-
Notifications
You must be signed in to change notification settings - Fork 1.4k
/
image_client.py
executable file
·324 lines (282 loc) · 13.3 KB
/
image_client.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
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
#!/usr/bin/env python
# Copyright (c) 2018-2020, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
import argparse
import numpy as np
import os
from builtins import range
from PIL import Image
from functools import partial
from tensorrtserver.api import *
import tensorrtserver.api.model_config_pb2 as model_config
if sys.version_info >= (3, 0):
import queue
else:
import Queue as queue
class UserData:
def __init__(self):
self._completed_requests = queue.Queue()
# Callback function used for async_run()
def completion_callback(input_filenames, user_data, infer_ctx, request_id):
user_data._completed_requests.put((request_id, input_filenames))
FLAGS = None
def model_dtype_to_np(model_dtype):
if model_dtype == model_config.TYPE_BOOL:
return np.bool
elif model_dtype == model_config.TYPE_INT8:
return np.int8
elif model_dtype == model_config.TYPE_INT16:
return np.int16
elif model_dtype == model_config.TYPE_INT32:
return np.int32
elif model_dtype == model_config.TYPE_INT64:
return np.int64
elif model_dtype == model_config.TYPE_UINT8:
return np.uint8
elif model_dtype == model_config.TYPE_UINT16:
return np.uint16
elif model_dtype == model_config.TYPE_FP16:
return np.float16
elif model_dtype == model_config.TYPE_FP32:
return np.float32
elif model_dtype == model_config.TYPE_FP64:
return np.float64
elif model_dtype == model_config.TYPE_STRING:
return np.dtype(object)
return None
def parse_model(url, protocol, model_name, batch_size, verbose=False):
"""
Check the configuration of a model to make sure it meets the
requirements for an image classification network (as expected by
this client)
"""
ctx = ServerStatusContext(url, protocol, model_name, verbose)
server_status = ctx.get_server_status()
if model_name not in server_status.model_status:
raise Exception("unable to get status for '" + model_name + "'")
status = server_status.model_status[model_name]
config = status.config
if len(config.input) != 1:
raise Exception("expecting 1 input, got {}".format(len(config.input)))
if len(config.output) != 1:
raise Exception("expecting 1 output, got {}".format(len(config.output)))
input = config.input[0]
output = config.output[0]
if output.data_type != model_config.TYPE_FP32:
raise Exception("expecting output datatype to be TYPE_FP32, model '" +
model_name + "' output type is " +
model_config.DataType.Name(output.data_type))
# Output is expected to be a vector. But allow any number of
# dimensions as long as all but 1 is size 1 (e.g. { 10 }, { 1, 10
# }, { 10, 1, 1 } are all ok). Variable-size dimensions are not
# currently supported.
non_one_cnt = 0
for dim in output.dims:
if dim == -1:
raise Exception("variable-size dimension in model output not supported")
if dim > 1:
non_one_cnt += 1
if non_one_cnt > 1:
raise Exception("expecting model output to be a vector")
# Model specifying maximum batch size of 0 indicates that batching
# is not supported and so the input tensors do not expect an "N"
# dimension (and 'batch_size' should be 1 so that only a single
# image instance is inferred at a time).
max_batch_size = config.max_batch_size
if max_batch_size == 0:
if batch_size != 1:
raise Exception("batching not supported for model '" + model_name + "'")
else: # max_batch_size > 0
if batch_size > max_batch_size:
raise Exception("expecting batch size <= {} for model {}".format(max_batch_size, model_name))
# Model input must have 3 dims, either CHW or HWC
if len(input.dims) != 3:
raise Exception(
"expecting input to have 3 dimensions, model '{}' input has {}".format(
model_name, len(input.dims)))
# Variable-size dimensions are not currently supported.
for dim in input.dims:
if dim == -1:
raise Exception("variable-size dimension in model input not supported")
if ((input.format != model_config.ModelInput.FORMAT_NCHW) and
(input.format != model_config.ModelInput.FORMAT_NHWC)):
raise Exception("unexpected input format " + model_config.ModelInput.Format.Name(input.format) +
", expecting " +
model_config.ModelInput.Format.Name(model_config.ModelInput.FORMAT_NCHW) +
" or " +
model_config.ModelInput.Format.Name(model_config.ModelInput.FORMAT_NHWC))
if input.format == model_config.ModelInput.FORMAT_NHWC:
h = input.dims[0]
w = input.dims[1]
c = input.dims[2]
else:
c = input.dims[0]
h = input.dims[1]
w = input.dims[2]
return (input.name, output.name, c, h, w, input.format, model_dtype_to_np(input.data_type))
def preprocess(img, format, dtype, c, h, w, scaling):
"""
Pre-process an image to meet the size, type and format
requirements specified by the parameters.
"""
#np.set_printoptions(threshold='nan')
if c == 1:
sample_img = img.convert('L')
else:
sample_img = img.convert('RGB')
resized_img = sample_img.resize((w, h), Image.BILINEAR)
resized = np.array(resized_img)
if resized.ndim == 2:
resized = resized[:,:,np.newaxis]
typed = resized.astype(dtype)
if scaling == 'INCEPTION':
scaled = (typed / 128) - 1
elif scaling == 'VGG':
if c == 1:
scaled = typed - np.asarray((128,), dtype=dtype)
else:
scaled = typed - np.asarray((123, 117, 104), dtype=dtype)
else:
scaled = typed
# Swap to CHW if necessary
if format == model_config.ModelInput.FORMAT_NCHW:
ordered = np.transpose(scaled, (2, 0, 1))
else:
ordered = scaled
# Channels are in RGB order. Currently model configuration data
# doesn't provide any information as to other channel orderings
# (like BGR) so we just assume RGB.
return ordered
def postprocess(results, filenames, batch_size):
"""
Post-process results to show classifications.
"""
if len(results) != 1:
raise Exception("expected 1 result, got {}".format(len(results)))
batched_result = list(results.values())[0]
if len(batched_result) != batch_size:
raise Exception("expected {} results, got {}".format(batch_size, len(batched_result)))
if len(filenames) != batch_size:
raise Exception("expected {} filenames, got {}".format(batch_size, len(filenames)))
for (index, result) in enumerate(batched_result):
print("Image '{}':".format(filenames[index]))
for cls in result:
print(" {} ({}) = {}".format(cls[0], cls[2], cls[1]))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('-v', '--verbose', action="store_true", required=False, default=False,
help='Enable verbose output')
parser.add_argument('-a', '--async', dest="async_set", action="store_true", required=False,
default=False, help='Use asynchronous inference API')
parser.add_argument('--streaming', action="store_true", required=False, default=False,
help='Use streaming inference API. ' +
'The flag is only available with gRPC protocol.')
parser.add_argument('-m', '--model-name', type=str, required=True,
help='Name of model')
parser.add_argument('-x', '--model-version', type=int, required=False,
help='Version of model. Default is to use latest version.')
parser.add_argument('-b', '--batch-size', type=int, required=False, default=1,
help='Batch size. Default is 1.')
parser.add_argument('-c', '--classes', type=int, required=False, default=1,
help='Number of class results to report. Default is 1.')
parser.add_argument('-s', '--scaling', type=str, choices=['NONE', 'INCEPTION', 'VGG'],
required=False, default='NONE',
help='Type of scaling to apply to image pixels. Default is NONE.')
parser.add_argument('-u', '--url', type=str, required=False, default='localhost:8000',
help='Inference server URL. Default is localhost:8000.')
parser.add_argument('-i', '--protocol', type=str, required=False, default='HTTP',
help='Protocol (HTTP/gRPC) used to ' +
'communicate with inference service. Default is HTTP.')
parser.add_argument('image_filename', type=str, nargs='?', default=None,
help='Input image / Input folder.')
FLAGS = parser.parse_args()
protocol = ProtocolType.from_str(FLAGS.protocol)
if FLAGS.streaming and protocol != ProtocolType.GRPC:
raise Exception("Streaming is only allowed with gRPC protocol")
# Make sure the model matches our requirements, and get some
# properties of the model that we need for preprocessing
input_name, output_name, c, h, w, format, dtype = parse_model(
FLAGS.url, protocol, FLAGS.model_name,
FLAGS.batch_size, FLAGS.verbose)
ctx = InferContext(FLAGS.url, protocol, FLAGS.model_name,
FLAGS.model_version, FLAGS.verbose, 0, FLAGS.streaming)
filenames = []
if os.path.isdir(FLAGS.image_filename):
filenames = [os.path.join(FLAGS.image_filename, f)
for f in os.listdir(FLAGS.image_filename)
if os.path.isfile(os.path.join(FLAGS.image_filename, f))]
else:
filenames = [FLAGS.image_filename,]
filenames.sort()
# Preprocess the images into input data according to model
# requirements
image_data = []
for filename in filenames:
img = Image.open(filename)
image_data.append(preprocess(img, format, dtype, c, h, w, FLAGS.scaling))
# Send requests of FLAGS.batch_size images. If the number of
# images isn't an exact multiple of FLAGS.batch_size then just
# start over with the first images until the batch is filled.
results = []
result_filenames = []
request_ids = []
image_idx = 0
last_request = False
user_data = UserData()
sent_count = 0
while not last_request:
input_filenames = []
input_batch = []
for idx in range(FLAGS.batch_size):
input_filenames.append(filenames[image_idx])
input_batch.append(image_data[image_idx])
image_idx = (image_idx + 1) % len(image_data)
if image_idx == 0:
last_request = True
# Send request
if not FLAGS.async_set:
results.append(ctx.run(
{ input_name : input_batch },
{ output_name : (InferContext.ResultFormat.CLASS, FLAGS.classes) },
FLAGS.batch_size))
result_filenames.append(input_filenames)
else:
ctx.async_run(partial(completion_callback, input_filenames, user_data),
{ input_name :input_batch },
{ output_name : (InferContext.ResultFormat.CLASS, FLAGS.classes) },
FLAGS.batch_size)
sent_count += 1
# For async, retrieve results according to the send order
if FLAGS.async_set:
processed_count = 0
while processed_count < sent_count:
(request_id, input_filenames) = user_data._completed_requests.get()
results.append(ctx.get_async_run_results(request_id))
result_filenames.append(input_filenames)
processed_count += 1
for idx in range(len(results)):
print("Request {}, batch size {}".format(idx, FLAGS.batch_size))
postprocess(results[idx], result_filenames[idx], FLAGS.batch_size)