-
Notifications
You must be signed in to change notification settings - Fork 337
/
Copy pathblip2.py
401 lines (338 loc) · 13.8 KB
/
blip2.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
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
import sys
import time
from typing import List
from logging import getLogger
import numpy as np
import cv2
from PIL import Image
# import original modules
sys.path.append('../../util')
from arg_utils import get_base_parser, update_parser # noqa
from model_utils import check_and_download_models, check_and_download_file # noqa
from detector_utils import load_image # noqa
import ailia
logger = getLogger(__name__)
# ======================
# Parameters
# ======================
WEIGHT_PATH = 'blip2-opt-2.7b.onnx'
WEIGHT_PB_PATH = 'blip2-opt-2.7b_weights.pb'
MODEL_PATH = 'blip2-opt-2.7b.onnx.prototxt'
WEIGHT_VIS_PATH = 'vision_model.onnx'
WEIGHT_VIS_PB_PATH = 'vision_model_weights.pb'
MODEL_VIS_PATH = 'vision_model.onnx.prototxt'
REMOTE_PATH = 'https://storage.googleapis.com/ailia-models/blip2/'
IMAGE_PATH = 'merlion.png'
IMG_SIZE = 224
# ======================
# Arguemnt Parser Config
# ======================
parser = get_base_parser(
'BLIP-2', IMAGE_PATH, None, fp16_support=False
)
parser.add_argument(
'--onnx',
action='store_true',
help='execute onnxruntime version.'
)
parser.add_argument(
'--disable_ailia_tokenizer',
action='store_true',
help='disable ailia tokenizer.'
)
args = update_parser(parser, check_input_type=False)
# ======================
# Main functions
# ======================
def preprocess(img):
im_h, im_w, _ = img.shape
h = w = IMG_SIZE
img = np.array(Image.fromarray(img).resize((w, h), Image.Resampling.BICUBIC))
img = img / 255
image_mean = (0.48145466, 0.4578275, 0.40821073)
image_std = (0.26862954, 0.26130258, 0.27577711)
img = (img - image_mean) / image_std
img = img.transpose(2, 0, 1) # HWC -> CHW
img = np.expand_dims(img, axis=0)
img = img.astype(np.float16)
return img
def decode(
net,
inputs_embeds: np.ndarray,
input_ids: np.ndarray,
attention_mask: np.ndarray,
past_key_values: List[np.ndarray]):
if not args.onnx:
decoder_output = net.predict([
attention_mask,
input_ids,
inputs_embeds,
past_key_values[0],
past_key_values[1],
past_key_values[2],
past_key_values[3],
past_key_values[4],
past_key_values[5],
past_key_values[6],
past_key_values[7],
past_key_values[8],
past_key_values[9],
past_key_values[10],
past_key_values[11],
past_key_values[12],
past_key_values[13],
past_key_values[14],
past_key_values[15],
past_key_values[16],
past_key_values[17],
past_key_values[18],
past_key_values[19],
past_key_values[20],
past_key_values[21],
past_key_values[22],
past_key_values[23],
past_key_values[24],
past_key_values[25],
past_key_values[26],
past_key_values[27],
past_key_values[28],
past_key_values[29],
past_key_values[30],
past_key_values[31],
past_key_values[32],
past_key_values[33],
past_key_values[34],
past_key_values[35],
past_key_values[36],
past_key_values[37],
past_key_values[38],
past_key_values[39],
past_key_values[40],
past_key_values[41],
past_key_values[42],
past_key_values[43],
past_key_values[44],
past_key_values[45],
past_key_values[46],
past_key_values[47],
past_key_values[48],
past_key_values[49],
past_key_values[50],
past_key_values[51],
past_key_values[52],
past_key_values[53],
past_key_values[54],
past_key_values[55],
past_key_values[56],
past_key_values[57],
past_key_values[58],
past_key_values[59],
past_key_values[60],
past_key_values[61],
past_key_values[62],
past_key_values[63],
])
else:
decoder_output = net.run(
None, {
'attention_mask': attention_mask,
'input_ids': input_ids,
'inputs_embeds': inputs_embeds,
'past_key_values_0_key': past_key_values[0],
'past_key_values_0_value': past_key_values[1],
'past_key_values_1_key': past_key_values[2],
'past_key_values_1_value': past_key_values[3],
'past_key_values_2_key': past_key_values[4],
'past_key_values_2_value': past_key_values[5],
'past_key_values_3_key': past_key_values[6],
'past_key_values_3_value': past_key_values[7],
'past_key_values_4_key': past_key_values[8],
'past_key_values_4_value': past_key_values[9],
'past_key_values_5_key': past_key_values[10],
'past_key_values_5_value': past_key_values[11],
'past_key_values_6_key': past_key_values[12],
'past_key_values_6_value': past_key_values[13],
'past_key_values_7_key': past_key_values[14],
'past_key_values_7_value': past_key_values[15],
'past_key_values_8_key': past_key_values[16],
'past_key_values_8_value': past_key_values[17],
'past_key_values_9_key': past_key_values[18],
'past_key_values_9_value': past_key_values[19],
'past_key_values_10_key': past_key_values[20],
'past_key_values_10_value': past_key_values[21],
'past_key_values_11_key': past_key_values[22],
'past_key_values_11_value': past_key_values[23],
'past_key_values_12_key': past_key_values[24],
'past_key_values_12_value': past_key_values[25],
'past_key_values_13_key': past_key_values[26],
'past_key_values_13_value': past_key_values[27],
'past_key_values_14_key': past_key_values[28],
'past_key_values_14_value': past_key_values[29],
'past_key_values_15_key': past_key_values[30],
'past_key_values_15_value': past_key_values[31],
'past_key_values_16_key': past_key_values[32],
'past_key_values_16_value': past_key_values[33],
'past_key_values_17_key': past_key_values[34],
'past_key_values_17_value': past_key_values[35],
'past_key_values_18_key': past_key_values[36],
'past_key_values_18_value': past_key_values[37],
'past_key_values_19_key': past_key_values[38],
'past_key_values_19_value': past_key_values[39],
'past_key_values_20_key': past_key_values[40],
'past_key_values_20_value': past_key_values[41],
'past_key_values_21_key': past_key_values[42],
'past_key_values_21_value': past_key_values[43],
'past_key_values_22_key': past_key_values[44],
'past_key_values_22_value': past_key_values[45],
'past_key_values_23_key': past_key_values[46],
'past_key_values_23_value': past_key_values[47],
'past_key_values_24_key': past_key_values[48],
'past_key_values_24_value': past_key_values[49],
'past_key_values_25_key': past_key_values[50],
'past_key_values_25_value': past_key_values[51],
'past_key_values_26_key': past_key_values[52],
'past_key_values_26_value': past_key_values[53],
'past_key_values_27_key': past_key_values[54],
'past_key_values_27_value': past_key_values[55],
'past_key_values_28_key': past_key_values[56],
'past_key_values_28_value': past_key_values[57],
'past_key_values_29_key': past_key_values[58],
'past_key_values_29_value': past_key_values[59],
'past_key_values_30_key': past_key_values[60],
'past_key_values_30_value': past_key_values[61],
'past_key_values_31_key': past_key_values[62],
'past_key_values_31_value': past_key_values[63],
}
)
logits, new_past_key_values = decoder_output[0], decoder_output[1:]
return logits, new_past_key_values
def stopping_criteria(
input_ids: np.array) -> bool:
max_length = 21
cur_len = input_ids.shape[-1]
is_done = cur_len >= max_length
return is_done
def greedy_search(net, inputs_embeds):
bos_token_id = 2
eos_token_id = np.array([50118])
shape = inputs_embeds.shape[:2]
batch_size = shape[0]
input_ids = np.ones((batch_size, 1), dtype=int) * bos_token_id
attention_mask = np.ones(shape, dtype=int)
past_key_values = [np.zeros((batch_size, shape[1] - 1, 0, 80), dtype=np.float16)] * 64
# keep track of which sequences are already finished
unfinished_sequences = np.ones(input_ids.shape[0], dtype=int)
this_peer_finished = False # used by synced_gpus only
while True:
logits, past_key_values = decode(
net, inputs_embeds, input_ids[:, 1:][:, -1:], attention_mask, past_key_values
)
next_tokens_scores = logits[:, -1, :]
# argmax
next_tokens = np.argmax(next_tokens_scores, axis=-1)
# update generated ids, model inputs, and length for next step
input_ids = np.concatenate([input_ids, next_tokens[:, None]], axis=-1)
attention_mask = np.concatenate(
[attention_mask, np.ones((attention_mask.shape[0], 1), dtype=int)],
axis=-1
)
inputs_embeds = inputs_embeds[:, :0, :]
# if eos_token was found in one sentence, set sentence to finished
unfinished_sequences = unfinished_sequences * np.prod(
np.tile(next_tokens, (eos_token_id.shape[0], 1)) != eos_token_id[:, None],
axis=0
)
# stop when each sentence is finished
if np.max(unfinished_sequences) == 0:
this_peer_finished = True
# stop if we exceed the maximum length
if stopping_criteria(input_ids):
this_peer_finished = True
if this_peer_finished:
break
return input_ids
def predict(models, img):
img = img[:, :, ::-1] # BGR -> RGB
img = preprocess(img)
net = models['vis']
if not args.onnx:
output = net.predict([img])
else:
output = net.run(None, {'pixel_values': img})
embeds = output[0]
net = models['net']
generated_ids = greedy_search(net, embeds)
tokenizer = models['tokenizer']
generated_text = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
generated_text = generated_text[0].strip()
return generated_text
def recognize_from_image(models):
# input image loop
for image_path in args.input:
logger.info(image_path)
# prepare input data
img = load_image(image_path)
img = cv2.cvtColor(img, cv2.COLOR_BGRA2BGR)
# inference
logger.info('Start inference...')
if args.benchmark:
logger.info('BENCHMARK mode')
total_time_estimation = 0
for i in range(args.benchmark_count):
start = int(round(time.time() * 1000))
generated_text = predict(models, img)
end = int(round(time.time() * 1000))
estimation_time = (end - start)
# Logging
logger.info(f'\tailia processing estimation time {estimation_time} ms')
if i != 0:
total_time_estimation = total_time_estimation + estimation_time
logger.info(f'\taverage time estimation {total_time_estimation / (args.benchmark_count - 1)} ms')
else:
generated_text = predict(models, img)
logger.info('### Caption ### ')
logger.info(generated_text)
logger.info('Script finished successfully.')
def main():
# model files check and download
check_and_download_models(WEIGHT_PATH, MODEL_PATH, REMOTE_PATH)
check_and_download_models(WEIGHT_VIS_PATH, MODEL_VIS_PATH, REMOTE_PATH)
check_and_download_file(WEIGHT_PB_PATH, REMOTE_PATH)
check_and_download_file(WEIGHT_VIS_PB_PATH, REMOTE_PATH)
env_id = args.env_id
# initialize
if not args.onnx:
memory_mode = ailia.get_memory_mode(
reduce_constant=True, ignore_input_with_initializer=True,
reduce_interstage=False, reuse_interstage=True)
net = ailia.Net(MODEL_PATH, WEIGHT_PATH, env_id=env_id, memory_mode=memory_mode)
vis_net = ailia.Net(MODEL_VIS_PATH, WEIGHT_VIS_PATH, env_id=env_id, memory_mode=memory_mode)
if args.profile:
net.set_profile_mode(True)
vis_net.set_profile_mode(True)
else:
import onnxruntime
cuda = 0 < ailia.get_gpu_environment_id()
providers = ['CUDAExecutionProvider', 'CPUExecutionProvider'] if cuda else ['CPUExecutionProvider']
net = onnxruntime.InferenceSession(WEIGHT_PATH, providers=providers)
vis_net = onnxruntime.InferenceSession(WEIGHT_VIS_PATH, providers=providers)
if args.disable_ailia_tokenizer:
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("tokenizer")
else:
from ailia_tokenizer import GPT2Tokenizer
tokenizer = GPT2Tokenizer.from_pretrained("./tokenizer/")
models = {
'net': net,
'vis': vis_net,
'tokenizer': tokenizer,
}
recognize_from_image(models)
if args.profile and not args.onnx:
print("--- profile net")
print(net.get_summary())
print("")
print("--- profile vis_net")
print(vis_net.get_summary())
if __name__ == '__main__':
main()