I implemented these changes to try to solve #6, returning several layers for several drawn shapes, but it is still not working.

The motivation for this is explained here and there is indication that this approach has been implemented here.

Does any of you guys have ideas to make this work? @haesleinhuepf @tdmorello

Hi Robert,

First of all, consider me a big fan of your napari plugins -- I really appreciate the number of tools you are creating and publishing for napari users!

Second, I was testing out this plugin and found it wasn't working on RGB images, so I dug into the code a little bit and thought of some ways to (hopefully) make it applicable to more scenarios. If you get some time to test it out, I'd really appreciate it!

I think the key features are:

1. it works with any number of dimensions
2. you can draw a cropping region in an orthogonal view (e.g. XZ plane) and it'll give you the expected results
3. the slice indices aren't hard-coded, so it should be a little easier to maintain and adapt

I also wrote some tests to help my dev'ing (and hope they will be useful to the repo as well).

Finally, I see @zoccoler 's PR and am planning on trying to fit my changes in with his (polygon cropping). The code I'm sharing can be easily modified to output more than 1 new layer when multiple shapes are draw.

Let me know what you think.

Best, Tim

I think this should fix #28 but I can't know for sure until I've started a run of the github CI.

Edit: Tests are passing, switching to non-draft mode.

Fixes #28

I have noticed that, for certain ellipses, the following cropped image is returned. This behavior is inconsistent: the line appears or not depending on where the ellipse is drawn.

I'm not sure whether this installation is necessary for packages that do not rely on GPU-functionality. I'll start a PR to see if tests still pass after removing this part from test_and_deploy.yaml.

Hey Robert @haesleinhuepf

Here's a new example. I didn't match your style with the highlight circle around the cursor -- I hope that's ok. The url to the animation will have to change before merge.

Hi Robert @haesleinhuepf

I modified the function adding 2 new functionalities:

1. if more than one shape is drawn in the layer shape, it crops all shapes and adds them as new layers;
2. if the shapes have irregular shapes (like ellipses or polygons), it crops according to that shape, clearing pixels outside the shape;

• I also changed a little the extent of the shapes position and size (from .astype(int) to np.ceil or np.around) to ensure the irregular drawn shape would be entirely captured;

Please let me know if it works and if you have suggestions/improvements to the code 😃

Hi guys,

This Draft PR is intended to fix #20 . The problem happened because after #7 because here we always assigned the same name for drawn shapes from the same Shapes layer, thus, napari replaces previous output Image layer data instead of creating a new layer if the function was executed again.

The proposed solution is to always provide a new unique name by checking layer names in the viewer (or layer names about to be added to the viewer in case of multiple shapes).

It works here. I will write a test before turning this into regular PR.

Best, Marcelo

Applying crop to shapes like these:

import numpy as np
arr_2d = np.arange(0, 25).reshape((5, 5))
shapes = [
    np.array([[1, 1], [1, 3], [5, 3], [5, 1]]),
    np.array([[0.5, 0.5], [0.5, 3.5], [4.51, 3.5], [4.51, 0.5]]),
    np.array([[0, 2], [5, 5], [5, 2], [2, 0]]),
    
]
shape_types = ["rectangle", "ellipse", "polygon"]

viewer.add_image(arr_2d)
shapes_layer = viewer.add_shapes(shapes, shape_type=shape_types, edge_width=0)

Only works with the rectangle. For irregular shapes drawn close or over image edge, it gives an error like this:

ValueError: operands could not be broadcast together with remapped shapes [original->remapped]: (6,6)  and requested shape (5,5)

Hi all,

does anybody see issues if we make crop_region part of the public API?

E.g. like this:

from napari_crop import crop_region

We could then use it from scripts as discssed in this thread.

Let me know what you think! If there are no concerns, I would just open the API.

Best, Robert

Fixes a problem where Labels layers could not be cropped.

The crop_region function was looking for layer_props["rgb"] raising a KeyError with Labels layers.

Also, adding a test for cropping Labels.

Both the image layer to crop from and the shape layer defining the crop regions could have an .affine property. I don't think these are currently taken into account.

This seems to depend on https://github.com/haesleinhuepf/napari-workflows by @haesleinhuepf, but when installing the plugin, this dependency is not automatically installed.

Proposed enhancement

napari-crop can crop in nD based on 2D shapes. It would be a good new feature to cut in irregular 3D shapes with the user providing points/shapes representing polygons at different z-slices.

Example of the current behavior

The code below is an attempt to reproduce it in the current state.

import napari
from napari_crop._function import crop_region
from skimage.data import cells3d
import numpy as np
from magicgui import magicgui

viewer = napari.Viewer()
image = cells3d()
image = image[:,1]

polygon1 = np.array([[ 24., 141., 100.],
                    [ 24., 135., 115.],
                    [ 24., 142., 130.],
                    [ 24., 155., 134.],
                    [ 24., 167., 129.],
                    [ 24., 173., 117.],
                    [ 24., 163., 103.],
                    [ 24., 156.,  95.]])

polygon2 = np.array([[ 33. , 136., 102.],
                    [ 33., 132., 115.7],
                    [ 33., 152., 135.],
                    [ 33., 165., 134.],
                    [ 33., 176., 122.],
                    [ 33., 180., 112.],
                    [ 33., 160.,  89.],
                    [ 33., 140.,  95.]])

polygon3 = np.array([[ 45., 146., 94.],
                    [ 45., 143., 109.],
                    [ 45., 156., 122.],
                    [ 45., 170., 126.],
                    [ 45., 179., 120.],
                    [ 45., 182., 108.],
                    [ 45., 177.,  99.],
                    [ 45., 154.,  89.]])

# This is how the shapes layer data would be if the user draw polygons along a z-stack
polygon_list = [polygon1, polygon2, polygon3]

# Transforms a list of polygons into a single 3D array of vertices
shapes = [polygon[np.newaxis,:] for polygon in polygon_list]
shape_3D = np.concatenate(shapes, axis=0)

viewer.add_image(image)
viewer.add_shapes(np.array(shape_3D), shape_type='polygon')

widget = magicgui(crop_region)
viewer.window.add_dock_widget(widget)

Two errors happen:

1. If the number of vertices per slice is not the same, there is a ValueError because the polygons cannot be concatenated into a 3D array with the same shape.
2. Even if the number of vertices match, it gives an interpolation error: ValueError: 'linear' is not a valid Interpolation

From this, it seems the shapes layer may not be the best choice for that. Turning it into a surface and cropping it seems more appropriate. Suggestions and feedback welcome :)

I believe napari-crop should be able to handle time crops as well. It is just a matter of properly slicing, right? For that case, maybe relying on the shapes layer wouldn't be so intuitive from the user point of view.

My first thought would be to add 2 spinboxes for start end end of time slice. It could be a rangeslider as well. Enabling this by means of a checkbox, or even better if the type of data can be auto-detected.

What do you guys think?

I have an 10888 x 6451 px image. When I draw small crops, it works as expected. When I draw large crops, the output shapes are not as expected.

E.g. Crop shape = 9917 x 5423 output shape = 9025 x 4164

Maybe it has something to do with how the image data is tiled in memory?

When I crop a region that extends beyond the bounds of the image to get the whole image, it works as expected.