Source code for kornia.color.histogram

from typing import Tuple, List
import torch
import torch.nn as nn


def marginal_pdf(values: torch.Tensor, bins: torch.Tensor, sigma: torch.Tensor,
                 epsilon: float = 1e-10) -> Tuple[torch.Tensor, torch.Tensor]:
    """
    Function that calculates the marginal probability distribution function of the input tensor based on the number
    of histogram bins.
    Args:
                    values: (torch.Tensor), shape [BxNx1]
                    bins: (torch.Tensor), shape [NUM_BINS]
                    sigma: (torch.Tensor), shape [1], gaussian smoothing factor
                    epsilon: (float), scalar, for numerical stability
    Returns:
                    pdf: (torch.Tensor), shape [BxN]
                    kernel_values: (torch.Tensor), shape [BxNxNUM_BINS]"""

    if not torch.is_tensor(values):
        raise TypeError("Input values type is not a torch.Tensor. Got {}"
                        .format(type(values)))
    if not torch.is_tensor(bins):
        raise TypeError("Input bins type is not a torch.Tensor. Got {}"
                        .format(type(bins)))
    if not torch.is_tensor(sigma):
        raise TypeError("Input sigma type is not a torch.Tensor. Got {}"
                        .format(type(sigma)))

    if not values.dim() == 3:
        raise ValueError("Input values must be a of the shape BxNx1."
                         " Got {}".format(values.shape))
    if not bins.dim() == 1:
        raise ValueError("Input bins must be a of the shape NUM_BINS"
                         " Got {}".format(bins.shape))
    if not sigma.dim() == 0:
        raise ValueError("Input sigma must be a of the shape 1"
                         " Got {}".format(sigma.shape))

    residuals = values - bins.unsqueeze(0).unsqueeze(0)
    kernel_values = torch.exp(-0.5 * (residuals / sigma).pow(2))

    pdf = torch.mean(kernel_values, dim=1)
    normalization = torch.sum(pdf, dim=1).unsqueeze(1) + epsilon
    pdf = pdf / normalization

    return (pdf, kernel_values)


def joint_pdf(kernel_values1: torch.Tensor, kernel_values2: torch.Tensor, epsilon: float = 1e-10) -> torch.Tensor:
    """
    Function that calculates the joint probability distribution function of the input tensors based on the number
    of histogram bins.
    Args:
                    kernel_values1: (torch.Tensor), shape [BxNxNUM_BINS]
                    kernel_values2: (torch.Tensor), shape [BxNxNUM_BINS]
                    epsilon: (float), scalar, for numerical stability
    Returns:
                    pdf: (torch.Tensor), shape [BxNUM_BINSxNUM_BINS]"""

    if not torch.is_tensor(kernel_values1):
        raise TypeError("Input kernel_values1 type is not a torch.Tensor. Got {}"
                        .format(type(kernel_values1)))
    if not torch.is_tensor(kernel_values2):
        raise TypeError("Input kernel_values2 type is not a torch.Tensor. Got {}"
                        .format(type(kernel_values2)))

    if not kernel_values1.dim() == 3:
        raise ValueError("Input kernel_values1 must be a of the shape BxN."
                         " Got {}".format(kernel_values1.shape))
    if not kernel_values2.dim() == 3:
        raise ValueError("Input kernel_values2 must be a of the shape BxN."
                         " Got {}".format(kernel_values2.shape))
    if kernel_values1.shape != kernel_values2.shape:
        raise ValueError("Inputs kernel_values1 and kernel_values2 must have the same shape."
                         " Got {} and {}".format(kernel_values1.shape, kernel_values2.shape))

    joint_kernel_values = torch.matmul(kernel_values1.transpose(1, 2), kernel_values2)
    normalization = torch.sum(joint_kernel_values, dim=(1, 2)).view(-1, 1, 1) + epsilon
    pdf = joint_kernel_values / normalization

    return pdf


[docs]def histogram(x: torch.Tensor, bins: torch.Tensor, bandwidth: torch.Tensor, epsilon: float = 1e-10) -> torch.Tensor: """ Function that estimates the histogram of the input tensor. The calculation uses kernel density estimation which requires a bandwidth (smoothing) parameter. Args: x: (torch.Tensor), shape [BxN] bins: (torch.Tensor), shape [NUM_BINS] bandwidth: (torch.Tensor), shape [1], gaussian smoothing factor epsilon: (float), scalar, for numerical stability Returns: pdf: (torch.Tensor), shape [BxNUM_BINS]""" pdf, _ = marginal_pdf(x.unsqueeze(2), bins, bandwidth, epsilon) return pdf
[docs]def histogram2d( x1: torch.Tensor, x2: torch.Tensor, bins: torch.Tensor, bandwidth: torch.Tensor, epsilon: float = 1e-10) -> torch.Tensor: """ Function that estimates the histogram of the input tensor. The calculation uses kernel density estimation which requires a bandwidth (smoothing) parameter. Args: x1: (torch.Tensor), shape [BxN1] x2: (torch.Tensor), shape [BxN2] bins: (torch.Tensor), shape [NUM_BINS] bandwidth: (torch.Tensor), scalar, gaussian smoothing factor epsilon: (float), scalar, for numerical stability Returns: pdf: (torch.Tensor), shape [BxNUM_BINSxNUM_BINS]""" pdf1, kernel_values1 = marginal_pdf(x1.unsqueeze(2), bins, bandwidth, epsilon) pdf2, kernel_values2 = marginal_pdf(x2.unsqueeze(2), bins, bandwidth, epsilon) pdf = joint_pdf(kernel_values1, kernel_values2) return pdf