-
Notifications
You must be signed in to change notification settings - Fork 6
/
eval_single_task.py
160 lines (131 loc) · 6.93 KB
/
eval_single_task.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
import os
import tensorflow as tf
import numpy as np
from PIL import Image
from nets import vgg_arg_scope
from nets import vgg_16_deeplab_st, vgg_16_shortcut_deeplab_st
from ops.errors import seg_error, normal_error
from util.img_encoder_st import ImageReader
from util.img_decoder import inv_preprocess, decode_labels
from util.input_arguments import arguments_st_eval
slim = tf.contrib.slim
IMG_MEAN = np.array((104.00698793,116.66876762,122.67891434), dtype=np.float32)
def eval():
"""Create the model and start the evaluation process."""
args = arguments_st_eval()
# Create queue coordinator.
coord = tf.train.Coordinator()
# Encode the data.
with tf.name_scope("create_inputs"):
reader = ImageReader(
args.data_dir,
args.data_list,
input_size=None, # No defined input size.
random_scale=False, # No random scale.
random_mirror=False, # No random mirror.
random_crop=False, # No random crop.
ignore_label=args.ignore_label,
img_mean=IMG_MEAN,
coord=coord,
task=args.task)
image, label = reader.image, reader.label
image_batch, label_batch = tf.expand_dims(image, dim=0), tf.expand_dims(label, dim=0) # Add one batch dimension.
# Create network.
with slim.arg_scope(vgg_arg_scope(weight_decay=0.0, use_batch_norm=True, is_training=False)):
if args.network == 'vgg_16_deeplab_st':
net, end_points = vgg_16_deeplab_st(image_batch, num_classes=args.num_classes, is_training=False, dropout_keep_prob=1.0)
elif args.network == 'vgg_16_shortcut_deeplab_st':
net, end_points = vgg_16_shortcut_deeplab_st(image_batch, num_classes=args.num_classes, is_training=False, dropout_keep_prob=1.0)
else:
raise Exception('network name is not recognized!')
# Which variables to load.
restore_var = tf.global_variables()
# Predictions.
raw_output = end_points['vgg_16/fc8_voc12']
raw_output = tf.image.resize_bilinear(raw_output, tf.shape(image_batch)[1:3,])
if args.task == 'normal':
save_pred = tf.nn.l2_normalize(raw_output, dim=-1) * 255
cos_distance = normal_error(raw_output, label_batch, args.num_classes, args.ignore_label)
elif args.task == 'seg':
save_pred = tf.expand_dims(tf.argmax(raw_output, axis=3), dim=3)
mIoU, update_op, correct_pixel, valid_pixel = seg_error(raw_output, label_batch, args.num_classes, args.ignore_label)
else:
raise Exception('task name is not recognized!')
# Save folder
if args.save_dir is not None and not os.path.exists(args.save_dir):
os.makedirs(args.save_dir)
# Set up tf session and initialize variables.
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
init = tf.global_variables_initializer()
sess.run(init)
sess.run(tf.local_variables_initializer())
# Load weights.
if args.restore_from is not None:
if tf.gfile.IsDirectory(args.restore_from):
folder_name = args.restore_from
checkpoint_path = tf.train.latest_checkpoint(args.restore_from)
else:
folder_name = args.restore_from.replace(args.restore_from.split('/')[-1], '')
checkpoint_path = args.restore_from
tf.train.Saver(var_list=restore_var).restore(sess, checkpoint_path)
print("Restored model parameters from {}".format(checkpoint_path))
# Start queue threads.
threads = tf.train.start_queue_runners(coord=coord, sess=sess)
# Iterate over evaluation steps.
correct_pixel_all = []
valid_pixel_all = []
cos_distance_all = []
for step in range(args.num_steps):
if args.task == 'seg':
imgs, save_preds, cp, vp, _ = sess.run([image_batch, save_pred, correct_pixel, valid_pixel, update_op])
correct_pixel_all.append(cp)
valid_pixel_all.append(vp)
elif args.task == 'normal':
imgs, save_preds, cd = sess.run([image_batch, save_pred, cos_distance])
cos_distance_all.append(cd)
else:
raise Exception('task name is not recognized!')
if step % 100 == 0:
print('step {:d}'.format(step))
if args.save_dir is not None:
org_imgs = inv_preprocess(imgs, 1, IMG_MEAN)
org_img = org_imgs[0]
org_img = Image.fromarray(org_img)
org_img.save(args.save_dir + '/org_' + str(step) + '.png')
save_imgs = decode_labels(save_preds, num_images=1, num_classes=args.num_classes, task=args.task)
save_img = save_imgs[0]
save_img = Image.fromarray(save_img)
save_img.save(args.save_dir + '/save_' + str(step) + '.png')
if args.task == 'seg':
mIoU_value = sess.run(mIoU)
print('Mean IoU: {:.3f}'.format(mIoU_value))
pixel_acc = sum(correct_pixel_all) / (sum(valid_pixel_all) + 0.0)
print('Pixel Acc: {:.3f}'.format(pixel_acc))
with open(folder_name + '/results.txt', 'a') as f:
f.write(checkpoint_path.split('/')[-1] + ' Mean IoU: {:.3f} \n'.format(mIoU_value))
f.write(checkpoint_path.split('/')[-1] + ' Pixel Acc: {:.3f} \n'.format(pixel_acc))
elif args.task == 'normal':
cos_distance_all = np.concatenate(cos_distance_all, axis=0)
cosine_distance = np.minimum(np.maximum(cos_distance_all, -1.0), 1.0)
angles = np.arccos(cosine_distance) / np.pi * 180.0
print('Mean: {:.3f}'.format(np.mean(angles)))
print('Median: {:.3f}'.format(np.median(angles)))
print('RMSE: {:.3f}'.format(np.sqrt(np.mean(angles ** 2))))
print('11.25: {:.3f}'.format(np.mean(np.less_equal(angles, 11.25)) * 100))
print('22.5: {:.3f}'.format(np.mean(np.less_equal(angles, 22.5)) * 100))
print('30: {:.3f}'.format(np.mean(np.less_equal(angles, 30.0)) * 100))
print('45: {:.3f}'.format(np.mean(np.less_equal(angles, 45.0)) * 100))
with open(folder_name + '/results.txt', 'a') as f:
f.write(checkpoint_path.split('/')[-1] + ' Mean: {:.3f} \n'.format(np.mean(angles)))
f.write(checkpoint_path.split('/')[-1] + ' Median: {:.3f} \n'.format(np.median(angles)))
f.write(checkpoint_path.split('/')[-1] + ' RMSE: {:.3f} \n'.format(np.sqrt(np.mean(angles ** 2))))
f.write(checkpoint_path.split('/')[-1] + ' 11.25: {:.3f} \n'.format(np.mean(np.less_equal(angles, 11.25)) * 100))
f.write(checkpoint_path.split('/')[-1] + ' 22.5: {:.3f} \n'.format(np.mean(np.less_equal(angles, 22.5)) * 100))
f.write(checkpoint_path.split('/')[-1] + ' 30: {:.3f} \n'.format(np.mean(np.less_equal(angles, 30.0)) * 100))
f.write(checkpoint_path.split('/')[-1] + ' 45: {:.3f} \n'.format(np.mean(np.less_equal(angles, 40.0)) * 100))
coord.request_stop()
coord.join(threads)
if __name__ == '__main__':
eval()