forked from danielgordon10/re3-tensorflow
-
Notifications
You must be signed in to change notification settings - Fork 1
/
batch_cache.py
250 lines (219 loc) · 8.85 KB
/
batch_cache.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
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
import argparse
import cv2
import numpy as np
import random
import struct
import sys
import threading
import time
import get_datasets
try:
import cPickle as pickle
import SocketServer
except ImportError:
# Python 3 compatibility
import pickle
import socketserver as SocketServer
HOST = 'localhost'
class BatchCacheHandler(SocketServer.BaseRequestHandler, object):
def handle(self):
# self.request is the TCP socket connected to the client
# self.server is used for communication with BatchCacheServer
# Serves until disconnect.
while not self.server.shut_down:
alive = self.request.recv(1024).strip()
print(alive)
if len(alive) == 0:
break
(key, val) = self.server.get_sample(self.server.batch_cache)
print('key', key)
self.server.lock.acquire()
try:
# Send the key.
keyPickle = pickle.dumps(key)
messageLength = struct.pack('>I', len(keyPickle))
self.request.sendall(messageLength)
self.request.sendall(keyPickle)
for image in val:
# Send the actual image.
messageLength = struct.pack('>I',len(image[0]))
self.request.sendall(messageLength)
self.request.sendall(image[0])
# Send the shape.
messageLength = struct.pack('>I', len(image[1]))
self.request.sendall(messageLength)
self.request.sendall(image[1])
finally:
self.server.lock.release()
class BatchCacheServer:
def __init__(self, args):
# Set constants
self.max_size = args.max_size
self.num_unrolls = args.num_unrolls
self.debug = args.debug
self.vals = []
self.keys = []
# key -> index in vals array
self.idxs = dict()
self.data_hits = None
self.data_lock = threading.Lock()
# Start the queue monitor.
self.keep_alive = True
self.worker = threading.Thread(target=self.__memory_monitor, args=())
#self.worker.daemon = True
self.worker.start()
# Flag for the handler to shut down
self.shut_down = False
self.image_paths = []
# dataset_id, vid_id, track_id, frame_id
self.all_keys = set()
self.create_keys()
#Load the first few samples.
for i in range(min(int(self.max_size / 2), 32)):
self.__random_load(force_append=True)
def __del__(self):
self.keep_alive = False
self.worker.join()
def __memory_monitor(self):
while self.keep_alive:
if self.data_hits is not None and np.sum(self.data_hits) > 0:
self.__random_load()
time.sleep(0.0001)
def __random_load(self, force_append=False):
# First find some data that hasn't been loaded already.
try:
key = random.sample(self.all_keys, 1)[0]
while key in self.idxs:
key = random.sample(self.all_keys, 1)[0]
val = self.lookup_func(key)
# Next check to see if we should append or replace existing data.
self.data_lock.acquire()
if len(self.keys) < self.max_size or force_append: # Append
if self.debug:
print('Appending new data. Num keys =', len(self.keys))
self.vals.append(val)
self.keys.append(key)
if self.data_hits is None:
self.data_hits = np.zeros(1)
else:
self.data_hits = np.append(self.data_hits, 0)
self.idxs[key] = len(self.vals) - 1
else: # Replace
if np.sum(self.data_hits) == 0:
sys.stderr.writ/lookupe(
('Something went horribly wrong. __random_load was '
'called and the cache is full, but none of the '
'elements have been hit!'))
sys.stderr.flush()
self.data_lock.release()
return
total_hits = np.sum(self.data_hits)
i = np.argmax(self.data_hits)
if self.debug:
print('Replacing data. Replacing spot', i)
del self.idxs[self.keys[i]]
self.vals[i] = val
self.keys[i] = key
self.data_hits[i] = 0
self.idxs[key] = i
if self.debug:
print('Total used elements:', len(self.data_hits[self.data_hits > 0]))
print(self.data_hits[self.data_hits > 0])
self.data_lock.release()
except Exception as ex:
import traceback
trace = traceback.format_exc()
print(trace)
self.shut_down = True
self.data_lock.release()
errorFile = open('error.txt', 'a+')
errorFile.write('exception in __random_load %s\n' % str(ex))
errorFile.write(str(trace))
def add_dataset(self, dataset_name):
dataset_ind = len(self.image_paths)
data = get_datasets.get_data_for_dataset(dataset_name, 'train')
gt = data['gt']
num_keys = 0
for xx in range(gt.shape[0] - self.num_unrolls):
start_line = gt[xx,:].astype(int)
end_line = gt[xx + self.num_unrolls,:].astype(int)
# Check that still in the same sequence.
# Video_id should match, track_id should match, and image number should be exactly num_unrolls frames later.
if (start_line[4] == end_line[4] and
start_line[5] == end_line[5] and
start_line[6] + self.num_unrolls == end_line[6]):
# Add the key.
self.all_keys.add((dataset_ind, start_line[4], start_line[5], start_line[6]))
num_keys += 1
if self.debug:
print('#%s keys: %d' % (dataset_name, num_keys))
image_paths = data['image_paths']
# Add the array to image_paths. Note that image paths is indexed by the dataset number THEN by the image line.
self.image_paths.append(image_paths)
def create_keys(self):
self.add_dataset('imagenet_video')
time.sleep(1)
def lookup_func(self, key):
images = None
try:
images = []
ind = key[-1]
if self.debug:
imageName = self.image_paths[key[0]][ind]
print('Reading image', imageName)
for dd in range(self.num_unrolls):
path = self.image_paths[key[0]][ind + dd]
image = cv2.imread(path)[:,:,::-1]
shape = pickle.dumps(image.shape)
string = image.tostring()
images.append((string, shape))
except Exception as ex:
import traceback
trace = traceback.format_exc()
print(trace)
errorFile = open('error.txt', 'a+')
errorFile.write('exception in lookup_func %s\n' % str(ex))
errorFile.write(str(trace))
finally:
return images
def get_sample(self, batch_cache):
try:
batch_cache.data_lock.acquire()
idx = random.randint(0, len(batch_cache.vals) - 1)
key = batch_cache.keys[idx]
val = batch_cache.vals[idx]
batch_cache.data_hits[idx] += 1
except Exception as ex:
import traceback
trace = traceback.format_exc()
print(trace)
errorFile = open('error.txt', 'a+')
errorFile.write('exception in lookup_func %s\n' % str(ex))
errorFile.write(str(trace))
finally:
batch_cache.data_lock.release()
return (key, val)
def serve(self, port):
if self.debug:
print('Server starting up')
handler = SocketServer.TCPServer((HOST, port), BatchCacheHandler)
handler.get_sample = self.get_sample
handler.batch_cache = self
handler.lock = self.data_lock
handler.shut_down = self.shut_down
handler.serve_forever()
def main(args):
server = BatchCacheServer(args)
server.serve(args.port)
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Server for network images.')
parser.add_argument('-s', '--max_size', action='store', default=100,
dest='max_size', type=int)
parser.add_argument('-n', '--num_unrolls', action='store', default=2,
dest='num_unrolls', type=int)
parser.add_argument('-p', '--port', action='store', default=9997,
dest='port', type=int)
parser.add_argument('-d', '--debug', action='store_true', default=False,
dest='debug')
args = parser.parse_args()
main(args)