-
Notifications
You must be signed in to change notification settings - Fork 11
/
train_u-net.py
158 lines (105 loc) · 5.08 KB
/
train_u-net.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
from __future__ import print_function
import tensorflow as tf
import custom_layers_unet
import read_sunrgbd_data
from PIL import Image
import argparse
from UNet import unet
import time
import numpy as np
headless = 'True'
img_width = 320
img_height = 240
if headless == False:
import matplotlib.pyplot as pl
import matplotlib as mpl
pl.close('all')
def tile_images(img, batch_size, rows, cols, rgb):
batchImages = np.random.random((img_height*rows,img_width*cols,rgb))
if rgb>1:
batchImages = np.random.random((img_height*rows,img_width*cols,rgb))
else:
batchImages = np.random.random((img_height*rows,img_width*cols))
for i in range(rows):
for j in range(cols):
if i*cols+j < batch_size:
if rgb > 1:
batchImages[0+i*img_height:(i+1)*img_height,0+j*img_width:(j+1)*img_width,:] = img[i*cols+j]
else:
batchImages[0+i*img_height:(i+1)*img_height,0+j*img_width:(j+1)*img_width] = img[i*cols+j]
return batchImages
# Training settings
parser = argparse.ArgumentParser(description='plotting example')
parser.add_argument('--batch-size', type=int, default=20, metavar='N',
help='input batch size for training (default: 64)')
args = parser.parse_args()
rows = np.int(np.ceil(np.sqrt(args.batch_size)))
cols = np.int(np.ceil(args.batch_size / rows))
#SUNRGBD_dataset = read_sunrgbd_data.dataset("SUNRGBD","/data/ahanda/sunrgbd-meta-data/sunrgbd_rgb_training.txt")
#SUNRGBD_dataset = read_sunrgbd_data.dataset("SUNRGBD","/data/workspace/sunrgbd-meta-data/sunrgbd_rgb_training.txt")
SUNRGBD_dataset = read_sunrgbd_data.dataset("SUNRGBD",
# "/data/ahanda/code/baxter_data_renderer/data/multijtdata/baxter_babbling_rarm_3.5hrs_Dec14_16/postprocessmotions/motion0",
"/se3netsproject/data/multijtdata/baxter_babbling_rarm_3.5hrs_Dec14_16/postprocessmotions/motion0",
img_type='depth')
# SUNRGBD_dataset = read_sunrgbd_data.dataset("SUNRGBD","/Users/ankurhanda/workspace/code/sunrgbd-meta-data/sunrgbd_training.txt")
max_labels = 23
#inspired by http://jdherman.github.io/colormap/
# colour_code = [(0, 0, 0),(0,0,1),(0.9137,0.3490,0.1882), (0, 0.8549, 0),
# (0.5843,0,0.9412),(0.8706,0.9451,0.0941),(1.0000,0.8078,0.8078),
# (0,0.8784,0.8980),(0.4157,0.5333,0.8000),(0.4588,0.1137,0.1608),
# (0.9412,0.1373,0.9216),(0,0.6549,0.6118),(0.9765,0.5451,0),
# (0.8824,0.8980,0.7608)]
if headless == 'False':
colour_code = np.random.rand(max_labels, 3)
colour_code[0] = [0, 0, 0]
cm = mpl.colors.ListedColormap(colour_code)
fig, ax = pl.subplots()
someImage = np.random.random((img_height*np.int(rows),img_width*np.int(cols),max_labels))
some_img_argmax = np.argmax(someImage, axis=2)
# Turn off axes and set axes limits
im = ax.imshow(some_img_argmax, interpolation='none', cmap=cm)
ax.axis('tight')
ax.axis('off')
# Set whitespace to 0
fig.subplots_adjust(left=0,right=1,bottom=0,top=1)
fig.show()
# config=tf.ConfigProto(allow_soft_placement=True, log_device_placement=False)
config=tf.ConfigProto(log_device_placement=False)
config.gpu_options.allow_growth = True
batch_size = 20
learning_rate = 1e-3
iter = 0
logs_path = '/tensorboard/tf-summary-logs/'
img_type = 'depth'
with tf.Session(config=config) as sess:
UNET = unet(batch_size, img_height, img_width, learning_rate, sess, num_classes=max_labels, is_training=True,
img_type=img_type)
sess.run(tf.global_variables_initializer())
summary_writer = tf.summary.FileWriter(logs_path, graph=tf.get_default_graph())
while True:
img, label = SUNRGBD_dataset.get_random_shuffle(batch_size)
batch_labels = label
label = np.reshape(label, [-1])
if iter <= 10:
UNET.set_learning_rate(learning_rate=1e-2)
elif (iter > 10 and iter <= 500):
UNET.set_learning_rate(learning_rate=1e-3)
else:
UNET.set_learning_rate(learning_rate=1e-4)
batch_start = time.time()
train_op, cost, pred, summary = UNET.train_batch(img, label)
time_taken = time.time() - batch_start
images_per_sec = batch_size / time_taken
summary_writer.add_summary(summary, iter)
if headless == 'False':
fig.show()
pl.pause(0.00001)
pred_class = np.argmax(pred, axis=3)
batch_labels[batch_labels > 0] = 1
pred_class_gt_mask = np.multiply(pred_class, batch_labels)
batchImage = tile_images(pred_class_gt_mask, batch_size, rows, cols, 1)
im.set_data(np.uint8(batchImage))
print('iter = ', iter, 'max = ', batchImage.max(), 'min = ', batchImage.min(), 'cost = ', cost, 'images/sec = ', images_per_sec)
else:
print('iter = ', iter, 'cost = ', cost, 'images/sec = ', images_per_sec)
iter = iter + 1