如何在不使用model.fit_generator的情况下将旋转应用于Keras中的图像? [英] How can I apply rotation to image in Keras without using model.fit_generator?

问题描述

我正在使用卷积神经网络研究图像像素分类问题. 我的训练images的大小为128x128x3，而 标签mask是128x128

I am working on an image pixel classification problem using convolution neural nets. The size of my training images is 128x128x3 and the size of the label mask is 128x128

我在Keras进行如下培训:

I do training in Keras as follows:

Xtrain, Xvalid, ytrain, yvalid = train_test_split(images, masks,test_size=0.3, random_state=567)

model.fit(Xtrain, ytrain, batch_size=32, epochs=20, verbose=1, shuffle=True, validation_data=(Xvalid, yvalid))

但是，我想对Xtrain和ytrain分别应用随机2D旋转，其大小也分别为128x128x3和128x128.更具体地说，我想将此旋转应用于每个时代迭代.

However, I want to apply a Random 2D rotation to Xtrain and ytrain which is also of size 128x128x3 and 128x128 respectively. More specifically, I want to apply this rotation for every epoch iteration.

暂时，我想继续使用model.fit而不使用model.fit_generator，因为我知道数据增强通常是使用.fit_generator进行的.

For the time being, I would like to continue using model.fit and not use model.fit_generator, as I know data augmentation is commonly done using .fit_generator.

所以从本质上讲，我想循环model.fit，以便每个时期随机旋转Xtrain和ytrain.我是Python和Keras的新手，所以即使有可能，欢迎提供任何见识.

So essentially, I want to loop model.fit so that Xtrain and ytrain is randomly rotated for every epoch. I am new to Python and Keras so any insights are welcome if this is even possible.

推荐答案

下面是使用ImageDataGenerator将输出保存到指定目录的示例，从而避免了使用model.fit_generator的要求.

Here's an exmaple of using ImageDataGenerator to save the output to a specified directory, thus getting around the requirement to use model.fit_generator.

from keras.preprocessing.image import ImageDataGenerator, array_to_img, img_to_array, load_img

datagen = ImageDataGenerator(
        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')

img = load_img('data/train/cats/cat.0.jpg')  # this is a PIL image
x = img_to_array(img)  # this is a Numpy array with shape (3, 150, 150)
x = x.reshape((1,) + x.shape)  # this is a Numpy array with shape (1, 3, 150, 150)

# the .flow() command below generates batches of randomly transformed images
# and saves the results to the `preview/` directory
i = 0
for batch in datagen.flow(x, batch_size=1,
                          save_to_dir='preview', save_prefix='cat', save_format='jpeg'):
    i += 1
    if i > 20:
        break  # otherwise the generator would loop indefinitely

来自此处: https ://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html

您可以更改参数以适合您的用例，然后生成X_train和X_valid或任何数据集，然后加载到内存中并使用简单的旧model.fit.

You can change the args to suit your use case and then generate your X_train and X_valid or whatever datasets, then load into memory and use plain old model.fit.

这篇关于如何在不使用model.fit_generator的情况下将旋转应用于Keras中的图像?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！