eadcat / apap-image-stitching Goto Github PK

I am working on multiple image stitching where my image is images taken from drone and I need to stitch to create a map.
However, most of the implementations i saw can only stitch dominantly in one orientation.

Hi, it's a nice code.
I am a beginner who is learning image-stitching.
I don't know how to use the code to calculate the RMSE of image1 and image2.
Could you please write it?
Thanks you very much.

Hi, it's a nice code.
I am a beginner who is learning image-stitching.
I don't know how to use the code to calculate the RMSE of image1 and image2.
Could you please write it?
Thanks you very much.

Thanks for your work! when I run the code in some samples, it will get the error: 'ValueError: Sample larger than population or is negative'. The last code executed is ‘final_src, final_dst = ransac.thread(src_match, dst_match, self.opt.ransac_max)’. Could you please give me a solution? Thanks you very much.

Does this work？How can I get him running？

hello, I have a questions about warp_local

for i in tqdm(range(self.final_height)) if progress else range(self.final_height): m = np.where(i < mesh_h)[0][0] for j in range(self.final_width): n = np.where(j < mesh_w)[0][0] homography = np.linalg.inv(local_homography[m-1, n-1, :]) x, y = j - self.offset_x, i - self.offset_y source_pts = np.array([x, y, 1]) target_pts = self.warp_coordinate_estimate(source_pts, homography) if 0 < target_pts[0] < ori_w and 0 < target_pts[1] < ori_h: warped_img[i, j, :] = ori_img[int(target_pts[1]), int(target_pts[0]), :]

The local_h is obtained by meshing the src, but in the end, it is obtained by reverse warp the mesh of the image of the large canvas, 1. Is this my misunderstanding? 2. Is this reasonable if I understand correctly?

i am read your excellent code of APAP. but i found the (std_x and std_y) are the same in the red box I painted。could u please tell
me sth about that? thanks

Hello! Thanks for your work!

 weight = np.exp(-(np.sqrt(dist[:, 0] ** 2 + dist[:, 1] ** 2) * inverse_sigma))

Is there an extra np.sqrt() here?