使用预训练模型预测时，cpu比gpu快

Question

我在使用Resnet50预训练模型预测的时候，用cpu花了1.6s，用gpu花了6.4s，机子比较老用的tensorflow1.12。代码用的是官方提供的，甚至不知道哪里出问题。

import time

# from tensorflow.keras.applications.resnet50 import ResNet50,preprocess_input, decode_predictions
# from tensorflow.keras.preprocessing import image
from keras.applications.resnet50 import ResNet50,preprocess_input, decode_predictions
from keras.preprocessing import image
import numpy as np

ss = time.time()
model = ResNet50(weights=None, include_top=True)
model.load_weights(r'E:\pycode\fight\resnet18_imagenet_1000.h5', by_name=True)
ee = time.time()

img_path = r'E:\pycode\fight\hashiqi.jpg'
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)

s = time.time()
preds = model.predict(x)
e = time.time()
# decode the results into a list of tuples (class, description, probability)
# (one such list for each sample in the batch)
print('Predicted:', decode_predictions(preds, top=3)[0])
print(e-s)
print(ee-ss)

cpu

gpu

Answer 1

1. 正常情况下e-s的实际肯定gpu更小
2. 先检测gpu有没有启动（运行的时候gpu使用率）
3. 如果第二步是对的，那就得看后处理部分，是不是实际花在GPU->CPU的过程