How to get the NPR images?

[Phoebe-cap]

Hi, first of all, excellent work! According to my understanding, the input of the training classifier is NPR images. However, do you have the code to get the NPR images? I didn't find it.

[chuangchuangtan]

We use the difference between an image and its resampled version as NPR. NPR = x - self.interpolate(x, 0.5)

Specifically, after one downsampling and one nearest-neighbor upsampling, the 2x2 local pixels will replicate the value of the top-left corner code.

    def interpolate(self, img, factor):
        return F.interpolate(F.interpolate(img, scale_factor=factor, mode='nearest', recompute_scale_factor=True), \
                   scale_factor=1/factor, mode='nearest', recompute_scale_factor=True)

For example

Image:
|  1.1   |  1.2   |  2.1   |  2.2   |
|  1.3   |  1.4   |  2.3   |  2.4   |
|  3.1   |  3.2   |  4.1   |  4.2   |
|  3.3   |  3.4   |  4.3   |  4.4   |

downsampling
|  1.1   |  2.1   | 
|  3.1   |  4.1   | 

nearest-neighbor upsampling  2x
|  1.1   |  1.1   |  2.1   |  2.1   |
|  1.1   |  1.1   |  2.1   |  2.1   |
|  3.1   |  3.1   |  4.1   |  4.1   |
|  3.1   |  3.1   |  4.1   |  4.1   |

Image - upsampling ( downsampling(Image) ):

|  1.1 - 1.1  |  1.2 - 1.1   |  2.1 - 2.1   |  2.2 - 2.1   |
|  1.3 - 1.1  |  1.4 - 1.1   |  2.3 - 2.1   |  2.4 - 2.1   |
|  3.1 - 3.1  |  3.2 - 3.1   |  4.1 - 4.1   |  4.2 - 4.1   |
|  3.3 - 3.1  |  3.4 - 3.1   |  4.3 - 4.1   |  4.4 - 4.1   |

[chuangchuangtan]

Evaluation code

import numpy as np
from torch.nn import functional as F
import torch

image = torch.from_numpy(np.array([[1.1, 1.2, 2.1, 2.2], \
                                   [1.3, 1.4, 2.3, 2.4], \
                                   [3.1, 3.2, 4.1, 4.2], \
                                   [3.3, 3.4, 4.3, 4.4]])).reshape([1, 1, 4, 4])
downsampling_image = F.interpolate(image, scale_factor=0.5, mode='nearest', recompute_scale_factor=True)
upsampling_downsampling_image = F.interpolate(downsampling_image, scale_factor=2, mode='nearest', recompute_scale_factor=True)

[Phoebe-cap]

OK, thank you very much.