Resize any image to 512 x 512

Question

I am trying to train a stable diffusion model, which receives 512 x 512 images as inputs.

I am downloading the bulk of images from the web, and they have multiple sizes and shapes, and so I need to preprocess them and convert to 512 x 512.

If I take this 2000 × 1434 image:

And I try to resize it, with:

from PIL import Image

# Open an image file
with Image.open("path/to/Image.jpg") as im:
    # Create a thumbnail of the image
    im.thumbnail((512, 512))
    # Save the thumbnail
    im.save("path/to/Image_resized.jpg")

I get this 512 × 367 image:

But I need 512 for both width and height, without distorting the image, like you can achieve on this website:

Birma

Any ideas on how I can achieve this conversion using python?

You have to crop the image. Resize the short dimension (in the example 1434px) to 512px and crop the other dimension 714px by 202px. — phibel, Jan 08 '23 at 15:52
if you answer with a workable code, I'll gladly accept and upvote it. — 8-Bit Borges, Jan 08 '23 at 16:02

phibel · Accepted Answer · 2023-01-08T17:00:53.733

I dont think a resize of an image with another aspect ratio of 1:1 (512:512) is possible without distorting the image. You can resize the shorter dimension of the image to 512px and crop the large dimension to 512px.

from PIL import Image

# Open an image file
with Image.open("image.jpg") as im:
    width, height = im.size
    if width < height:
        newWidth = 512
        newHeight = int(height / (width / 512))
        cropTop = (newHeight - 512) // 2
        cropBottom = cropTop + 512
        crop = (0, cropTop, 512, cropBottom)
    else:
        newHeight = 512
        newWidth = int(width / (height / 512))
        cropLeft = (newWidth - 512) // 2
        cropRight = cropLeft + 512
        crop = (cropLeft, 0, cropRight, 512)
    imResize = im.resize((newWidth, newHeight))
    imCrop = imResize.crop(crop)
    imCrop.save("image_resized_cropped.jpg")

Resize any image to 512 x 512

1 Answers1