Gaussian filtering in pytorch

gaussian_filter_pytorch.py

import torch
import matplotlib.pyplot as plt

inp = torch.FloatTensor(1,1,32,32).uniform_(-1,1)
plt.imshow(inp[0,0])
plt.show()

# noyau gaussien
gauss_ker_7 = torch.FloatTensor(1,1,7,7)
x,y = torch.meshgrid(torch.linspace(-3,3,7), torch.linspace(-3,3,7), indexing='xy')
gauss_ker_7[0,0] = torch.exp(-0.5*((x**2 + y**2)))
gauss_ker_7 /= gauss_ker_7.sum()

# filtrage sans padding
inp_filt = torch.nn.functional.conv2d(inp, gauss_ker_7)
plt.imshow(inp_filt[0,0])
plt.show()

# filtrage avec padding
inp_pad_filt = torch.nn.functional.conv2d(
        torch.nn.functional.pad(inp, (3,3,3,3), mode='reflect'),
        gauss_ker_7)

plt.imshow(inp_pad_filt[0,0])
plt.show()