Use same color map for different images to more accurately compare them #68

New Issue

Benjamin_Loison · 2024-05-09T23:18:36+02:00

Benjamin_Loison commented

2024-05-09 23:18:36 +02:00

help(plt.imsave)

...
    vmin, vmax : float, optional
        *vmin* and *vmax* set the color scaling for the image by fixing the
        values that map to the colormap color limits. If either *vmin*
        or *vmax* is None, that limit is determined from the *arr*
        min/max value.
    cmap : str or `~matplotlib.colors.Colormap`, default: :rc:`image.cmap`
        A Colormap instance or registered colormap name. The colormap
        maps scalar data to colors. It is ignored for RGB(A) data.
...

```python help(plt.imsave) ``` ``` ... vmin, vmax : float, optional *vmin* and *vmax* set the color scaling for the image by fixing the values that map to the colormap color limits. If either *vmin* or *vmax* is None, that limit is determined from the *arr* min/max value. cmap : str or `~matplotlib.colors.Colormap`, default: :rc:`image.cmap` A Colormap instance or registered colormap name. The colormap maps scalar data to colors. It is ignored for RGB(A) data. ... ```

Benjamin_Loison added the

enhancement

low priority

quick

labels 2024-05-09 23:18:36 +02:00

Benjamin_Loison commented

2024-05-09 23:28:47 +02:00

plt.imsave(PREFIX + 'multiple_colors.png', multipleColorsImage, vmin = multipleColorsImage.min(), vmax = multipleColorsImage.max())

is identical (according to diff) to:

plt.imsave(PREFIX + 'multiple_colors.png', multipleColorsImage)

```python plt.imsave(PREFIX + 'multiple_colors.png', multipleColorsImage, vmin = multipleColorsImage.min(), vmax = multipleColorsImage.max()) ``` is identical (according to `diff`) to: ```python plt.imsave(PREFIX + 'multiple_colors.png', multipleColorsImage) ```

Benjamin_Loison commented

2024-05-09 23:51:45 +02:00

denoiser = bilateral np.min(multipleColorsImage) = -0.04890543970650529 np.max(multipleColorsImage) = 0.047219058067642594
denoiser = wavelet np.min(multipleColorsImage) = -0.05360332845280251 np.max(multipleColorsImage) = 0.03556257229735231
denoiser = mean np.min(multipleColorsImage) = -69.4864864864864 np.max(multipleColorsImage) = 36.67181467181476

If use the same color scale, then only mean is interesting as it enforces its own color scale and 2 other images are monocolor.

Mean:

Bilateral:

When remove mean denoiser:

Bilateral:

Wavelet:

``` denoiser = bilateral np.min(multipleColorsImage) = -0.04890543970650529 np.max(multipleColorsImage) = 0.047219058067642594 denoiser = wavelet np.min(multipleColorsImage) = -0.05360332845280251 np.max(multipleColorsImage) = 0.03556257229735231 denoiser = mean np.min(multipleColorsImage) = -69.4864864864864 np.max(multipleColorsImage) = 36.67181467181476 ``` If use the same color scale, then only `mean` is interesting as it enforces its own color scale and 2 other images are monocolor. Mean: ![mean_rafael_230424_mean_multiple_colors](/attachments/2823f3c5-b68f-4d4b-8035-d04bc1ff1794) Bilateral: ![mean_rafael_230424_bilateral_multiple_colors](/attachments/366a53f3-bd3a-479d-9682-540c8de9e242) When remove mean denoiser: Bilateral: ![mean_rafael_230424_bilateral_multiple_colors](/attachments/78fe2daf-0458-474c-b033-09e477097906) Wavelet: ![mean_rafael_230424_wavelet_multiple_colors](/attachments/04e13889-9ae0-413e-b5df-0c53ece15cfc)

mean_rafael_230424_mean_multiple_colors.png

50 MiB

mean_rafael_230424_bilateral_multiple_colors.png

102 KiB

mean_rafael_230424_wavelet_multiple_colors.png

36 MiB

mean_rafael_230424_bilateral_multiple_colors.png

36 MiB

Benjamin_Loison referenced this issue from a commit

2024-05-10 00:03:13 +02:00

Fix #68: Use same color map for different images to more accurately compare them

Benjamin_Loison closed this issue

2024-05-10 00:03:13 +02:00

Benjamin_Loison commented

2024-05-10 00:27:53 +02:00

![prnu_estimate_distribution_of_values_for_rafael_230424_images](/attachments/52c1f5a5-cdad-4f9a-9be3-d3469cf4921a)

prnu_estimate_distribution_of_values_for_rafael_230424_images.png

47 KiB

Benjamin_Loison commented

2024-05-10 02:25:41 +02:00

Should make similar plot for RAISE flat-field to maybe understand why Rafael images are absorbing in #63.

Why have such an outlier for mean blue? Maybe because of borders? I would not say so as they do not behave differently as far as I understand. Maybe because of point at (644, 1200)? It seems to:

import numpy as np
import matplotlib.pyplot as plt

fileName = 'means/mean_flat-field_NEF_mean_blue'
npArray = np.load(f'{fileName}.npy')

plt.title('Distribution of PRNU estimation of RAISE flat-field raw blue channel')
plt.hist(npArray.flatten(), 'auto')
plt.show()

>>> index = -1; np.partition(npArray.flatten(), index)[index]
5042.010000000002

>>> index = -2; np.partition(npArray.flatten(), index)[index]
423.28999999999934

So the first and second maximum are 5042 and 423.29.

Should make similar plot for RAISE flat-field to maybe understand why Rafael images are absorbing in #63. ![prnu_estimate_distribution_of_values_for_raise_flat-field_images](/attachments/bac0a3cc-d60c-4a89-9ae5-1e6036147b40) ![prnu_estimate_distribution_of_values_for_raise_flat-field_images](/attachments/b1b46f64-d5e9-44c2-bff4-53e31087dccd) ![Figure_1](/attachments/44261368-e522-4aca-b3a9-4e1bf1ff3869) Why have such an outlier for mean blue? Maybe because of borders? I would not say so as they do not behave differently as far as I understand. Maybe because of point at (644, 1200)? It seems to: ![image](/attachments/d3f797f4-a361-443c-8c88-0b70e7184664) ![image](/attachments/946eacdc-a2f2-43fa-aaca-76601a930592) ```python import numpy as np import matplotlib.pyplot as plt fileName = 'means/mean_flat-field_NEF_mean_blue' npArray = np.load(f'{fileName}.npy') plt.title('Distribution of PRNU estimation of RAISE flat-field raw blue channel') plt.hist(npArray.flatten(), 'auto') plt.show() ``` ![distribution_of_prnu_estimation_of_raise_flat-field_raw_blue_channel](/attachments/ff5fa212-5673-4b12-b66e-c5cdcf8a962c) ![distribution_of_prnu_estimation_of_raise_flat-field_raw_blue_channel](/attachments/1109ecfc-0a42-409f-af27-5e59e9b7b181) ```pycon >>> index = -1; np.partition(npArray.flatten(), index)[index] 5042.010000000002 >>> index = -2; np.partition(npArray.flatten(), index)[index] 423.28999999999934 ``` So the first and second maximum are 5042 and 423.29.

prnu_estimate_distribution_of_values_for_raise_flat-field_images.png

46 KiB

image.png

176 KiB

image.png

199 KiB

distribution_of_prnu_estimation_of_raise_flat-field_raw_blue_channel.png

24 KiB

distribution_of_prnu_estimation_of_raise_flat-field_raw_blue_channel.png

20 KiB

distribution_of_prnu_estimation_of_raise_flat-field_raw_blue_channel.png

49 KiB

Figure_1.png

46 KiB

Benjamin_Loison commented

2024-05-12 02:54:58 +02:00

Could be interesting to just add a line with raw images.

Benjamin_Loison commented

2024-05-12 03:10:39 +02:00

>>> index = -100; np.partition(npArray.flatten(), index)[index]
27.28
>>> index = 100; np.partition(npArray.flatten(), index)[index]
-26.35

So width of 53.63 and compared to other denoisers width of about 0.1, the width is 563 times wider...

```pycon >>> index = -100; np.partition(npArray.flatten(), index)[index] 27.28 >>> index = 100; np.partition(npArray.flatten(), index)[index] -26.35 ``` So width of 53.63 and compared to other denoisers width of about 0.1, the width is 563 times wider...

Benjamin_Loison commented

2024-05-12 03:15:38 +02:00

Could locate such maximums to potentially see visually other outliers.

>>> np.unravel_index(npArray.argmax(), npArray.shape)
(1200, 644)
>>> index = -0; np.where(npArray == np.partition(npArray.flatten(), index)[index])
(array([1202]), array([643]))
>>> index = -1; np.array(np.where(npArray == np.partition(npArray.flatten(), index)[index])[::-1]) * 2
array([[1288],
       [2400]])

They are not clearly visible as the first maximum, this is explainable as the intensity is 133 times less.

Could locate such maximums to potentially see visually other outliers. ```pycon >>> np.unravel_index(npArray.argmax(), npArray.shape) (1200, 644) >>> index = -0; np.where(npArray == np.partition(npArray.flatten(), index)[index]) (array([1202]), array([643])) >>> index = -1; np.array(np.where(npArray == np.partition(npArray.flatten(), index)[index])[::-1]) * 2 array([[1288], [2400]]) ``` They are not clearly visible as the first maximum, this is explainable as the intensity is 133 times less.

Benjamin_Loison commented

2024-05-12 03:45:34 +02:00

Related to Benjamin_Loison/gimp/issues/29.

Related to [Benjamin_Loison/gimp/issues/29](https://gitlab.gnome.org/Benjamin_Loison/gimp/-/issues/29).

Benjamin_Loison commented

2024-05-13 01:43:33 +02:00

Could compute proportion of pixels for mean outside largest other denoisers band.

But it may be a significant proportion exceeding not much, so should maybe just plot distribution of pixel values and zoom on the largest other denoisers band.

Could compute proportion of pixels for mean outside largest other denoisers band. But it may be a significant proportion exceeding not much, so should maybe just plot distribution of pixel values and zoom on the largest other denoisers band.

Sign in to join this conversation.

No Label

bug

Context-Adaptive Interpolator