我正在尝试使用tflearn使用卷积神经网络对图像进行分类。输入图像的大小不同,我将它们全部调整为32x32x3(我正在使用所有RGB通道)。这是我的代码。
import os
import numpy as np
from random import shuffle
from tqdm import tqdm
import tflearn
import cv2
from tflearn.layers.conv import conv_2d, max_pool_2d
from tflearn.layers.core import input_data, dropout, fully_connected
from tflearn.layers.estimator import regression
from tflearn.data_preprocessing import ImagePreprocessing
from tflearn.data_augmentation import ImageAugmentation
import tensorflow as tf
import matplotlib.pyplot as plt
import pickle
TRAIN_DIR = 'some_dir'
def label_img(img):
word_label = img.split('.')[0]
if 'metal' in word_label:
return np.array([1,0,0,0])
elif 'plastic' in word_label:
return np.array([0,1,0,0])
elif 'paper' in word_label:
return np.array([0,0,1,0])
elif 'glass' in word_label:
return np.array([0,0,0,1])
def create_train_data():
training_data = []
for img in tqdm(os.listdir(TRAIN_DIR)):
label = label_img(img)
path = os.path.join(TRAIN_DIR,img)
img = cv2.imread(path,1)
img = cv2.resize(img, (32,32), interpolation = cv2.INTER_AREA)
float_img = img.astype(float)
float_img = float_img.reshape([-1,3,32,32])
float_img = float_img.transpose([0,2,3,1])
training_data.append([float_img, label])
shuffle(training_data)
with open('training_data.pickle','wb') as f:
pickle.dump(training_data,f)
return training_data
if os.path.exists('training_data.pickle'):
pickle_in = open('training_data.pickle','rb')
train_data = pickle.load(pickle_in)
else:
train_data = create_train_data()
img_prep = ImagePreprocessing()
img_prep.add_featurewise_zero_center()
img_prep.add_featurewise_stdnorm()
img_aug = ImageAugmentation()
img_aug.add_random_flip_leftright()
img_aug.add_random_rotation(max_angle=15.)
network = input_data(shape=[None, 32, 32, 3], data_preprocessing=img_prep, data_augmentation=img_aug)
network = conv_2d(network, 32, 3, strides=1, padding='same', activation='relu', bias=True, bias_init='zeros', weights_init='uniform_scaling')
network = max_pool_2d(network, 2 , strides=None, padding='same')
network = conv_2d(network, 64, 3, strides=1, padding='same', activation='relu', bias=True, bias_init='zeros', weights_init='uniform_scaling')
network = conv_2d(network, 64, 3 , strides=1, padding='same', activation='relu', bias=True, bias_init='zeros', weights_init='uniform_scaling')
network = max_pool_2d(network, 2 , strides=None, padding='same')
network = fully_connected(network, 512, activation='relu')
network = dropout(network, 0.5)
network = fully_connected(network, 4, activation='softmax')
network = regression(network, optimizer='adam', loss='categorical_crossentropy', learning_rate=0.001)
train = train_data[:-200]
test = train_data[-200:]
X = np.array([i[0] for i in train]).reshape(-1, 32, 32, 3)
Y = np.array([i[1] for i in train])
test_x = np.array([i[0] for i in test]).reshape(-1, 32, 32, 3)
test_y = np.array([i[1] for i in test])
model = tflearn.DNN(network, tensorboard_dir='log')
model.fit(X, Y, n_epoch=50, shuffle=True, validation_set= (test_x, test_y) , show_metric=True, batch_size=100 , run_id='aa2')
我收到了函数
代码的最后一行的错误model.fit(....)
错误表示“值错误:设置带序列的数组元素”。 我查看了tflearn文档。它表示输入(在我的例子中,函数 model.fit 中的 X,Y 参数)和验证( test_x,test_y 函数中的参数都需要数组的数据类型。所以,我已将它们全部转换为numpy数组,但我仍然收到此错误。 另外,我在另一台安装了tensorflow的virtualenv的机器上运行此代码。代码完全在那台机器上运行。我搜索了许多论坛,但没有人得到答案。
还有一件事,我使用过tflearn的图像预处理和图像增强。它们不应该通过添加更多图像(15度的旋转图像)在技术上增加数据集大小吗?数据集大小也没有增加。使用这些功能时是否有任何错误?
提前致谢!
答案 0 :(得分:0)
更改此行:
float_img = float_img.reshape([-1,3,32,32])
到
float_img = float_img.reshape([3,32,32])
你也不需要再次重塑:
X = np.array([i[0] for i in train]).reshape(-1, 32, 32, 3)