3

根据[本教程]（https://towardsdatascience.com/keras-transfer-learning-for-beginners-6c9b8b7143e ），我正在使用迁移学习方法在 Keras 中使用按训练的 VGG19 模型。它展示了如何训练模型，但不展示如何为预测准备测试图像。

在评论部分它说：

获取图像，使用相同的preprocess_image函数对图像进行预处理，然后调用model.predict(image). 这将为您提供该图像上模型的预测。使用argmax(prediction)，您可以找到图像所属的类。

我在代码中找不到名为preprocess_imageused 的函数。我做了一些搜索并想到使用本教程提出的方法。

但这给出了一个错误说：

decode_predictions expects a batch of predictions (i.e. a 2D array of shape (samples, 1000)). Found array with shape: (1, 12)

我的数据集有 12 个类别。这是训练模型的完整代码以及我是如何得到这个错误的：

import pandas as pd
import numpy as np
import os
import keras
import matplotlib.pyplot as plt

from keras.layers import Dense, GlobalAveragePooling2D
from keras.applications.vgg19 import VGG19
from keras.preprocessing import image
from keras.applications.vgg19 import preprocess_input
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Model
from keras.optimizers import Adam

base_model = VGG19(weights='imagenet', include_top=False)

x=base_model.output                                                          
x=GlobalAveragePooling2D()(x)                                                
x=Dense(1024,activation='relu')(x)                                           
x=Dense(1024,activation='relu')(x)                                           
x=Dense(512,activation='relu')(x)        

preds=Dense(12,activation='softmax')(x)                                      
model=Model(inputs=base_model.input,outputs=preds)                           

# view the layer architecture
# for i,layer in enumerate(model.layers):
#   print(i,layer.name)

for layer in model.layers:
    layer.trainable=False

for layer in model.layers[:20]:
    layer.trainable=False

for layer in model.layers[20:]:
    layer.trainable=True

train_datagen=ImageDataGenerator(preprocessing_function=preprocess_input)

train_generator=train_datagen.flow_from_directory('dataset',
                    target_size=(96,96), # 224, 224
                    color_mode='rgb',
                    batch_size=64,
                    class_mode='categorical',
                    shuffle=True)

model.compile(optimizer='Adam',loss='categorical_crossentropy',metrics=['accuracy'])

step_size_train=train_generator.n//train_generator.batch_size

model.fit_generator(generator=train_generator,
    steps_per_epoch=step_size_train,
    epochs=5)

# model.predict(new_image)

IPython：

In [3]: import classify_tl                                                                                                                                                   
Found 4750 images belonging to 12 classes.
Epoch 1/5
74/74 [==============================] - 583s 8s/step - loss: 2.0113 - acc: 0.4557
Epoch 2/5
74/74 [==============================] - 576s 8s/step - loss: 0.8222 - acc: 0.7170
Epoch 3/5
74/74 [==============================] - 563s 8s/step - loss: 0.5875 - acc: 0.7929
Epoch 4/5
74/74 [==============================] - 585s 8s/step - loss: 0.3897 - acc: 0.8627
Epoch 5/5
74/74 [==============================] - 610s 8s/step - loss: 0.2689 - acc: 0.9071

In [6]: model = classify_tl.model                                                                                                                                            

In [7]: print(model)                                                                                                                                                         
<keras.engine.training.Model object at 0x7fb3ad988518>

In [8]: from keras.preprocessing.image import load_img                                                                                                                       

In [9]: image = load_img('examples/0021e90e4.png', target_size=(96,96))                                                                                                      

In [10]: from keras.preprocessing.image import img_to_array                                                                                                                  

In [11]: image = img_to_array(image)                                                                                                                                         

In [12]: image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))                                                                                          

In [13]: from keras.applications.vgg19 import preprocess_input                                                                                                               

In [14]: image = preprocess_input(image)                                                                                                                                     

In [15]: yhat = model.predict(image)                                                                                                                                         

In [16]: print(yhat)                                                                                                                                                         
[[1.3975363e-06 3.1069856e-05 9.9680350e-05 1.7175063e-03 6.2767825e-08
  2.6133494e-03 7.2859187e-08 6.0187017e-07 2.0794137e-06 1.3714411e-03
  9.9416250e-01 2.6067207e-07]]

In [17]: from keras.applications.vgg19 import decode_predictions                                                                                                             

In [18]: label = decode_predictions(yhat)

IPython 提示中的最后一行导致以下错误：

ValueError: `decode_predictions` expects a batch of predictions (i.e. a 2D array of shape (samples, 1000)). Found array with shape: (1, 12)

我应该如何正确输入我的测试图像并获得预测？

4

1 回答 1

1

decode_predictions用于根据具有 1000 个类别的 ImageNet 数据集中的类别标签对模型的预测进行解码。但是，您的微调模型只有 12 个类。因此，在这里使用是没有意义的decode_predictions。当然，您必须知道这 12 个类别的标签是什么。因此，只需在预测中取最大分数的索引并找到它的标签：

# create a list containing the class labels
class_labels = ['class1', 'class2', 'class3', ...., 'class12']

# find the index of the class with maximum score
pred = np.argmax(class_labels, axis=-1)

# print the label of the class with maximum score
print(class_labels[pred[0]])
于 2018-12-27T08:45:20.630 回答