我如何从filepath格式化我的图像到与python中的mnist.load_data（）相同的方式？

Question

How can I format my images from file path to the same way as mnist.load_data() ? I'm currently getting my images from a driectory/file path, how can I format these images the same way as mnist uses for mnist.load_data() ?

Answer 1

The keras.datasets.mnist.load_data actually just loads a preprocessed pickle file . If you check the data type of X_train & X_test they are just a numpy 2D array of float representing images pixel value (0-255). While y_train & y_test are just numpy 1D array representing the classes/labels (0-9).

So the first way to imitate that functionality is read you images using image processing library (eg. opencv ) into numpy array & finally split them using sklearn :

import numpy as np
import cv2
from sklearn.model_selection import train_test_split

X = []
y = []

# convert color image to 2D array (grayscale) & rescale
data = cv2.imread('zero.jpg',0) / 255.0
label = 0 # label/class of the image
X.append(data)
y.append(label)

# loop trough all images ...

# split for training & testing
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33)

Another way that you can try is using keras ImageDataGenerator.flow_from_directory(color_mode='grayscale') . The output is an ImageDataGenerator object that can be passed to keras model.fit_generator() function. In order to make use this function, you should arrange your dataset into train & test directories where each of them contains subdirectories representing classes of the images inside them. Please find detailed explanation in here .