QUAD Deformation

[ceccopierangiolieugenio]

I hope someone would be able to answer this because I could not find any info about this.
(nor chat-gpt)

I am trying to perform this quad (or mesh) deformation:

but when I try to use the QUAD transform operation using the 4 dest points as shown in the picture

from PIL import Image

img = Image.open('img5.png')

out = img.transform((800,600), Image.Transform.QUAD, [50, 100, 100, 500, 600, 400, 700, 50])

out.show()

I got this:

That looks like the opposite result than the one I wanted to achieve:

Is it a way to define the initial quad and the end quad?
Or to perform a deformation as shown i the first picture?
Or a way to transform the 4 destination point in a way to deform the mesh as in the first image?

Thanks!!!

[radarhere]

Hi. Have you seen https://pillow.readthedocs.io/en/stable/reference/ImageTransform.html? It explains that the parameter you're passing in is

An 8-tuple (x0, y0, x1, y1, x2, y2, x3, y3) which contain the upper left, lower left, lower right, and upper right corner of the source quadrilateral.

The key word there for this discussion is 'source'. Your expectation was that it was the destination quadrilateral.

As far as achieving your effect, could you upload a copy of your initial image?

[ceccopierangiolieugenio]

This is the image i made for reference,

but you can use any image, I spent many hours tweaking with the Perspective,Quad,Mesh deforation and I could not find a way to define the initial and final deformation.

Thanks

[radarhere]

To achieve your effect, I found code on StackOverflow. You can find a copy at https://stackoverflow.com/questions/42827978/image-perspective-transform-using-pillow

import numpy
from PIL import Image

def find_coeffs(pa, pb):
    matrix = []
    for p1, p2 in zip(pa, pb):
        matrix.append([p1[0], p1[1], 1, 0, 0, 0, -p2[0]*p1[0], -p2[0]*p1[1]])
        matrix.append([0, 0, 0, p1[0], p1[1], 1, -p2[1]*p1[0], -p2[1]*p1[1]])

    A = numpy.matrix(matrix, dtype=float)
    B = numpy.array(pb).reshape(8)

    res = numpy.dot(numpy.linalg.inv(A.T * A) * A.T, B)
    return numpy.array(res).reshape(8)

with Image.open("input.png") as im:
    coeffs = find_coeffs(
        [(50, 100), (100, 500), (600, 400), (650, 50)],
        [(0, 0), (im.width, 0), (im.width, im.height), (0, im.height)]
    )
    im = im.transform(im.size, Image.PERSPECTIVE, coeffs)
im.save("output.png")

With this, and your input image, I get

[ceccopierangiolieugenio]

That's amazing,
I spent so much time trying to find this solution.
Thanks!!!

[ceccopierangiolieugenio]

It works SuperWell!!!

Screen.Recording.2025-05-15.at.16.08.57.mov