forked from isl-org/Open3D-PointNet2-Semantic3D
-
Notifications
You must be signed in to change notification settings - Fork 0
/
benchmark.py
72 lines (58 loc) · 2.14 KB
/
benchmark.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
import json
import numpy as np
import tensorflow as tf
import time
from predict import Predictor
if __name__ == "__main__":
checkpoint = "logs/semantic_backup_full_submit_dec_10/best_model_epoch_275.ckpt"
hyper_params = json.loads(open("semantic.json").read())
predictor = Predictor(
checkpoint_path=checkpoint, num_classes=9, hyper_params=hyper_params
)
batch_size = 64
# Init data
points_with_colors = np.random.randn(batch_size, hyper_params["num_point"], 6)
# Warm up
pd_labels = predictor.predict(points_with_colors)
# Benchmark
s = time.time()
profiler = tf.profiler.Profiler(predictor.sess.graph)
run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
_ = predictor.predict(
points_with_colors, run_options=run_options, run_metadata=run_metadata
)
profiler.add_step(0, run_metadata)
batch_time = time.time() - s
sample_time = batch_time / batch_size
print(
"Batch size: {}, batch_time: {}, sample_time: {}".format(
batch_size, batch_time, sample_time
)
)
option_builder = tf.profiler.ProfileOptionBuilder
opts = (
option_builder(option_builder.time_and_memory())
.with_step(-1) # with -1, should compute the average of all registered steps.
.with_file_output("tf-profile.txt")
.select(["micros", "bytes", "occurrence"])
.order_by("micros")
.build()
)
# Profiling info about ops are saved in 'test-%s.txt' % FLAGS.out
profiler.profile_operations(options=opts)
for batch_size in [2 ** n for n in range(8)]:
# Init data
points_with_colors = np.random.randn(batch_size, hyper_params["num_point"], 6)
# Warm up
pd_labels = predictor.predict(points_with_colors)
# Benchmark
s = time.time()
_ = predictor.predict(points_with_colors)
batch_time = time.time() - s
sample_time = batch_time / batch_size
print(
"Batch size: {}, batch_time: {}, sample_time: {}".format(
batch_size, batch_time, sample_time
)
)