python - 使用 CNN 和 Keras-Tf python 的 dogs_cats 分类数据集的准确性不够高

Question

伙计们，我正在尝试使用 CNN 对 Dogs vs Cats 数据集进行分类。顺便说一句，我是深度学习初学者。

数据集链接可以从here获得.我还使用 MLP 对上述数据集进行了分类，训练精度为 70%，测试精度为 62%。所以我决定使用 CNN 来提高分数。

但不幸的是，我仍然得到非常相似的结果。这是我的代码:

from sklearn.cross_validation import train_test_split
from sklearn.preprocessing import LabelEncoder
from keras.layers import Dense, Activation, Flatten, Dropout
from keras.layers.convolutional import Convolution2D
from keras.layers.convolutional import MaxPooling2D
from keras.models import Sequential
from keras.utils import np_utils
from keras.optimizers import SGD
from keras.datasets import mnist
from keras import backend as K
from imutils import paths
import numpy as np
import argparse
import cPickle
import h5py
import sys
import cv2
import os

K.set_image_dim_ordering('th')

def image_to_feature_vector(image, size=(28, 28)):
    return cv2.resize(image, size)

print("[INFO] pre-processing images...")
imagePaths = list(paths.list_images(raw_input('path to dataset: ')))

data   = []
labels = []

for (i, imagePath) in enumerate(imagePaths):
    image = cv2.imread(imagePath)
    label = imagePath.split(os.path.sep)[-1].split(".")[0]
    features = image_to_feature_vector(image)
    data.append(features)
    labels.append(label)

    if i > 0 and i % 1000 == 0:
        print("[INFO] processed {}/{}".format(i, len(imagePaths)))

le     = LabelEncoder()
labels = le.fit_transform(labels)
labels = np_utils.to_categorical(labels, 2)
data   = np.array(data) / 255.0

print("[INFO] constructing training/testing split...")
(X_train, X_test, y_train, y_test) = train_test_split(data, labels, test_size=0.25, random_state=42)

X_train = X_train.reshape(X_train.shape[0], 3, 28, 28).astype('float32')
X_test  = X_test.reshape(X_test.shape[0], 3, 28, 28).astype('float32')
num_classes = y_test.shape[1]

def basic_model():
    model = Sequential()
    model.add(Convolution2D(32, 3, 3, border_mode='valid', init='uniform', bias=True, input_shape=(3, 28, 28), activation='relu'))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Dropout(0.25))
    model.add(Flatten())
    model.add(Dense(128, activation='relu'))
    model.add(Dense(num_classes, activation='softmax'))

    sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
    model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
    return model

model = basic_model()

model.fit(X_train, y_train, validation_data=(X_test, y_test), nb_epoch=25, batch_size=50, shuffle=True, verbose=1)

print('[INFO] Evaluating the model on test data...')
scores = model.evaluate(X_test, y_test, batch_size=100, verbose=1)
print("\nAccuracy: %.4f%%\n\n"%(scores[1]*100))

我使用的 CNN 模型非常基础，但我认为它还不错。我按照各种教程来学习它。我什至使用了这种架构，但得到了类似的结果(65% 的测试准确率):

def baseline_model():
    model = Sequential()
    model.add(Convolution2D(30, 5, 5, border_mode='valid', input_shape=(3, 28, 28), activation='relu'))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Convolution2D(15, 3, 3, activation='relu'))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Dropout(0.2))
    model.add(Flatten())
    model.add(Dense(128, activation='relu'))
    model.add(Dense(50, activation='relu'))
    model.add(Dense(num_classes, activation='softmax'))

    sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
    model.compile(loss='binary_crossentropy', optimizer=sgd, metrics=['accuracy'])
    return model

对于优化器，我还尝试了使用默认参数的 adam 和 model.complie 损失函数，我也尝试了 categorical_crossentropy 但没有(或非常轻微)改进。

你能建议我哪里出错了或者我可以做些什么来提高效率吗？(如果可能，在几个 epoch 中)

(我是深度学习和keras编程的初学者...)

编辑:所以我设法达到了 70.224% 的测试准确率和 74.27% 的训练准确率。 CNN架构是CONV => CONV => POOL => DROPOUT => FLATTEN => DENSE*3

(几乎没有过拟合，因为训练 acc:74%，测试是:70% 🙂)

但仍然对进一步增加它的建议持开放态度，70% 肯定偏低...

Answer 1

使用 (128,3,3) 或 (64,3,3) 可以解决精度问题。您使用了多少个纪元？如果你使用 20 个以上的 epoch，那就太好了。

试试这个:

model.add(Convolution2D(32, 3, 3, 3, border_mode='full'))
model.add(Activation('relu'))
model.add(Convolution2D(32, 32, 3, 3))
model.add(Activation('relu'))
model.add(MaxPooling2D(poolsize=(2, 2)))
model.add(Dropout(0.25))

model.add(Convolution2D(64, 32, 3, 3, border_mode='full'))
model.add(Activation('relu'))
model.add(Convolution2D(64, 64, 3, 3))
model.add(Activation('relu'))
model.add(MaxPooling2D(poolsize=(2, 2)))
model.add(Dropout(0.25))

model.add(Flatten())
model.add(Dense(64*8*8, 512))
model.add(Activation('relu'))
model.add(Dropout(0.5))

model.add(Dense(512, 2))
model.add(Activation('softmax'))