Today, I came across a weird thing that I cannot explain and cannot find an answer to. Maybe I have some fundamental misunderstanding and I am not seeing the obvious, but I guess I am at the point where I will bite the bullet of being called out over now knowing what is going on.
I am loading some image data using PIL and converting it into a numpy array. Nothing fancy here. However, I am then changing the data type from integer to float so I can normalize the RGB channels in a later step.
I would expect that changing the datatype from int to float wouldn't really change the image in any way. However ...
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image
im = Image.open("test.jpg")
before the dtype change:
plt.imshow(np.array(im))
after the dtype change:
plt.imshow(np.array(im).astype(float))
Is this what one would expect? If so ... why? What am I missing?
Cheers & thanks in advance!!
If you check the docs you'll see that imshow expects:
(M, N): an image with scalar data. The values are mapped to colors using normalization and a colormap. See parameters norm, cmap, vmin, vmax.
(M, N, 3): an image with RGB values (0-1 float or 0-255 int).
So when the are of dtype float, it expects them to be in the range 0-1, hence we have to divide it by 255 to get the expected result, otherwise you will see the negative of the image.
Here's an example using a sample image from sklearn.datasets:
from sklearn.datasets import load_sample_images
dataset = load_sample_images()
first_img_data = dataset.images[0]
plt.imshow(first_img_data)
In order to limit the values in the range 0-1, we just have to divide by 255:
plt.imshow(first_img_data.astype(float)/255)
