Warp an image to cylindrical coordinates for cylindrical panorama stitching, using Python OpenCV

cylindricalWarping.py

import cv2
import numpy as np

def cylindricalWarp(img, K):
    """This function returns the cylindrical warp for a given image and intrinsics matrix K"""
    h_,w_ = img.shape[:2]
    # pixel coordinates
    y_i, x_i = np.indices((h_,w_))
    X = np.stack([x_i,y_i,np.ones_like(x_i)],axis=-1).reshape(h_*w_,3) # to homog
    Kinv = np.linalg.inv(K) 
    X = Kinv.dot(X.T).T # normalized coords
    # calculate cylindrical coords (sin\theta, h, cos\theta)
    A = np.stack([np.sin(X[:,0]),X[:,1],np.cos(X[:,0])],axis=-1).reshape(w_*h_,3)
    B = K.dot(A.T).T # project back to image-pixels plane
    # back from homog coords
    B = B[:,:-1] / B[:,[-1]]
    # make sure warp coords only within image bounds
    B[(B[:,0] < 0) | (B[:,0] >= w_) | (B[:,1] < 0) | (B[:,1] >= h_)] = -1
    B = B.reshape(h_,w_,-1)
    
    img_rgba = cv2.cvtColor(img,cv2.COLOR_BGR2BGRA) # for transparent borders...
    # warp the image according to cylindrical coords
    return cv2.remap(img_rgba, B[:,:,0].astype(np.float32), B[:,:,1].astype(np.float32), cv2.INTER_AREA, borderMode=cv2.BORDER_TRANSPARENT)
  
if __name__ == '__main__':
  img = cv2.imread("image.png")
  h, w = img.shape[:2]
  K = np.array([[800,0,w/2],[0,800,h/2],[0,0,1]]) # mock intrinsics
  img_cyl = cylindricalWarp(img, K)
  cv2.imwrite("image_cyl.png", img_cyl)

Any simple way that you recommend, to turn cylindrical warping to transverse-cylindrical warping?

@thennk can't see a simple way, but line 13 https://gist.github.com/royshil/0b21e8e7c6c1f46a16db66c384742b2b#file-cylindricalwarping-py-L13 is where you'd need to make a change

Any simple way that you recommend, to turn cylindrical warping to transverse-cylindrical warping?

@thennk can't see a simple way, but line 13 https://gist.github.com/royshil/0b21e8e7c6c1f46a16db66c384742b2b#file-cylindricalwarping-py-L13 is where you'd need to make a change

Yes, thank you.

For future reference, for transverse cylindrical propjection (wrapping around cylinder flipped by 90), line 13 should be:

A = np.stack([X[:,0],np.sin(X[:,1]), np.cos(X[:,1])],axis=-1).reshape(w_*h_,3)

Thanks for this. For future reference, to wrap an image with a transparent background:
img = cv2.imread("image.png", cv2.IMREAD_UNCHANGED)

I really want reverse method, cylinderical to flat image. Is there any way?

@hanish3464 you can invert the mapping B and use that for cv2.remap. here's an answer on stackoverflow with an implementation for how to do that: https://stackoverflow.com/a/65566295

Did not work. Only outputted the same image without the outer pixels. I used this image:

And got this: