模型预测在拟合后给出相同的结果

Question

嗨，我是ML和Tensorflow的初学者，请原谅我不理解复杂的理论。

我正在构建一个图像分类器CNN作为一种实践形式。该模型是使用MobileNetv2进行训练的，它应该对猫、狗和熊猫的图像进行分类。在训练了我的模型(准确率达到92%(后，我尝试使用model.product((来评估它对新图像的处理效果，但我注意到我所有的输出都是1。即使我使用了以前相同的训练数据，也会发生这种情况。顺便说一下，我使用了2700张(每节课900张(图像进行训练，使用了300张进行验证。

这是我的代码

%tensorflow_version 2.x  # this line is not required unless you are in a notebook
import tensorflow as tf
from tensorflow.keras import datasets, layers, models
from keras.preprocessing.image import load_img
from keras.preprocessing.image import img_to_array
from keras.preprocessing import image
from keras.preprocessing.image import ImageDataGenerator
from PIL import Image
import matplotlib.pyplot as plt
import numpy as np
import os
import PIL
import PIL.Image
import tensorflow_datasets as tfds
import pathlib
from google.colab import drive
drive.mount('/content/gdrive')
IMAGE_SIZE=[150,150]
train_path = "/content/gdrive/MyDrive/Colab Notebooks/Cat_Dog_Panda_CNN/train"
test_path = "/content/gdrive/MyDrive/Colab Notebooks/Cat_Dog_Panda_CNN/test"
IMAGE_SHAPE=[150,150,3]
base_model = tf.keras.applications.MobileNetV2(input_shape=IMAGE_SHAPE,
include_top=False,
weights='imagenet')
base_model.trainable = False
global_average_layer = tf.keras.layers.GlobalAveragePooling2D()
prediction_layer = tf.keras.layers.Dense(3)
model = tf.keras.Sequential([
base_model,
global_average_layer,
prediction_layer
])
model.summary()

base_learning_rate = 0.0001
model.compile(optimizer=tf.keras.optimizers.RMSprop(lr=base_learning_rate),
loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
metrics=['accuracy'])
# creates a data generator object that transforms images
train_datagen = ImageDataGenerator(
rescale=1./255,
rotation_range=40,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
fill_mode='nearest')
test_datagen = ImageDataGenerator (rescale=1./255)
training_set = train_datagen.flow_from_directory(train_path, target_size=IMAGE_SIZE, batch_size=32, class_mode='categorical')
testing_set = test_datagen.flow_from_directory(test_path, target_size=IMAGE_SIZE, batch_size=32, class_mode='categorical')
model.fit(
training_set,
epochs=3,
validation_data=testing_set)
img = Image.open("/content/gdrive/MyDrive/Colab Notebooks/Cat_Dog_Panda_CNN/test/panda/panda_00094.jpg").convert('RGB').resize((150, 150), Image.ANTIALIAS)
img = np.array(img)
predictions = model.predict(img[None,:,:])
np.argmax(predictions[0])

Answer 1

在model.compile中，您有loss=tf.keras.loss.BinaryCrossentropy(from_gits=True(。既然你有3个班，你就应该有

loss=tf.keras.losses.CategoricalCrossentropy()

Mobilenet模型使用缩放在-1到+1范围内的像素值进行训练。所以你应该将重缩放设置为

rescale=1/127.5-1

如果要将图像输入model.predict，则必须执行相同的预处理即重新缩放图像，并将图像大小调整为与您在训练中使用的大小相同，即(150150(。