Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

wrong result for morphology opening

See original GitHub issue

🐛 Bug

To Reproduce

Steps to reproduce the behavior:

import torch
import numpy as np
from scipy.ndimage.morphology import grey_opening
from kornia.morphology import opening
from skimage.morphology import opening as sk_opening
kernel = np.array([[0, 0, 0, 0, 1, 0, 0, 0, 0],
                   [0, 0, 1, 1, 1, 1, 1, 0, 0],
                   [0, 1, 1, 1, 1, 1, 1, 1, 0],
                   [0, 1, 1, 1, 1, 1, 1, 1, 0],
                   [1, 1, 1, 1, 1, 1, 1, 1, 1],
                   [0, 1, 1, 1, 1, 1, 1, 1, 0],
                   [0, 1, 1, 1, 1, 1, 1, 1, 0],
                   [0, 0, 1, 1, 1, 1, 1, 0, 0],
                   [0, 0, 0, 0, 1, 0, 0, 0, 0]], dtype=np.float32)

input = np.zeros([1, 1, 128, 128])
output_scipy = grey_opening(input[0, 0, :, :], structure=kernel)
output_sk = sk_opening(input[0, 0, :, :], kernel)
output_kornia = opening(torch.tensor(input), torch.tensor(kernel))
print(np.all(output_scipy == output_sk))
print(np.all(output_sk == output_kornia))
print(np.all(output_scipy == output_kornia))

this bug present on master branch, opening is not yet in any release version

Issue Analytics

State:
Created 2 years ago
Comments:8 (3 by maintainers)

Top GitHub Comments

1reaction

Manza12commented, Aug 2, 2021

I think it is solved. But I don’t remeber well what was this about. I am going to check it and I Will answer properly.

De: Edgar @.> Enviado: lunes, 2 de agosto de 2021 14:56 Para: @.> CC: @.>; @.> Asunto: Re: [kornia/kornia] wrong result for morphology opening (#1045)

@Manza12 https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FManza12&data=04|01||3e2d6dd25c6f4abb7f0908d955b4ee2d|84df9e7fe9f640afb435aaaaaaaaaaaa|1|0|637635057835268856|Unknown|TWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D|1000&sdata=UEjEuOkGvjO2ASrQP%2BGw7dl3Dh0R3fPh4BVPB3xDVWA%3D&reserved=0 is this still relevant ?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fkornia%2Fkornia%2Fissues%2F1045%23issuecomment-891002195&data=04|01||3e2d6dd25c6f4abb7f0908d955b4ee2d|84df9e7fe9f640afb435aaaaaaaaaaaa|1|0|637635057835278809|Unknown|TWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D|1000&sdata=dJzK3uIAhkcLnoErr7enX0g8XfbHgmhCZBvUtRMwd%2FA%3D&reserved=0, or unsubscribehttps://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FALAHVFRLDPYYFN2TVUM45BDT22IXLANCNFSM452HSMSA&data=04|01||3e2d6dd25c6f4abb7f0908d955b4ee2d|84df9e7fe9f640afb435aaaaaaaaaaaa|1|0|637635057835278809|Unknown|TWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D|1000&sdata=%2BuEGe4CbK0Vay1GJm0yQhhUy0Vnco6h%2FwsOlSpT60tY%3D&reserved=0.

1reaction

Manza12commented, Jun 1, 2021

This bug is produced by the padding of the border. I will implement a custom border value for all the morphological operations as it is done in nnMorpho. This way you would be able to fully control the border value.

Also, comparing PyTorch tensors with numpy arrays won’t work for np.all; last two lines should be

print(np.all(output_sk == output_kornia.numpy()))
print(np.all(output_scipy == output_kornia.numpy()))

Nevertheless, I checked the images and it is true that they don’t give the same result. When testing with input = np.ones([1, 1, 128, 128]) they do, though