如何重塑 ndarray 以预测 model？

Question

I need to read two images, convert them to size 150x150 and add them to an array that needs to be reshaped into a shape of (2, 150, 150, 3) in order to fit a keras model. Im having trouble understanding how numpy's reshape method works and how do i need to make use of it.

My code:

import cv2
import numpy

def loadAndReshape(target, path):
    targetImage = cv2.imread(path)
    targetImage = cv2.cvtColor(targetImage, cv2.COLOR_BGR2RGB)
    targetImage = cv2.resize(targetImage, dsize=(150, 150)) / 255
    targetImage = targetImage.reshape(1, 150, 150, 3).astype('float32')
    numpy.append(target, targetImage)

targetImages = numpy.ndarray((2, 150, 150, 3))
loadAndReshape(targetImages, './/test1.jpg')
loadAndReshape(targetImages, './/test2.jpg')

Reshaping targetImage works without issues but in the end targetImages will still be an empty ndarray. How do i go about outputting the array needed for my model?

Answer 1

The function 'numpy.append' does not work inplace as I think you expect it to. Instead, you can do smth like:

mport cv2
import numpy as np

def loadAndReshape(image_list, path):
    targetImage = cv2.imread(path)
    targetImage = cv2.cvtColor(targetImage, cv2.COLOR_BGR2RGB)
    targetImage = cv2.resize(targetImage, dsize=(150, 150)) / 255
    targetImage = targetImage.reshape(1, 150, 150, 3).astype('float32')
    image_list.append(targetImage)

targetImages = []
loadAndReshape(targetImages, './/test1.jpg')
loadAndReshape(targetImages, './/test2.jpg')
.
.
.
targetImages = np.concatenate(targetImages)

Answer 2

numpy.append return a copy, see https://numpy.org/doc/stable/reference/generated/numpy.append.html

You can try this:

import cv2
import numpy

def loadAndReshape(path):
    targetImage = cv2.imread(path)
    targetImage = cv2.cvtColor(targetImage, cv2.COLOR_BGR2RGB)
    targetImage = cv2.resize(targetImage, dsize=(150, 150)) / 255
    targetImage = targetImage.reshape(1, 150, 150, 3).astype('float32')
    return targetImage

li = []
li.append(loadAndReshape('.//test1.jpg'))
li.append(loadAndReshape('.//test2.jpg'))
targetImages = np.array(li)