Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Image.from_array() fails on one of two seemingly identical arrays #583

nek11 opened this issue Jul 19, 2022 · 1 comment

Image.from_array() fails on one of two seemingly identical arrays #583

nek11 opened this issue Jul 19, 2022 · 1 comment


Copy link

nek11 commented Jul 19, 2022

I have this notebook:

import cv2
import numpy as np
import matplotlib.pyplot as plt
import glob
from wand.image import Image
class BipedStitcher:
def _init_(self):
        self._width = 270
        self._height = 480
        M_01 = np.array([[ 9.73331494e-01, -8.01939095e-01,  2.23343892e+01],
         [ 2.72445372e-02,  5.90069057e-02,  6.44154273e+02],
         [ 2.28304381e-05, -1.25424366e-03,  1.00000000e+00]])
        M_21 = np.array([[ 1.58019986e+01,  1.35065361e+01, -9.44831124e+03],
         [-8.03741771e-02,  1.67267572e+01, -1.08396169e+04],
         [-3.94436455e-05,  2.12285957e-02,  1.00000000e+00]])

        S = np.array([
            [self._width / 720, 0, 0],
            [0, self._width / 720, 0],
            [0, 0, 1]

        self._matrix_0 =
        self._matrix_2 =
        K = np.array([
            [239.417, 0.0, 239.291],
            [0.0, 239.417, 128.642],
            [0.0, 0.0, 1.0]

        num, R, T, N  = cv2.decomposeHomographyMat(M_01, K)
        self._canvas_w, self._canvas_h = 10 * self._width, 10 * self._height

        x_offset = self._canvas_w // 2 - self._height // 2
        y_offset = self._canvas_h // 2
        self._translation_matrix = np.array([ # translation matrix
        ], dtype=np.float32)
    def _overlay_images(self, left_warped, middle, right_warped):
        # Select all non-black pixels
        left_mask = np.float32(np.any(left_warped != [0, 0, 0], axis=-1))
        middle_mask = np.float32(np.any(middle != [0, 0, 0], axis=-1))
        right_mask = np.float32(np.any(right_warped != [0, 0, 0], axis=-1))
        # XOR mask in the middle
        xor_mask_left_mid = cv2.bitwise_xor(left_mask, middle_mask)
        xor_mask_final = cv2.bitwise_xor(xor_mask_left_mid, right_mask).astype(int)
        final_mask = np.stack([xor_mask_final, xor_mask_final, xor_mask_final], axis=2)
        # Stitch the three images
        stitched_image = left_warped*final_mask + right_warped*final_mask + middle
        stitched_image = np.rot90(stitched_image, k=1)
        return stitched_image
    def _undistort_image(self, stitched_image):
        # Cut image to center it and transform to int for wand
        stitched_image_cut = (stitched_image[:, 270:, :]).astype(np.uint8)
        cv2.imwrite("temp_save.png", stitched_image_cut)
        array = cv2.imread("temp_save.png")
        print(np.array_equal(stitched_image_cut, array))
        with Image.from_array(np.copy(array)) as wand_image:
            wand_image.virtual_pixel = 'transparent'
            wand_image.distort('barrel_inverse', (0.0, 0.0, -2.5, 4.5))
            wand_image.distort('barrel', (1, 0.0, 0.0, 1.5))
            undistorted_image = np.array(wand_image)
        cv2.imwrite("loaded_from_cv2.png", undistorted_image)
        with Image.from_array(stitched_image_cut) as wand_image_2:
            wand_image_2.virtual_pixel = 'transparent'
            wand_image_2.distort('barrel_inverse', (0.0, 0.0, -2.5, 4.5))
            wand_image_2.distort('barrel', (1, 0.0, 0.0, 1.5))
            undistorted_image_2 = np.array(wand_image_2)
        cv2.imwrite("loaded_from_array.png", undistorted_image_2)
        return undistorted_image
    def run(self, left_image, middle_image, right_image):
        warped_img_right = cv2.warpPerspective(right_image,, dsize=(self._canvas_w, self._canvas_h), borderMode=cv2.BORDER_CONSTANT, borderValue=(0,0,0))
        translated_img_middle = cv2.warpPerspective(middle_image, self._translation_matrix, dsize=(self._canvas_w, self._canvas_h), borderMode=cv2.BORDER_CONSTANT, borderValue=(0,0,0))
        warped_img_left = cv2.warpPerspective(left_image,, dsize=(self._canvas_w, self._canvas_h), borderMode=cv2.BORDER_CONSTANT, borderValue=(0,0,0))
        stitched_image = self._overlay_images(warped_img_left, translated_img_middle, warped_img_right)
        undistorted_image = self._undistort_image(stitched_image)
left_image = cv2.imread("test/left.jpg", cv2.COLOR_RGB2GRAY)
middle_image = cv2.imread("test/middle.jpg", cv2.COLOR_RGB2GRAY)
right_image = cv2.imread("test/right.jpg", cv2.COLOR_RGB2GRAY)

stitcher = BipedStitcher(), middle_image, right_image)

For which you will need the zip file I have attached.

Now, a very weird output is produced:
In line 65, I save an array called "stitched_image_cut" with cv2 as "temp_save.png", then load it again with cv2 just after (named "array"), and test that both arrays are equal, which they are.

Then, I use the "Image.from_array" function on the loaded from cv2 array "array", apply some transformations and save it as "loaded_from_cv2.png". This is the result:

Just after, I do the same thing but this time with "stitched_image_cut", which is supposed to be exactly the same array as "array". This time, this is what I get:

Can anyone tell me what in the world is going on?

Copy link

emcconville commented Jul 19, 2022

Looks like a bug with stride handling. Although the ultimate data is the same, one data-buffer is laid out in memory with K order, and the other as 'C' order.

Short fix

Just set the order when copying data (np.copy(img, order='C'))

with Image.from_array(np.copy(array, order='C')) as wand_image:
with Image.from_array(np.copy(stitched_image_cut, order='C')) as wand_image_2:

@emcconville emcconville self-assigned this Jul 19, 2022
@emcconville emcconville added this to the Wand 0.6.9 milestone Jul 19, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
None yet

No branches or pull requests

2 participants