Not getting degradation function

[nanmehta]

thanks al ton for the code and for the excellent paper, but going through the code, I was not able to get the code wriiten in degradation.py for building the kernel. Can you help me out please, I am new in this field

[sunreef]

What part of the degradation function is confusing to you ?

The degradation is formed by a convolution with a certain kernel (anisotropic Gaussian in our case) followed by a downsampling operation (taking one pixel out of two in both spatial dimensions).

[nanmehta]

Thanks for your reply sir, actually i am not ab;e to understand how is this being defined in the code. a = cos_theta_2 / sigma_x_2 + sin_theta_2 / sigma_y_2 b = sin_theta * cos_theta * (1.0 / sigma_y_2 - 1.0 / sigma_x_2) c = sin_theta_2 / sigma_x_2 + cos_theta_2 / sigma_y_2 gaussian = lambda x,y: (- ( a * (x ** 2) + 2.0 * b * x * y + c * (y ** 2))).exp()

Sorry sir i am new to this so I am having trouble in understanding , like gaussian kernel as i have known is defined as -exp(x^2)/2(variance^2), so why we are using a,b,c

[sunreef]

This is because we are using 2D anisotropic gaussian kernels in this code. The coefficients depend on the orientation of the kernel as well as the two standard deviations.

[nanmehta]

okay. thanks alot sir