-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathevaluate.py
1653 lines (1433 loc) · 64.2 KB
/
evaluate.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
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements. See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership. The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
import os
import numpy as np
import tvm
from tvm import relay
from tvm.relay import testing
from tvm import autotvm
from tvm.contrib import utils, ndk
from tvm.topi import testing
from common import convert_to_dtype, advanced_time_evaluator
class ModelImporter(object):
def available_models(self):
import inspect
models = []
for method in inspect.getmembers(type(self)):
if "import_" in method[0]:
models.append(method[0].split("import_")[1])
return models
def __call__(self, model, *args, **kwargs):
import inspect
for method in inspect.getmembers(type(self)):
if "import_" + model == method[0]:
return method[1](self, *args, **kwargs)
raise ValueError("import_" + model + " not found.")
def get_onnx_from_tf1(self, model_url, filename, input_names, output_names, shape_override = None):
tf_model_file = os.path.abspath(
os.path.dirname(os.path.realpath(__file__))
+ "/models/{}.pb".format(filename)
)
from tvm.contrib import download
download.download(model_url, tf_model_file)
# converted using command line:
# python -m tf2onnx.convert --graphdef mace_resnet-v2-50.pb --output mace_resnet-v2-50.onnx --inputs input:0[1,224,224,3] --outputs resnet_v2_50/predictions/Reshape_1:0
onnx_model_file = os.path.abspath(
os.path.dirname(os.path.realpath(__file__))
+ "/models/{}.onnx".format(filename))
if os.path.exists(onnx_model_file) == False:
import tf2onnx
import tensorflow as tf
try:
tf_compat_v1 = tf.compat.v1
except ImportError:
tf_compat_v1 = tf
# Tensorflow utility functions
import tvm.relay.testing.tf as tf_testing
with tf_compat_v1.gfile.GFile(tf_model_file, "rb") as f:
graph_def = tf_compat_v1.GraphDef()
graph_def.ParseFromString(f.read())
#graph = tf.import_graph_def(graph_def, name="")
# Call the utility to import the graph definition into default graph.
graph_def = tf_testing.ProcessGraphDefParam(graph_def)
model_proto, external_tensor_storage = tf2onnx.convert.from_graph_def(graph_def,
name=filename, input_names=input_names, output_names=output_names,
shape_override = shape_override,
output_path=onnx_model_file)
return onnx_model_file
def get_graphdef_from_tf1(self, model_url, filename):
graph_def = None
tf_model_file = os.path.abspath(
os.path.dirname(os.path.realpath(__file__))
+ "/models/{}.pb".format(filename)
)
from tvm.contrib import download
download.download(model_url, tf_model_file)
# converted using command line:
# python -m tf2onnx.convert --graphdef mace_resnet-v2-50.pb --output mace_resnet-v2-50.onnx --inputs input:0[1,224,224,3] --outputs resnet_v2_50/predictions/Reshape_1:0
onnx_model_file = os.path.abspath(
os.path.dirname(os.path.realpath(__file__))
+ "/../models/{}.onnx".format(filename))
import tensorflow as tf
try:
tf_compat_v1 = tf.compat.v1
except ImportError:
tf_compat_v1 = tf
# Tensorflow utility functions
import tvm.relay.testing.tf as tf_testing
with tf_compat_v1.gfile.GFile(tf_model_file, "rb") as f:
graph_def = tf_compat_v1.GraphDef()
graph_def.ParseFromString(f.read())
graph_def = tf_testing.ProcessGraphDefParam(graph_def)
return graph_def
def import_mace_mobilenetv1_nhwc(self, target="llvm", dtype="float32"):
model_url = "https://cnbj1.fds.api.xiaomi.com/mace/miai-models/mobilenet-v1/mobilenet-v1-1.0.pb"
filename = "mace_mobilenet-v1-1.0"
graph_def = self.get_graphdef_from_tf1(model_url, filename)
shape_dict = {"input": (1, 224, 224, 3)}
mod, params = relay.frontend.from_tensorflow(graph_def, shape=shape_dict,
outputs=["MobilenetV1/Predictions/Reshape_1"])
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
dtype = "float32" if dtype == "float32" else "float16"
return (mod, params, shape_dict, dtype, target, ImageNetValidator(shape_dict, "NHWC", preproc="keras_mobilenetv1"))
def import_mace_mobilenetv1_nchw(self, target="llvm", dtype="float32"):
model_url = "https://cnbj1.fds.api.xiaomi.com/mace/miai-models/mobilenet-v1/mobilenet-v1-1.0.pb"
filename = "mace_mobilenet-v1-1.0"
input_names = ["input:0"]
output_names = ["MobilenetV1/Predictions/Reshape_1:0"]
onnx_model_file = self.get_onnx_from_tf1(model_url, filename, input_names, output_names)
import onnx
model = onnx.load(onnx_model_file)
shape_dict = {'input:0': [1, 224, 224, 3]}
mod, params = relay.frontend.from_onnx(model, shape_dict, freeze_params=True)
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
dtype = "float32" if dtype == "float32" else "float16"
return (mod, params, shape_dict, dtype, target, ImageNetValidator(shape_dict, "NHWC", preproc="keras_mobilenetv1"))
def import_conv2d_deeplabv3(self, target="llvm", dtype="float32"):
dtype_init="float32"
input_shape = (1, 513, 513, 3)
filter_shape = (3, 3, 3, 32)
bias_shape = (1, 1, 1, 32)
A = relay.var("data", shape=input_shape, dtype=dtype_init)
B = relay.var("weight", shape=filter_shape, dtype=dtype_init)
bias = relay.var("bias", shape=bias_shape, dtype=dtype_init)
#C = relay.nn.relu(A)
conv = relay.nn.conv2d(A, B, data_layout="NHWC", kernel_layout="HWIO",
padding=[1,1,1,1],strides=[2,2],
out_dtype=dtype_init, channels=32, kernel_size=(3,3))
D = relay.op.add(conv, bias)
D = relay.op.nn.relu(D)
mod = relay.Function([A, B, bias], D)
np.random.seed(0)
initializer = relay.testing.init.Xavier()
filter_data = np.zeros(filter_shape).astype(dtype_init)
bias_data = np.zeros(bias_shape).astype(dtype_init)
initializer("weight", filter_data)
initializer("bias", bias_data)
params = {
"weight": tvm.nd.array(filter_data),
"bias" : tvm.nd.array(bias_data),
}
# downcast to float16
mod = convert_to_dtype(mod, dtype)
dtype = "float32" if dtype == "float32" else "float16"
return (mod, params, {"data": input_shape}, dtype, target)
def import_mace_resnet50_v2(self, target="llvm", dtype="float32"):
model_url = "https://cnbj1.fds.api.xiaomi.com/mace/miai-models/resnet-v2-50/resnet-v2-50.pb"
filename = "mace_resnet-v2-50"
input_names = ["input:0"]
shape_override = {"input:0": [1, 299, 299, 3]}
output_names = ["resnet_v2_50/predictions/Reshape_1:0"]
onnx_model_file = self.get_onnx_from_tf1(model_url, filename, input_names, output_names, shape_override)
import onnx
model = onnx.load(onnx_model_file)
mod, params = relay.frontend.from_onnx(model, shape_override, freeze_params=True)
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
dtype = "float32" if dtype == "float32" else "float16"
return (mod, params, shape_override, dtype, target, \
ImageNetValidator(shape_override, "NHWC", preproc="keras"))
def import_ac_resnet50_tf(self, target="llvm", dtype="float32"):
model_url = "https://download.01.org/opencv/public_models/012020/resnet-50-tf/resnet_v1-50.pb"
filename = "resnet_v1-50"
input_names = ["map/TensorArrayStack/TensorArrayGatherV3:0"]
shape_override = {"map/TensorArrayStack/TensorArrayGatherV3:0": [1, 224, 224, 3]}
output_names = ["softmax_tensor:0"]
onnx_model_file = self.get_onnx_from_tf1(model_url, filename, input_names, output_names, shape_override)
import onnx
model = onnx.load(onnx_model_file)
shape_dict = {'map/TensorArrayStack/TensorArrayGatherV3:0': [1, 224, 224, 3]}
mod, params = relay.frontend.from_onnx(model, shape_dict, freeze_params=True)
mod = relay.quantize.prerequisite_optimize(mod, params)
return (mod, params, shape_dict, dtype, target, ImageNetValidator(shape_dict, "NHWC", preproc="keras_mobilenetv1"))
def import_mace_inceptionv3(self, target="llvm", dtype="float32"):
model_url = "https://cnbj1.fds.api.xiaomi.com/mace/miai-models/inception-v3/inception-v3.pb"
filename = "mace_inception-v3"
input_names = ["input:0"]
output_names = ["InceptionV3/Predictions/Reshape_1:0"]
onnx_model_file = self.get_onnx_from_tf1(model_url, filename, input_names, output_names)
import onnx
model = onnx.load(onnx_model_file)
shape_dict = {'input:0': [1, 299, 299, 3]}
mod, params = relay.frontend.from_onnx(model, shape_dict, freeze_params=True)
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
dtype = "float32" if dtype == "float32" else "float16"
return (mod, params, shape_dict, dtype, target, ImageNetValidator(shape_dict, "NHWC", preproc="keras"))
def import_mxnet_vgg16(self, target="llvm", dtype="float32"):
model, input_shape = gluon_model("vgg16", batch_size=1)
shape_dict = {"data": input_shape}
mod, params = relay.frontend.from_mxnet(model, shape_dict)
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
dtype = "float32" if dtype == "float32" else "float16"
return (mod, params, shape_dict, dtype, target, ImageNetValidator(shape_dict, preproc="mxnet"))
def import_mace_deeplabv3(self, target="llvm", dtype="float32"):
model_url = "https://cnbj1.fds.api.xiaomi.com/mace/miai-models/deeplab-v3-plus/deeplab-v3-plus-mobilenet-v2.pb"
filename = "mace_deeplab-v3-plus-mobilenet-v2"
graph_def = self.get_graphdef_from_tf1(model_url, filename)
shape_dict = {"sub_7": (1, 513, 513, 3)}
mod, params = relay.frontend.from_tensorflow(graph_def, shape=shape_dict,
outputs=["ResizeBilinear_2"])
# hack for insufficient pattern support in FlattenAtrousConv
# if it is called after convert to fp16 with mixed precision, it will not be able
# to catch cast.
# TODO(amalyshe) We need to extend FlattenAtrousConv but for now we are calling it
# explicitly
mod = tvm.relay.transform.FlattenAtrousConv()(mod)
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
dtype = "float32" if dtype == "float32" else "float16"
return (mod, params, shape_dict, dtype, target, Deeplabv3Validator(shape_dict, dtype))
def import_mace_yolov3(self, target="llvm", dtype="float32"):
model_url = "http://cnbj1.fds.api.xiaomi.com/mace/miai-models/yolo-v3/yolo-v3.pb"
filename = "mace_yolo-v3"
graph_def = self.get_graphdef_from_tf1(model_url, filename)
shape_dict = {"input_1": (1, 416, 416, 3)}
mod, params = relay.frontend.from_tensorflow(graph_def, shape=shape_dict,
outputs=["conv2d_59/BiasAdd","conv2d_67/BiasAdd","conv2d_75/BiasAdd"])
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
dtype = "float32" if dtype == "float32" else "float16"
return (mod, params, shape_dict, dtype, target, Yolov3Validator(shape_dict))
def import_onnx_ssd_resnet34(self, target="llvm", dtype="float32"):
archive_url = "https://github.com/onnx/models/raw/main/vision/object_detection_segmentation/ssd/model/ssd-12.tar.gz"
filename = "ssd-12.tar.gz"
from tvm.contrib import download
import onnx
import tarfile
download.download(archive_url, filename)
archive = tarfile.open(filename)
directory = "ssd_resnet34"
archive.extractall(directory)
archive.close()
directory = os.path.join(directory, "ssd-12")
model_file = os.path.join(directory, "ssd-12.onnx")
onnx_model = onnx.load(model_file)
shape_dict = {"image": (1, 3, 1200, 1200)}
mod, params = relay.frontend.from_onnx(onnx_model, shape_dict, freeze_params=True)
test_files_dir = os.path.join(directory, "test_data_set_0")
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
dtype = "float32" if dtype == "float32" else "float16"
#return (mod, params, shape_dict, dtype, target, ONNXTestSamplesValidator(test_files_dir, input_names=list(shape_dict.keys())))
return (mod, params, shape_dict, dtype, target, SSDResnetValidator())
def import_onnx_yolo_v3(self, target="llvm", dtype="float32"):
archive_url = "https://github.com/onnx/models/raw/main/vision/object_detection_segmentation/yolov3/model/yolov3-12.tar.gz"
filename = "yolov3-12.tar.gz"
from tvm.contrib import download
import onnx
import tarfile
download.download(archive_url, filename)
archive = tarfile.open(filename)
directory = "onnx_yolov3"
archive.extractall(directory)
archive.close()
directory = os.path.join(directory, "yolov3-12")
model_file = os.path.join(directory, "yolov3-12.onnx")
onnx_model = onnx.load(model_file)
shape_dict = {
"input_1": (1, 3, 416, 416),
"image_shape": (1, 2),
}
mod, params = relay.frontend.from_onnx(onnx_model, shape_dict, freeze_params=True)
test_files_dir = os.path.join(directory, "test_data_set_0")
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
dtype = "float32" if dtype == "float32" else "float16"
print("=" * 10)
print(mod)
print("=" * 10)
return (mod, params, shape_dict, dtype, target, ONNXYolov3Validator())
#return (mod, params, shape_dict, dtype, target, ONNXTestSamplesValidator(test_files_dir, input_names=list(shape_dict.keys())))
def import_onnx_faster_rcnn(self, target="llvm", dtype="float32"):
min_shape = 800.0
def _get_shape():
from PIL import Image
from tvm.contrib import download
# Download test image
image_url = "https://raw.githubusercontent.com/zhreshold/mxnet-ssd/master/data/demo/dog.jpg"
image_fn = "dog.png"
image_url = "https://raw.githubusercontent.com/onnx/models/main/vision/object_detection_segmentation/faster-rcnn/dependencies/demo.jpg"
image_fn = "demo.png"
download.download(image_url, image_fn)
# Prepare test image for inference
#import ipdb; ipdb.set_trace()
image = Image.open(image_fn)
print(image.size)
ratio = min_shape / min(image.size[0], image.size[1])
#return (3, int(ratio * image.size[1]), int(ratio * image.size[0])) # [c, h, w]
return (3, int(min_shape), int(min_shape)) # [c, h, w]
archive_url = "https://github.com/onnx/models/raw/main/vision/object_detection_segmentation/faster-rcnn/model/FasterRCNN-12.onnx"
filename = "FasterRCNN-12"
from tvm.contrib import download
import onnx
download.download(archive_url, filename)
onnx_model = onnx.load(filename)
shape_dict = {
"image": _get_shape(),
}
mod_file = f"onnx_faster_rcnn_mod_{dtype}.json"
params_file = f"onnx_faster_rcnn_params_{dtype}.json"
if not os.path.exists(mod_file):
mod, params = relay.frontend.from_onnx(onnx_model, shape_dict, freeze_params=True)
# downcast to float16
mod = convert_to_dtype(mod["main"], dtype)
with open(mod_file, "w") as file:
file.write(tvm.ir.save_json(mod))
with open(params_file, "wb") as file:
file.write(relay.save_param_dict(params))
else:
with open(mod_file, "r") as file:
mod = tvm.ir.load_json(file.read())
with open(params_file, "rb") as file:
params = relay.load_param_dict(file.read())
dtype = "float32" if dtype == "float32" else "float16"
print("=" * 10)
print(mod)
print("=" * 10)
return (mod, params, shape_dict, dtype, target, FasterRCNNValidator(min_shape))
def get_args():
import argparse
parser = argparse.ArgumentParser(
description="Tune and/or evaluate a curated set of models"
)
models = ModelImporter().available_models()
parser.add_argument(
"-m",
"--model",
type=str,
default=None,
required=True,
help="Model to tune and/or evaluate",
choices=models,
)
parser.add_argument(
"-t",
"--type",
type=str,
default="float16",
choices=["float32", "float16", "float16_acc32"],
help="Specify whether the model should be run with single or half precision floating point values",
)
parser.add_argument(
"-l", "--log", type=str, default=None, help="AutoTVM tuning logfile name"
)
parser.add_argument(
"-k", "--rpc_key", type=str, default="android", help="RPC key to use"
)
parser.add_argument(
"-r",
"--rpc_tracker_host",
type=str,
default=os.environ["TVM_TRACKER_HOST"],
help="RPC tracker host IP address",
)
parser.add_argument(
"-p",
"--rpc_tracker_port",
type=str,
default=os.environ["TVM_TRACKER_PORT"],
help="RPC tracker host port",
)
parser.add_argument(
"-T",
"--target",
type=str,
default="opencl --device=mali",
help="Compilation target",
)
parser.add_argument(
"--tune", action="store_true", help="Whether or not to run autotuning"
)
parser.add_argument(
"--debug",
action="store_true",
help="Use graph runtime debugger to output per layer perf. data and other statistics",
)
parser.add_argument(
"--VM",
action="store_true",
help="Use VM compiling and benchmarking",
)
args = parser.parse_args()
if args.rpc_tracker_port != None:
args.rpc_tracker_port = int(args.rpc_tracker_port)
args.tuning_options = {
"log_filename": args.log,
"early_stopping": None,
"measure_option": autotvm.measure_option(
builder=autotvm.LocalBuilder(build_func=ndk.create_shared, timeout=15),
runner=autotvm.RPCRunner(
args.rpc_key,
host=args.rpc_tracker_host,
port=args.rpc_tracker_port,
number=50,
timeout=15,
#min_repeat_ms=150,
#cooldown_interval=150
),
),
}
return args
args = get_args()
def main():
if "opencl" in args.target:
executor = Executor(use_tracker="android")
else:
executor = Executor()
executor.schedule(args.model, target=args.target, dtype=args.type)
if args.tune:
executor.tune_pending_benchmarks()
else:
executor.tune_pending_benchmarks(apply_previous_tune=True)
executor.run_pending_benchmarks()
def downcast_fp16(func, module):
from tvm.relay.expr_functor import ExprMutator
from tvm.relay.expr import Call, Var, Constant, TupleGetItem
from tvm.relay import transform as _transform
from tvm.relay import cast
from tvm.ir import IRModule
from tvm.relay import function as _function
"""Downcast to fp16 mutator
Parameters
---------
graph: Function
The original graph.
Retruns
-------
The graph after dowmcasting to half-precision floating-point.
"""
filter_list = ["vision.get_valid_counts", "vision.non_max_suppression"]
class DowncastMutator(ExprMutator):
"""Downcast to fp16 mutator"""
def visit_call(self, call):
dtype = "float32" if call.op.name in filter_list else "float16"
new_fn = self.visit(call.op)
# Collect the original dtypes
type_list = []
if call.op.name in filter_list:
# For NMS
for arg in call.args:
if isinstance(arg, TupleGetItem) and isinstance(
arg.tuple_value, Call
):
tuple_types = arg.tuple_value.checked_type.fields
type_list.append(tuple_types[arg.index].dtype)
if call.op.name == "vision.get_valid_counts":
tuple_types = call.checked_type.fields
for cur_type in tuple_types:
type_list.append(cur_type.dtype)
args = [self.visit(arg) for arg in call.args]
new_args = list()
arg_idx = 0
for arg in args:
if isinstance(arg, (Var, Constant)):
new_args.append(cast(arg, dtype=dtype))
else:
if call.op.name in filter_list:
if (
isinstance(arg, TupleGetItem)
and type_list[arg_idx] == "int32"
):
new_args.append(arg)
else:
new_args.append(cast(arg, dtype=dtype))
else:
new_args.append(arg)
arg_idx += 1
if (
call.op.name in filter_list
and call.op.name != "vision.get_valid_counts"
):
return cast(Call(new_fn, new_args, call.attrs), dtype="float16")
return Call(new_fn, new_args, call.attrs)
class UpcastMutator(ExprMutator):
"""upcast output back to fp32 mutator"""
def visit_call(self, call):
return cast(call, dtype="float32")
def infer_type(node, mod=None):
"""A method to infer the type of an intermediate node in the relay graph."""
if isinstance(mod, IRModule):
mod["main"] = _function.Function(tvm.relay.analysis.free_vars(node), node)
mod = _transform.InferType()(mod)
entry = mod["main"]
ret = entry.body
else:
new_mod = IRModule.from_expr(node)
if mod is not None:
new_mod.update(mod)
new_mod = _transform.InferType()(new_mod)
entry = new_mod["main"]
ret = entry if isinstance(node, _function.Function) else entry.body
return ret
func = infer_type(func, module)
downcast_pass = DowncastMutator()
func = downcast_pass.visit(func)
upcast_pass = UpcastMutator()
func = upcast_pass.visit(func)
func = infer_type(func, module)
new_mod = IRModule.from_expr(func)
# new_mod.update(module)
return new_mod
def get_input_data_shape_dict(graph_def, input_shape):
if isinstance(input_shape, list):
input_names = {}
shape_dict = {}
for i in range(len(input_shape)):
input_names[i] = graph_def.graph.input[i].name
shape_dict[input_names[i]] = input_shape[i]
else:
input_names = graph_def.graph.input[0].name
shape_dict = {input_names: input_shape}
return input_names, shape_dict
def gluon_model(name, batch_size=None):
if "resnet50_v1" in name or "mobilenet1.0" in name or "resnet50_v2" in name or "vgg16" in name:
model = gluon.model_zoo.vision.get_model(name, pretrained=True)
data_shape = (batch_size, 3, 224, 224)
elif "inceptionv3" in name:
model = gluon.model_zoo.vision.inception_v3(pretrained=True)
data_shape = (batch_size, 3, 299, 299)
else:
raise ValueError("Input shape unknown for gluon model: " + name)
return model, data_shape
def gluoncv_model(name, batch_size=None):
from gluoncv import model_zoo
if "yolo3" in name:
model = model_zoo.get_model(name, pretrained=True)
data_shape = (batch_size, 3, 416, 416)
return model, data_shape
class Validator(object):
def __init__(self, inputs):
if isinstance(inputs, dict):
self.inputs = inputs
else:
assert len(inputs) == 1
self.inputs = {"data" : inputs[0]}
def GetReference(self):
return []
def Validate(self):
return None
def GetInputDictionary(self):
return self.inputs
class ImageNetValidator(Validator):
def __init__(self, shape_dict, layout="NCHW", preproc=None):
assert layout in ("NCHW", "NHWC"), "Requested layout is not currently supported: " + layout
assert len(shape_dict) == 1
from PIL import Image
from tvm.contrib import download
from os.path import join, isfile
from matplotlib import pyplot as plt
name = list(shape_dict.keys())[0]
# Download ImageNet categories
categ_url = "https://github.com/uwsampl/web-data/raw/main/vta/models/"
categ_fn = "synset.txt"
download.download(join(categ_url, categ_fn), categ_fn)
self.synset = eval(open(categ_fn).read())
# Download test image
image_url = "https://homes.cs.washington.edu/~moreau/media/vta/cat.jpg"
image_fn = "cat.png"
download.download(image_url, image_fn)
# Prepare test image for inference
#import ipdb; ipdb.set_trace()
image = Image.open(image_fn)
if layout == "NHWC":
image = image.resize(shape_dict[name][1:-1])
elif layout == "NCHW":
image = image.resize(shape_dict[name][2:])
#image = self.preprocess(np.array(image))
if "mxnet" in preproc:
image = np.array(image) - np.array([123.0, 117.0, 104.0])
image /= np.array([58.395, 57.12, 57.375])
image = image.transpose((2, 0, 1))
image = image[np.newaxis, :]
elif "keras" in preproc:
image = np.array(image)[np.newaxis, :].astype("float32")
from tensorflow.keras.applications.inception_v3 import preprocess_input
image = preprocess_input(image)
elif "keras_mobilenetv1" in preproc:
image = np.array(image)[np.newaxis, :].astype("float32")
from tensorflow.keras.applications.mobilenet import preprocess_input
image = preprocess_input(image)
self.inputs = {name : image}
def Validate(self, m, ref_outputs=[]):
if isinstance(m, tvm.runtime.vm.VirtualMachine) or isinstance(m, tvm.runtime.profiler_vm.VirtualMachineProfiler):
tvm_output = m.get_outputs()[0]
else:
tvm_output = m.get_output(0)
#import ipdb; ipdb.set_trace()
top_categories = np.argsort(tvm_output.asnumpy()[0])
# Report top-5 classification results
print("\nTop5 predictions: \n")
top5 = np.flip(top_categories, axis=0)[:5]
# print("\t#1:", self.synset[top_categories[-1]])
# print("\t#2:", self.synset[top_categories[-2]])
# print("\t#3:", self.synset[top_categories[-3]])
# print("\t#4:", self.synset[top_categories[-4]])
# print("\t#5:", self.synset[top_categories[-5]])
print("\t#1:", self.synset[top5[1-1]])
print("\t#2:", self.synset[top5[2-1]])
print("\t#3:", self.synset[top5[3-1]])
print("\t#4:", self.synset[top5[4-1]])
print("\t#5:", self.synset[top5[5-1]])
print("\t", top5)
ImageNetClassifier = False
for k in top_categories[-5:]:
if "cat" in self.synset[k]:
ImageNetClassifier = True
assert ImageNetClassifier, "Failed ImageNet classifier validation check"
class VOCValidator(Validator):
# this function is from yolo3.utils.letterbox_image
def letterbox_image(self, image, size):
'''resize image with unchanged aspect ratio using padding'''
iw, ih = image.size
w, h = size
scale = min(w/iw, h/ih)
nw = int(iw*scale)
nh = int(ih*scale)
from PIL import Image
image = image.resize((nw,nh), Image.BICUBIC)
new_image = Image.new('RGB', size, (128,128,128))
new_image.paste(image, ((w-nw)//2, (h-nh)//2))
return new_image
def preprocess(self, img):
model_image_size = (416, 416)
boxed_image = self.letterbox_image(img, tuple(reversed(model_image_size)))
image_data = np.array(boxed_image, dtype='float32')
image_data /= 255.
image_data = np.transpose(image_data, [2, 0, 1])
image_data = np.expand_dims(image_data, 0)
return image_data
def __init__(self, shape_dict, layout="NCHW", preproc=None):
assert layout in ("NCHW", "NHWC"), "Requested layout is not currently supported: " + layout
assert len(shape_dict) == 1
from PIL import Image
from tvm.contrib import download
from os.path import join, isfile
from matplotlib import pyplot as plt
name = list(shape_dict.keys())[0]
# Download test image
image_url = "https://raw.githubusercontent.com/zhreshold/mxnet-ssd/master/data/demo/dog.jpg"
image_fn = "dog.png"
download.download(image_url, image_fn)
# Prepare test image for inference
#import ipdb; ipdb.set_trace()
image = Image.open(image_fn)
image_data = self.preprocess(image)
self.inputs = {name : image_data}
def Validate(self, m, ref_outputs=[]):
# class_IDs, scores, bounding_boxs
classid = m.get_output(0)
scores = m.get_output(1)
bounding_boxs = m.get_output(2)
for a in classid:
print(a)
class Deeplabv3Validator(Validator):
def __init__(self, input_shape, dtype):
from os.path import join
from tvm.contrib import download
assert isinstance(input_shape, dict)
assert dtype in ["float16", "float32"]
np.random.seed(1)
self.dtype = dtype
self.inputs = {}
for key in input_shape:
self.inputs[key] = np.random.normal(size=input_shape[key]).astype("float32")
categ_url = "https://github.com/Deelvin/qualcomm/raw/avoronov/rebase_master_v2/"
categ_fn = "deeplabv3_reference_output_{}".format(dtype)
download.download(join(categ_url, categ_fn), categ_fn)
# genered by target="llvm -keys=cpu" at np.random.seed(1)
self.ref_outputs = eval(open(categ_fn).read())
def GetReference(self):
return self.ref_outputs
def Validate(self, m, ref_outputs=[]):
if self.dtype == "float16":
rtol=1e-1
atol=1e-1
if self.dtype == "float32":
rtol=1e-3
atol=1e-3
if isinstance(m, tvm.runtime.vm.VirtualMachine) or isinstance(m, tvm.runtime.profiler_vm.VirtualMachineProfiler):
outputs = m.get_outputs()
for i in range(len(outputs)):
tvm_output = outputs[i]
np.testing.assert_allclose(tvm_output.asnumpy(), ref_outputs[i], rtol=rtol, atol=atol)
print("Deeplabv3Validator pass:", "rtol", rtol, "atol",atol)
else:
for i in range(m.get_num_outputs()):
tvm_output = m.get_output(i)
np.testing.assert_allclose(tvm_output.asnumpy(), ref_outputs[i], rtol=rtol, atol=atol)
print("Deeplabv3Validator pass:", "rtol", rtol, "atol",atol)
class Yolov3Validator(Validator):
class BoundBox:
def __init__(self, xmin, ymin, xmax, ymax, objness = None, classes = None):
self.xmin = xmin
self.ymin = ymin
self.xmax = xmax
self.ymax = ymax
self.objness = objness
self.classes = classes
self.label = -1
self.score = -1
def get_label(self):
if self.label == -1:
self.label = np.argmax(self.classes)
return self.label
def get_score(self):
if self.score == -1:
self.score = self.classes[self.get_label()]
return self.score
def decode_netout(netout, anchors, obj_thresh, net_h, net_w):
grid_h, grid_w = netout.shape[:2]
nb_box = 3
netout = netout.reshape((grid_h, grid_w, nb_box, -1))
nb_class = netout.shape[-1] - 5
boxes = []
def _sigmoid(x):
return 1. / (1. + np.exp(-x))
netout[..., :2] = _sigmoid(netout[..., :2])
netout[..., 4:] = _sigmoid(netout[..., 4:])
netout[..., 5:] = netout[..., 4][..., np.newaxis] * netout[..., 5:]
netout[..., 5:] *= netout[..., 5:] > obj_thresh
for i in range(grid_h*grid_w):
row = i / grid_w
col = i % grid_w
for b in range(nb_box):
# 4th element is objectness score
objectness = netout[int(row)][int(col)][b][4]
if(objectness.all() <= obj_thresh): continue
# first 4 elements are x, y, w, and h
x, y, w, h = netout[int(row)][int(col)][b][:4]
x = (col + x) / grid_w # center position, unit: image width
y = (row + y) / grid_h # center position, unit: image height
w = anchors[2 * b + 0] * np.exp(w) / net_w # unit: image width
h = anchors[2 * b + 1] * np.exp(h) / net_h # unit: image height
# last elements are class probabilities
classes = netout[int(row)][col][b][5:]
box = Yolov3Validator.BoundBox(x-w/2, y-h/2, x+w/2, y+h/2, objectness, classes)
boxes.append(box)
return boxes
def correct_yolo_boxes(boxes, image_h, image_w, net_h, net_w):
new_w, new_h = net_w, net_h
for i in range(len(boxes)):
x_offset, x_scale = (net_w - new_w)/2./net_w, float(new_w)/net_w
y_offset, y_scale = (net_h - new_h)/2./net_h, float(new_h)/net_h
boxes[i].xmin = int((boxes[i].xmin - x_offset) / x_scale * image_w)
boxes[i].xmax = int((boxes[i].xmax - x_offset) / x_scale * image_w)
boxes[i].ymin = int((boxes[i].ymin - y_offset) / y_scale * image_h)
boxes[i].ymax = int((boxes[i].ymax - y_offset) / y_scale * image_h)
def bbox_iou(box1, box2):
def _interval_overlap(interval_a, interval_b):
x1, x2 = interval_a
x3, x4 = interval_b
if x3 < x1:
if x4 < x1:
return 0
else:
return min(x2,x4) - x1
else:
if x2 < x3:
return 0
else:
return min(x2,x4) - x3
intersect_w = _interval_overlap([box1.xmin, box1.xmax], [box2.xmin, box2.xmax])
intersect_h = _interval_overlap([box1.ymin, box1.ymax], [box2.ymin, box2.ymax])
intersect = intersect_w * intersect_h
w1, h1 = box1.xmax-box1.xmin, box1.ymax-box1.ymin
w2, h2 = box2.xmax-box2.xmin, box2.ymax-box2.ymin
union = w1*h1 + w2*h2 - intersect
return float(intersect) / union
def do_nms(boxes, nms_thresh):
if len(boxes) > 0:
nb_class = len(boxes[0].classes)
else:
return
for c in range(nb_class):
sorted_indices = np.argsort([-box.classes[c] for box in boxes])
for i in range(len(sorted_indices)):
index_i = sorted_indices[i]
if boxes[index_i].classes[c] == 0: continue
for j in range(i+1, len(sorted_indices)):
index_j = sorted_indices[j]
if Yolov3Validator.bbox_iou(boxes[index_i], boxes[index_j]) >= nms_thresh:
boxes[index_j].classes[c] = 0
# load and prepare an image
@staticmethod
def load_image_pixels(filename, shape):
try:
from keras.preprocessing.image import load_img
from keras.preprocessing.image import img_to_array
except:
from tensorflow.keras.utils import load_img
from tensorflow.keras.utils import img_to_array
# load the image to get its shape
image = load_img(filename)
width, height = image.size
# load the image with the required size
image = load_img(filename, target_size=shape)
# convert to numpy array
image = img_to_array(image)
# scale pixel values to [0, 1]
image = image.astype('float32')
image /= 255.0
# add a dimension so that we have one sample
image = np.expand_dims(image, 0)
return image, width, height
# get all of the results above a threshold
@staticmethod
def get_boxes(boxes, labels, thresh):
v_boxes, v_labels, v_scores = list(), list(), list()
# enumerate all boxes
for box in boxes:
# enumerate all possible labels
for i in range(len(labels)):
# check if the threshold for this label is high enough
if box.classes[i] > thresh:
v_boxes.append(box)
v_labels.append(labels[i])
v_scores.append(box.classes[i]*100)
# don't break, many labels may trigger for one box
return v_boxes, v_labels, v_scores
# draw all results
@staticmethod
def draw_boxes(filename, v_boxes, v_labels, v_scores):
from matplotlib import pyplot
from matplotlib.patches import Rectangle
# load the image
from PIL import Image
if ".png" not in filename:
name, extension = filename.rsplit('.', 1)
im1 = Image.open(filename)
filename = "{}.png".format(name)
im1.save(filename)
data = pyplot.imread(filename)
# plot the image
pyplot.imshow(data)
# get the context for drawing boxes
ax = pyplot.gca()
# plot each box
for i in range(len(v_boxes)):
box = v_boxes[i]
# get coordinates
y1, x1, y2, x2 = box.ymin, box.xmin, box.ymax, box.xmax
# calculate width and height of the box
width, height = x2 - x1, y2 - y1
# create the shape
rect = Rectangle((x1, y1), width, height, fill=False, color='white')
# draw the box
ax.add_patch(rect)
# draw text and score in top left corner
label = "%s (%.3f)" % (v_labels[i], v_scores[i])
pyplot.text(x1, y1, label, color='white')
# show the plot
pyplot.show()
def __init__(self, input_shape, dtype="float32"):
from tvm.contrib import download
from os.path import join
n, h, w, c = list(input_shape.values())[0]
self.input_w, self.input_h = h, w
# Download Coco names
names_url = "https://github.com/pjreddie/darknet/raw/master/data/"
names_fn = "coco.names"
download.download(join(names_url, names_fn), names_fn, overwrite=True)