Hi Laurynas, What version of scikit-image are you using? threshold_otsu should work with arrays of any dimensionality, including 1D arrays: In [2]: filters.threshold_otsu(np.random.random((11746,))) Out[2]: 0.4980217396247033 Juan.

On 19 Dec 2020, at 1:37 am, Laurynas Gedminas wrote:

Hey, the suggested code reshapes the original array from (171, 169) to (11746,) and does not work in otsu threshold

Thank you, Laurynas

On Thu, Dec 17, 2020 at 9:56 PM Juan Nunez-Iglesias mailto:jni@fastmail.com> wrote: Some methods, like Li, have some special handling for nan values, but that is not uniform across all our methods, as you have seen. 😂 I would suggest:

```python t = threshold_otsu(image[np.isfinite(image)]) ```

That will only pass the non-nan, non-inf pixels to Otsu to do the calculation.

Hope this helps!

Juan.

...

On 18 Dec 2020, at 7:19 am, Laurynas Gedminas mailto:laurynasgedminas@gmail.com> wrote:

Hello, I have NDVI image, its not a square, and has some None values, I am trying to create otsu threshold, but getting error: autodetected range of [nan, nan] is not finite

how to go around that?

Thank you,

Yes, you need to use `t = threshold_otsu(image[np.isfinite(image)])` in order to get rid of the nan values. Then it should work: In [1]: image = np.random.random((100, 100)) In [2]: image_nan = np.pad(image, 20, mode='constant', constant_values=np.nan) In [3]: filters.threshold_otsu(image_nan[np.isfinite(image_nan)]) Out[3]: 0.501983630941274

On 22 Dec 2020, at 12:56 am, Laurynas Gedminas wrote:

Hello Juan,

the scikit-image is 0.18 version. I think the problem is the nan values in the array? ndvi.shape (171, 169) ndvi array([[nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan, ..., nan, nan, nan], ..., [nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan, ..., nan, nan, nan]])

it also does not work with masked arrays: t3 = np.ma.masked_invalid(ndvi): Thank you

On Fri, Dec 18, 2020 at 7:33 PM Juan Nunez-Iglesias mailto:jni@fastmail.com> wrote: Hi Laurynas,

What version of scikit-image are you using? threshold_otsu should work with arrays of any dimensionality, including 1D arrays:

In [2]: filters.threshold_otsu(np.random.random((11746,))) Out[2]: 0.4980217396247033

Juan.

...

On 19 Dec 2020, at 1:37 am, Laurynas Gedminas mailto:laurynasgedminas@gmail.com> wrote:

Hey, the suggested code reshapes the original array from (171, 169) to (11746,) and does not work in otsu threshold

Thank you, Laurynas

On Thu, Dec 17, 2020 at 9:56 PM Juan Nunez-Iglesias mailto:jni@fastmail.com> wrote: Some methods, like Li, have some special handling for nan values, but that is not uniform across all our methods, as you have seen. 😂 I would suggest:

```python t = threshold_otsu(image[np.isfinite(image)]) ```

That will only pass the non-nan, non-inf pixels to Otsu to do the calculation.

Hope this helps!

Juan.

...

On 18 Dec 2020, at 7:19 am, Laurynas Gedminas mailto:laurynasgedminas@gmail.com> wrote:

Hello, I have NDVI image, its not a square, and has some None values, I am trying to create otsu threshold, but getting error: autodetected range of [nan, nan] is not finite

how to go around that?

Thank you,