the error in segmented a ebsd png

[Yuanguo1994]

Hi,
I got an error when I made the segmentation to a ebsd microstructure. The picture is shown below:

The error message is shown like this:

segmented = GS.initial_segmentation(filtered)

---

ValueError Traceback (most recent call last)
in
----> 1 segmented = GS.initial_segmentation(filtered)

~\CristalX\grains\segmentation.py in initial_segmentation(self, *args)
131 else:
132 image = self.original_image
--> 133 segment_mask = segmentation.quickshift(image)
134 if self.__interactive_mode:
135 io.imshow(color.label2rgb(segment_mask, self.original_image, kind='avg'))

~\Anaconda3\envs\CristalX\lib\site-packages\skimage\segmentation_quickshift.py in quickshift(image, ratio, kernel_size, max_dist, return_tree, sigma, convert2lab, random_seed)
61 if image.shape[2] != 3:
62 ValueError("Only RGB images can be converted to Lab space.")
---> 63 image = rgb2lab(image)
64
65 if kernel_size < 1:

~\Anaconda3\envs\CristalX\lib\site-packages\skimage\color\colorconv.py in rgb2lab(rgb, illuminant, observer)
1090 .. [1] https://en.wikipedia.org/wiki/Standard_illuminant
1091 """
-> 1092 return xyz2lab(rgb2xyz(rgb), illuminant, observer)
1093
1094

~\Anaconda3\envs\CristalX\lib\site-packages\skimage\color\colorconv.py in rgb2xyz(rgb)
679 # Follow the algorithm from http://www.easyrgb.com/index.php
680 # except we don't multiply/divide by 100 in the conversion
--> 681 arr = _prepare_colorarray(rgb).copy()
682 mask = arr > 0.04045
683 arr[mask] = np.power((arr[mask] + 0.055) / 1.055, 2.4)

~\Anaconda3\envs\CristalX\lib\site-packages\skimage\color\colorconv.py in _prepare_colorarray(arr, force_copy)
148
149 if arr.shape[-1] != 3:
--> 150 raise ValueError("Input array must have a shape == (..., 3)), "
151 f"got {arr.shape}")
152

ValueError: Input array must have a shape == (..., 3)), got (519, 584, 4)

---

The 3D numpy.ndarray is (519, 584, 4), it is not in the shape of (...,3). Is it associated with the png I uploaded?

[CsatiZoltan]

Hi,

I witnessed this type of error before. This is not related to CristalX, rather to your input image. The image is expected to be a 3-channel image, such as RGB. So yes, the shape should be of (..., ..., 3).

[Yuanguo1994]

Hi, I have followed the process to repair the geometry, but still I cant finish the repair work. Let me take your example microstructure for instance.
The hole in the microstructure is shown below (the first picture), as writen in the documentation, I selected the sides edges of the hole to create the closed contour (second picture), but more holes appeared, shown in the third picture. And then I suppressed the hole, the hole I want to supress was not sucessfully removed, shown in the fourth picture.

[CsatiZoltan]

The hole in the microstructure is shown below (the first picture), as writen in the documentation, I selected the sides edges of the hole to create the closed contour (second picture)

I should have emphasized in the documentation that closing a hole makes sense when you have a grain that contains a smaller grain. In that case, you can select the larger grain, which contains the hole, and suppress that hole. In your case, you can see that the hole (corresponding to an unidentified grain) is not an embedded grain (it has more than one neighbor). Hence, there is no hole to suppress. What you rather want is creating a new grain in place of that hole. You can do it in a few seconds by creating a closed wire and then a planar face on it. I demonstrate it on a different microstructure, but you can see the idea. Check the GIF below.

more holes appeared

In most cases, that's just a visualization glitch -- something that happens with other software too (e.g. CAD Assistant). If the subsequent mesh does not contain such holes, you are good to go. If, in some rare cases, those are actual holes, you will need to repair the geometry by sewing it. But for now, don't bother about it.

[Yuanguo1994]

The hole in the microstructure is shown below (the first picture), as writen in the documentation, I selected the sides edges of the hole to create the closed contour (second picture)

I should have emphasized in the documentation that closing a hole makes sense when you have a grain that contains a smaller grain. In that case, you can select the larger grain, which contains the hole, and suppress that hole. In your case, you can see that the hole (corresponding to an unidentified grain) is not an embedded grain (it has more than one neighbor). Hence, there is no hole to suppress. What you rather want is creating a new grain in place of that hole. You can do it in a few seconds by creating a closed wire and then a planar face on it. I demonstrate it on a different microstructure, but you can see the idea. Check the GIF below.

more holes appeared

In most cases, that's just a visualization glitch -- something that happens with other software too (e.g. CAD Assistant). If the subsequent mesh does not contain such holes, you are good to go. If, in some rare cases, those are actual holes, you will need to repair the geometry by sewing it. But for now, don't bother about it.

Hi,

Thanks for the reply, after modifying the parameters of search_neighbor function, I have got a good geometry, shown in the first picture. However, after I did explode to create the faces, some gaps between grains appeared (the second and third picture), even though I tried different limit tolerance. I dont know how to solve the problem, could u please give me some advice?

[CsatiZoltan]

Hi,

True gaps should not appear. I have not noticed anything like this done by the explode tool in Salome. My suspicion is still that this is a visualization glitch. If you are worried about it, export the exploded geometry (select it in the Object Browser and then File --> Export --> STEP) and open it in another CAD program/viewer, such as FreeCAD or CAD Assistant. However, if you want to generate a mesh on the microstructure, you can simply go to the Mesh module in Salome and create the mesh. If there was a real gap between the grains, the mesh will not be conforming -- an easy way to check whether your gap is real or not.

If you have further questions on meshing or other topics not related to this thread, please open a new one.

[Yuanguo1994]

Yes, the geometry could be successfully meshed, so it should be visualization glitch. Thanks for ur so many patient answers. I might have some questions about export the mesh to the format reqiured by FEA software such as abaqus and fepx. I will open a new discussion if I need ur help.