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.

How to properly use kornia.enhance.histogram?

See original GitHub issue

🐛 Bug

hi,

i am comparing kornia.enhance.histogram to numpy.histogram and torch.histc. the last 2 give the same result, but kornia’s histogram is different and unexpected…
also, in the documentation, it is not clear what is the impact of the bandwidth on the histogram? could you please make it clear. does it have something to do with the histogram precision. if yes, how (low/high values)?

To Reproduce

all the 3 methods are required to produce histogram within the same range.

Steps to reproduce the behavior:

import numpy as np
import kornia
from kornia.enhance import histogram

import torch


if __name__ == '__main__':
    print('torch:', torch.__version__)
    print('numpy:', np.__version__)
    print('kornia:', kornia.__version__)

    seed = 0
    torch.manual_seed(seed)
    np.random.seed(seed)
    torch.cuda.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)

    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    nbins = 256
    a = torch.rand((1, 5))
    print(a)
    anp = a.numpy()
    nphist, bins = np.histogram(anp, bins=nbins, range=(anp.min(), anp.max()))
    print('np hist', nphist)

    a = a.to(device)
    thistc = torch.histc(a, bins=nbins, min=a.min(), max=a.max())
    print('torch hist', thistc)

    _bins = torch.linspace(a.min(), a.max(), 256).to(device)
    bandwith = torch.tensor(.9).to(device)

    khist = histogram(x=a, bins=_bins,
                      bandwidth=bandwith)
    print('kornia hist', khist[0])

    print('dif torch numpy: ', torch.abs(thistc - torch.from_numpy(nphist).to(
        device)).sum())

Expected behavior

i expect the results to be close to torch and numpy and to be consistent. the obtained histogram is not. all bins have values…

Environment

Collecting environment information...
PyTorch version: 1.9.0
Is debug build: False
CUDA used to build PyTorch: 11.1
ROCM used to build PyTorch: N/A

OS: Ubuntu 18.04.5 LTS (x86_64)
GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.10

Python version: 3.7.10 (default, Feb 26 2021, 18:47:35)  [GCC 7.3.0] (64-bit runtime)
Python platform: Linux-4.15.0-122-generic-x86_64-with-debian-buster-sid
Is CUDA available: True
CUDA runtime version: Could not collect
GPU models and configuration: 
GPU 0: Tesla P100-PCIE-16GB
GPU 1: Tesla P100-PCIE-16GB

Nvidia driver version: 455.32.00
cuDNN version: /usr/lib/x86_64-linux-gnu/libcudnn.so.7.4.2
HIP runtime version: N/A
MIOpen runtime version: N/A

Versions of relevant libraries:
[pip3] efficientnet-pytorch==0.7.0
[pip3] numpy==1.20.1
[pip3] torch==1.9.0
[pip3] torchvision==0.10.0
[conda] Could not collect

Additional context

Obtained results:

torch: 1.9.0
numpy: 1.20.1
kornia: 0.5.5+9a70e35
tensor([[0.4963, 0.7682, 0.0885, 0.1320, 0.3074]])
np hist [1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1]
torch hist tensor([1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 1.], device='cuda:0')
kornia hist tensor([0.0038, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039,
        0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039,
        0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039,
        0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039,
        0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0040,
        0.0040, 0.0040, 0.0040, 0.0040, 0.0040, 0.0039, 0.0039, 0.0039, 0.0039,
        0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039,
        0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039,
        0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039,
        0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039, 0.0039,
        0.0039, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038,
        0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038,
        0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038, 0.0038,
        0.0038, 0.0038, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037,
        0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037,
        0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0037, 0.0036,
        0.0036, 0.0036, 0.0036, 0.0036], device='cuda:0')
dif torch numpy:  tensor(0., device='cuda:0')

thanks

Issue Analytics

State:
Created 2 years ago
Comments:12 (12 by maintainers)

Top GitHub Comments

1reaction

edgarribacommented, Sep 23, 2021

@bsuleymanov let’s do that - feel free to propose a PR

0reactions

connorlee77commented, Oct 11, 2021

@bsuleymanov thanks for the clarification. /cc @connorlee77 now I recall that you proposed the original implementation - could you give some tips here and how eventually we match (if possible) numpy/torch implementations ? /cc @ducha-aiki @lferraz any opinions here ?

Yeah, the kornia histogram function I implemented is a differentiable approximate of a histogram. It won’t match the numpy or torch implementation exactly. I believe bandwidth is the smoothing parameter (higher value = less noise) of the gaussian kernel used in the kernel density estimation process.

Top Results From Across the Web

kornia.enhance.histogram

enhance.histogram. from typing import Optional, Tuple import torch ...

Image Augmentation for Deep Learning using Keras and ...

In this post we'll be looking at three image augmentation techniques for improving contrast in images. These approaches are sometimes also referred to...

A survey on Kornia: an Open Source Differentiable Computer ...

We have introduced Kornia, a library for computer vision in PyTorch that implements traditional vision algorithms in a differentiable manner making use of ......

Enhance.histogram input and output? - kornia - PyTorch Forums

I can't wrap my head around kornia.enhance.histogram. I want to use it to compare histograms of two images in a differentiable way.

Image histogram and equalizations techniques - Google Colab

on the lightness with Kornia ... We use kornina.enhance.normalize_min_max to normalize the image Luminance. Note: compare the histrograms with the original image.