-
Notifications
You must be signed in to change notification settings - Fork 2
/
labels_from_segnet.py
174 lines (151 loc) · 5.86 KB
/
labels_from_segnet.py
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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
import sys # NOQA isort:skip
sys.path.insert(0, 'datasets') # NOQA isort:skip
import argparse
import glob
import json
import os
import zipfile
from PIL import Image
import chainer
from chainer import serializers
from chainercv import evaluations
import matplotlib.pyplot as plt
import numpy as np
from tqdm import tqdm
import train_segnet
from zipped_cityscapes_road_dataset import ZippedCityscapesRoadDataset
chainer.config.train = False
def save_labels(param_dir, iteration, gpu, img_zip_fn, label_zip_fn, out_dir,
start_index, end_index, soft_label, eval_shape,
save_each=False):
train_args = json.load(open(os.path.join(param_dir, 'args.txt')))
if not os.path.exists(out_dir):
try:
os.makedirs(out_dir)
except Exception:
pass
# Find the target snapshot
snapshots = sorted(glob.glob(os.path.join(param_dir, 'snapshot_*')))
for snapshot in snapshots:
if 'iter_{}'.format(iteration) in snapshot:
break
# Create model
if train_args['model'] == 'basic':
model = train_segnet.SegNetBasic(n_class=2, pred_shape=eval_shape)
elif train_args['model'] == 'normal':
model = train_segnet.SegNet(n_class=2)
# Load model parameters
serializers.load_npz(
snapshot, model, path='updater/model:main/predictor/')
if gpu >= 0:
chainer.cuda.get_device_from_id(gpu).use()
model.to_gpu(gpu)
# Create dataset
d = ZippedCityscapesRoadDataset(
img_zip_fn, label_zip_fn, train_args['input_shape'])
if end_index > len(d):
raise ValueError(
'end_index option should be less than the length of dataset '
'{} but {} was given.'.format(len(d), end_index))
if not save_each:
pred_and_scores = {}
for i in tqdm(range(start_index, end_index)):
img, label = d[i]
pred, score = model.predict([img], True)[0]
assert pred.ndim == 2, pred.ndim
assert pred.shape == tuple(eval_shape), \
'pred:{} but eval_shape:{}'.format(pred.shape, eval_shape)
assert score.ndim == 3, score.ndim
assert score.shape[1:] == tuple(eval_shape), \
'score[1:]:{} but eval_shape: {}'.format(
score.shape[1:], eval_shape)
# Evaluate prediction
ret = evaluations.calc_semantic_segmentation_confusion([pred], [label])
TP = int(ret[1, 1])
FP = int(ret[0, 1])
FN = int(ret[1, 0])
precision = float(TP / (TP + FP)) if TP + FP > 0 else None
recall = float(TP / (TP + FN)) if TP + FN > 0 else None
iou = evaluations.calc_semantic_segmentation_iou(ret)
pred = pred.astype(np.bool)
score = score.astype(np.float32)
fn_base = os.path.splitext(os.path.basename(d.img_fns[i]))[0]
save_fn = os.path.join(out_dir, fn_base)
if save_each:
np.save(save_fn, pred)
np.save(save_fn + '_scores', pred)
else:
pred_and_scores[save_fn] = pred
pred_and_scores[save_fn + '_scores'] = score
plt.clf()
fig, axes = plt.subplots(1, 3)
fig.set_dpi(300)
axes[0].axis('off')
axes[1].axis('off')
axes[2].axis('off')
# Show result
img = np.array(Image.open(d.img_fns[i]), dtype=np.uint8)
axes[0].imshow(img)
axes[0].imshow(pred, alpha=0.4, cmap=plt.cm.Set1_r)
axes[0].set_title('Estimated road mask (input image overlayed)',
fontsize=4)
# Show labels
axes[1].imshow(label == 1)
axes[1].set_title('Ground truth road mask', fontsize=4)
# Show road estimation
axes[2].imshow(pred)
axes[2].set_title('Estimated road mask', fontsize=4)
plt.savefig(os.path.join(out_dir, os.path.basename(d.img_fns[i])),
bbox_inches='tight')
plt.close()
with open(os.path.join(out_dir, 'result.json'), 'a') as fp:
result_info = {
'img_fn': d.img_fns[i],
'label_fn': d.label_fns[i],
'road_iou': iou[1],
'non_road_iou': iou[0],
'precision': precision,
'recall': recall,
'TP': TP,
'FP': FP,
'FN': FN
}
result_info.update({
'param_dir': param_dir,
'iteration': iteration,
'gpu': gpu,
'img_zip_fn': img_zip_fn,
'label_zip_fn': label_zip_fn,
'out_dir': out_dir,
'start_index': start_index,
'end_index': end_index,
'soft_label': soft_label,
'eval_shape': eval_shape,
'save_each': save_each,
})
result_info.update({'train_args': train_args})
print(json.dumps(result_info), file=fp)
chainer.cuda.memory_pool.free_all_blocks()
del model
if not save_each:
return pred_and_scores
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--param_dir', type=str)
parser.add_argument('--iteration', type=int)
parser.add_argument('--gpu', type=int, default=-1)
parser.add_argument('--img_zip_fn', type=str)
parser.add_argument('--label_zip_fn', type=str)
parser.add_argument('--out_dir', type=str)
parser.add_argument('--start_index', type=int)
parser.add_argument('--end_index', type=int)
parser.add_argument('--soft_label', action='store_true', default=False)
parser.add_argument(
'--eval_shape', type=int, nargs=2, default=[1024, 2048])
args = parser.parse_args()
save_labels(
args.param_dir, args.iteration, args.gpu, args.img_zip_fn,
args.label_zip_fn, args.out_dir, args.start_index, args.end_index,
args.soft_label, args.eval_shape, True)