-
Notifications
You must be signed in to change notification settings - Fork 2
/
projector.py
229 lines (201 loc) · 7.08 KB
/
projector.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
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
# -*- coding:utf-8 -*-
from __future__ import print_function
import os
import argparse
import json
import numpy as np
def save_tensor_bytes(save_path, x):
y = x.flatten()
n = len(y)
with open(save_path, 'wb') as f:
for i in range(n):
v = y[i]
f.write(v)
def save_label_tsv(save_path, x):
y = x.flatten()
n = len(y)
with open(save_path, 'w') as f:
for i in range(n):
v = y[i]
f.write('%s\n' % (str(v)))
def create_sprite_image(images, image_size=(-1, -1)):
import cv2
"""Returns a sprite image consisting of images passed as argument. Images should be count x width x height"""
# image NHWC
if isinstance(images, list):
images = np.array(images)
if image_size[0] > 0:
img_h = image_size[0]
img_w = image_size[1]
else:
img_h = images.shape[1]
img_w = images.shape[2]
img_c = images.shape[3]
n_plots = int(np.ceil(np.sqrt(images.shape[0])))
spriteimage = np.ones((img_h * n_plots, img_w * n_plots, img_c)) * 255
for i in range(n_plots):
for j in range(n_plots):
this_filter = i * n_plots + j
if this_filter < images.shape[0]:
this_img = images[this_filter]
this_img = cv2.resize(this_img, (img_w, img_h))
this_img = this_img.reshape(img_w, img_h, img_c)
if img_c > 1:
this_img = cv2.cvtColor(this_img, cv2.COLOR_RGB2BGR)
spriteimage[i * img_h:(i + 1) * img_h, j * img_w:(j + 1) * img_w, :] = this_img
return spriteimage, [img_h, img_w]
def write_image_embeddings(root, title, feats, labels, imgs=None, sprite_size=(-1, -1), mode='w+'):
'''
Write embedding data for the `Embedding Project` tool to visualize
:param root: root dir of `Embedding Project` tool
:param title: name of the tensor
:param feats: embedding tensor NxDim
:param labels: labels for each sample NxNumClasses
:param [optional] imgs: images in format NHWC
:param [optional] sprite_size: image sprite size
:param mode: 'w' -- write, 'w+' -- update or append, '+' -- append
:return: None
'''
import cv2
prefix = 'oss_data/' + title
try:
os.makedirs(os.path.join(root, 'oss_data/'))
except:
pass
tensorPath = prefix + '.bytes'
metadataPath = prefix + '.tsv'
if imgs is not None:
imagePath = prefix + '.png'
# save sprites
sprite, singleImageDim = create_sprite_image(imgs, sprite_size)
cv2.imwrite(os.path.join(root,imagePath), sprite)
# save tensor
save_tensor_bytes(os.path.join(root,tensorPath), feats)
# save meta
save_label_tsv(os.path.join(root,metadataPath), labels)
# config
config_path = os.path.join(root,'oss_data/oss_demo_projector_config.json')
config = {"modelCheckpointPath": "Demo datasets", "embeddings": []}
# update or append mode
if mode == 'w+' or mode == '+':
if os.path.exists(config_path):
with open(config_path, 'r') as f:
config = json.load(f)
# new tensor item
item = {
"tensorName": title,
"tensorShape": list(feats.shape),
"tensorPath": tensorPath,
"metadataPath": metadataPath,
}
if imgs is not None:
item["sprite"] = {
"imagePath": imagePath,
"singleImageDim": list(singleImageDim)
}
embeddings = config["embeddings"]
if mode == 'w+':
# check existence
id = -1
for i in range(len(embeddings)):
if embeddings[i]["tensorName"] == title:
id = i
break
if id >= 0:
embeddings[id] = item
else:
config["embeddings"].append(item)
else:
config["embeddings"].append(item)
# write out
with open(config_path, 'w') as f:
json.dump(config, f, sort_keys=True, indent=4)
def save_tensor_tsv(path, feats):
with open(path, 'w') as f:
num, dim = feats.shape
for line in range(num):
for d in range(dim):
if d > 0:
f.write('\t')
v = str(feats[line][d])
f.write(v)
f.write('\n')
def write_tsv_embeddings(prefix, feats, labels=None):
'''
Write a tensor (or meta) to a tsv file for the `Embedding Project` tool
:param prefix: output file prefix
:param feats: embedding tensor NxDim
:param labels: meta data
:return: None
'''
feat_path = prefix + '_data.tsv'
save_tensor_tsv(feat_path, feats)
if labels is None:
return
dims = len(labels.shape)
label_path = prefix + '_meta.tsv'
if dims == 1:
save_label_tsv(label_path, labels)
else:
save_tensor_tsv(label_path, labels)
def _gen_fake_feats(num_class, total_size, feat_dim, var=0.2):
num_each_class = total_size // num_class
# generate centers
centers = np.random.randn(num_class, feat_dim)
feats = np.random.randn(total_size, feat_dim)
labels = []
for i in range(total_size):
cid = i // num_each_class
cid = min(cid, num_class-1)
feats[i] = feats[i] * var + centers[cid]
labels.append(cid)
labels = np.array(labels)
feats = np.float32(feats)
return feats, labels
def demo_test_word(root):
title = 'words'
batch_size = 500
feat_dim = 48
color_names = ['Red', 'Orange', 'Yellow', 'Green', 'Blue', 'Violet', 'Brown', 'Black', 'Grey', 'White']
num_colors = len(color_names)
# Fake metas
feats, labels = _gen_fake_feats(num_colors, batch_size, feat_dim)
metas = [color_names[labels[i]] for i in range(batch_size)]
metas = np.array(metas)
# write tsv files
#write_tsv_embeddings(title, feats, metas)
write_image_embeddings(root, title, feats, metas)
def demo_test_image(root):
import cv2
title = 'images'
num_class = 10
total_size = 500
feat_dim = 48
feats, labels = _gen_fake_feats(num_class, total_size, feat_dim)
# generate fake images
imgs = []
for i in range(total_size):
cid = labels[i]
img = np.ones((40, 40, 1), dtype=np.uint8) * 255
ox = np.random.randint(0, 25)
oy = np.random.randint(20, 35)
cv2.putText(img, str(cid), (ox, oy), 0, 1, (0, 0, 0))
imgs.append(img)
imgs = np.stack(imgs)
write_image_embeddings(root, title, feats, labels, imgs, sprite_size=(24, 24))
if __name__ == '__main__':
parser = parser = argparse.ArgumentParser(description='projector', conflict_handler='resolve')
parser.add_argument('--port', type=int, default=8000, help='server port')
parser.add_argument('--root', default='.', type=str, help='projector root dir')
parser.add_argument('--demo', default=False, type=bool, help='write demo data')
args = parser.parse_args()
if args.demo:
demo_test_image(args.root)
demo_test_word(args.root)
# start server
import sys
import os
if sys.version_info.major == 2:
os.system('python -m SimpleHTTPServer %s' % args.port)
else:
os.system('python -m http.server %s' % args.port)