Created
November 8, 2017 06:52
-
-
Save akirasosa/b31c3096d9cc1959cbbd8af40993c92d to your computer and use it in GitHub Desktop.
Benchmark conv block and depthwise conv block.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import time | |
import numpy as np | |
from keras import Input | |
from keras.applications.mobilenet import DepthwiseConv2D | |
from keras.engine import Model | |
from keras.layers import Conv2D | |
batch_num = 16 | |
def conv_net(inputs): | |
x = Conv2D(128, (3, 3), padding='same', strides=(2, 2))(inputs) | |
return Model(inputs, x) | |
def depthwise_conv_net(inputs): | |
x = DepthwiseConv2D((3, 3), padding='same', strides=(2, 2))(inputs) | |
x = Conv2D(128, (1, 1))(x) | |
return Model(inputs, x) | |
def main(): | |
input_tensor = Input(shape=(64, 64, 64)) | |
experiments = [ | |
{'name': 'conv', 'model': conv_net(input_tensor)}, | |
{'name': 'depthwise_conv', 'model': depthwise_conv_net(input_tensor)}, | |
] | |
inputs = np.random.randn(batch_num, 64, 64, 64) | |
for e in experiments: | |
time_per_batch = [] | |
for i in range(10): | |
start = time.time() | |
e['model'].predict(inputs, batch_size=batch_num) | |
elapsed = time.time() - start | |
time_per_batch.append(elapsed) | |
print(elapsed, e['name']) | |
time_per_batch = np.array(time_per_batch) | |
print(e['name']) | |
print(time_per_batch[1:].mean()) | |
print(time_per_batch[1:].std()) | |
if __name__ == '__main__': | |
main() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment