h, :image_w] / count[:image_h, :image_w]
    return score_map