Create Keras input array, numpy.append does not wo

问题:

I'm trying to create an input array for a functional Keras model. I have a set of images that I collect in a single np array, so the array has the shape: (nr_images,img_width,img_height,nr_channels)

I use this code:

files = glob.glob ("data/train/part2/*.png")
for myFile in files:
    image = cv2.imread (myFile)
    image=cv2.resize(image,(256,256))
    train.append (image)
train = np.array(train,dtype='float32')
np.save('train',train)

The resulting array dimension is (426, 256, 256, 3). So it seems to work.

But if I look at the images stored in the array by:

image_train=np.load("train.npy")
image=image_train[0]       #Look at the first image
img = Image.fromarray(image,'RGB')
img.show()

I get rubbish:

My Keras results are super bad, so I suspect that it has something to do with the input.

Am I doing something wrong?