-
Notifications
You must be signed in to change notification settings - Fork 8
/
infer.py
175 lines (144 loc) · 5.19 KB
/
infer.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
175
import numpy as np
#np.random.seed(777)
import chainer
from chainer import cuda
from chainer import serializers
import chainer.functions as F
import argparse
import matplotlib.pyplot as plt
plt.style.use('ggplot')
import PIL
from PIL import ImageDraw
parser = argparse.ArgumentParser()
group = parser.add_mutually_exclusive_group()
group.add_argument('--original', action='store_true',
default=True, help='train on original MNIST')
group.add_argument('--translated', action='store_true',
default=False, help='train on translated MNIST')
group.add_argument('--cluttered', action='store_true',
default=False, help='train on translated & cluttered MNIST')
parser.add_argument('--lstm', type=bool, default=False,
help='use LSTM units in core layer')
parser.add_argument('-m', '--model', type=str,
default='models/ram_original_epoch800.chainermodel',
help='load model weights from given file')
parser.add_argument('-g', '--gpuid', type=int, default=-1,
help='GPU device ID (default is CPU)')
args = parser.parse_args()
train, test = chainer.datasets.get_mnist()
train_data, train_targets = np.array(train).transpose()
test_data, test_targets = np.array(test).transpose()
train_data = np.array(list(train_data)).reshape(train_data.shape[0],1,28,28)
train_data.flags.writeable = False
test_data = np.array(list(test_data)).reshape(test_data.shape[0],1,28,28)
train_targets = np.array(train_targets).astype(np.int32)
test_targets = np.array(test_targets).astype(np.int32)
# hyper-params for each task
if args.original:
filename = 'ram_original'
# RAM params for original MNIST
g_size = 8
n_steps = 6
n_scales = 1
def process(batch):
return batch
if args.translated:
filename = 'ram_translated'
g_size = 12
n_steps = 6
n_scales = 3
# create translated MNIST
def translate(batch):
n, c, w_i = batch.shape[:3]
w_o = 60
data = np.zeros(shape=(n,c,w_o,w_o), dtype=np.float32)
for k in range(n):
i, j = np.random.randint(0, w_o-w_i, size=2)
data[k, :, i:i+w_i, j:j+w_i] += batch[k]
return data
process = translate
if args.cluttered:
filename = 'ram_cluttered'
g_size = 12
n_steps = 6
n_scales = 3
# create cluttered MNIST
def clutter(batch):
n, c, w_i = batch.shape[:3]
w_o = 60
data = np.zeros(shape=(n,c,w_o,w_o), dtype=np.float32)
for k in range(n):
i, j = np.random.randint(0, w_o-w_i, size=2)
data[k, :, i:i+w_i, j:j+w_i] += batch[k]
for _ in range(4):
clt = train_data[np.random.randint(0, train_data.shape[0]-1)]
c1, c2 = np.random.randint(0, w_i-8, size=2)
i1, i2 = np.random.randint(0, w_o-8, size=2)
data[k, :, i1:i1+8, i2:i2+8] += clt[:, c1:c1+8, c2:c2+8]
data = np.clip(data, 0., 1.)
return data
process = clutter
# init RAM model
from ram import RAM
model = RAM(
g_size=g_size, n_steps=n_steps, n_scales=n_scales, use_lstm=args.lstm)
print('load model from {}'.format(args.model))
serializers.load_hdf5(args.model, model)
gpuid = args.gpuid
if gpuid >= 0:
cuda.get_device(gpuid).use()
model.to_gpu()
# inference
test_data = process(test_data)
test_data.flags.writeable = False
index = np.random.randint(0, 9999)
image = PIL.Image.fromarray(test_data[index][0]*255).convert('RGB')
x = test_data[index][np.newaxis,:,:,:]
init_l = np.random.uniform(low=-1, high=1, size=2)
y, ys, ls = model.infer(x, init_l)
locs = ((ls+1) / 2) * (np.array(test_data.shape[2:4])+1)
# plot results
from crop import crop
plt.subplots_adjust(wspace=0.35, hspace=0.05)
for t in range(0, n_steps):
# digit with glimpse
plt.subplot(3+n_scales, n_steps, t+1)
# green if correct otherwise red
if np.argmax(ys[t]) == test_targets[index]:
color = (0, 255, 0)
else:
color = (255, 0, 0)
canvas = image.copy()
draw = ImageDraw.Draw(canvas)
xy = np.array([locs[t,1],locs[t,0],locs[t,1],locs[t,0]])
wh = np.array([-g_size//2, -g_size//2, g_size//2, g_size//2])
xys = [xy + np.power(2,s)*wh for s in range(n_scales)]
for xy in xys:
draw.rectangle(xy=list(xy), outline=color)
del draw
plt.imshow(canvas)
plt.axis('off')
# glimpse at each scale
gs = crop(x, center=ls[t:t+1], size=g_size)
plt.subplot(3+n_scales, n_steps, n_steps + t+1)
plt.imshow(gs.data[0,0], cmap='gray')
plt.axis('off')
for k in range(1, n_scales):
s = np.power(2,k)
patch = crop(x, center=ls[t:t+1], size=g_size*s)
patch = F.average_pooling_2d(patch, ksize=s)
gs = F.concat((gs, patch), axis=1)
plt.subplot(3+n_scales, n_steps, n_steps*(k+1) + t+1)
plt.imshow(gs.data[0,k], cmap='gray')
plt.axis('off')
# output probability
plt.subplot2grid((3+n_scales,n_steps), (1+n_scales,t), rowspan=2)
plt.barh(np.arange(10), ys[t], align='center')
plt.xlim(0, 1)
plt.ylim(-0.5, 9.5)
if t == 0:
plt.yticks(np.arange(10))
else:
plt.yticks(np.arange(10), ['' for _ in range(10)])
plt.xticks([])
plt.show()