Adding points in 3D is wacky

🐛 Bug

In #3508, the mode guard in 3D was removed in order to be able to select points, but the ability to add points was inadvertently added, and it’s completely broken.

To Reproduce

from skimage import data
from scipy import ndimage as ndi
import napari


blobs = data.binary_blobs(
        length=128, volume_fraction=0.1, n_dim=3
        )[::2].astype(float)
labeled = ndi.label(blobs)[0]

viewer = napari.Viewer(ndisplay=3)
viewer.add_image(blobs, name='blobs', scale=(2, 1, 1))
viewer.add_labels(labeled, name='blob ID', scale=(2, 1, 1))
pts = viewer.add_points()

viewer.camera.angles = (0, -65, 85)
pts.mode = 'add'

napari.run()

Now click on a point and you should see behaviour like the below:

https://user-images.githubusercontent.com/492549/154223487-99da4922-107b-4682-8a75-f8b64870f1fe.mov

Expected behavior

Adding a point in 3D without context is undefined, so one of two things should happen imho:

• you would need two clicks with parallax to define a point, or
• use another layer such as an image or labels layer to define a point in 3D, as done in zarpaint (though that appears to be a bit broken at the moment, still investigating)

Environment

napari: 0.4.13rc1.dev59+g79ef36175
Platform: macOS-11.4-x86_64-i386-64bit
System: MacOS 11.4
Python: 3.9.7 | packaged by conda-forge | (default, Sep  2 2021, 17:58:46)  [Clang 11.1.0 ]
Qt: 5.15.2
PyQt5: 5.15.6
NumPy: 1.20.3
SciPy: 1.7.1
Dask: 2021.09.0
VisPy: 0.9.6

OpenGL:
  - GL version:  2.1 ATI-4.5.14
  - MAX_TEXTURE_SIZE: 16384

Screens:
  - screen 1: resolution 2304x1296, scale 2.0
  - screen 2: resolution 1792x1120, scale 2.0

Plugins:
  - animation: 0.0.2
  - console: 0.0.4
  - explode-view: 0.0.1
  - napari-midi: 0.1.dev5+g16c91f6.d20210914
  - napari-skeleton-curator: 0.1.dev17+gd073290.d20211212
  - nd2-reader: 0.0.6
  - platelet-unet-watershed: 0.0.3
  - scikit-image: 0.4.13rc1.dev59+g79ef36175
  - svg: 0.1.5
  - tracks: 0.0.2

[kevinyamauchi]

Hey @jni . Sorry for the late reply. Things have been busy here.

I was looking into the issue and I think the problem is when the points layer has an ndim of 2. We currently don’t have a well defined behavior for changing the dimensionality of layers. For example, if you try to add a 3D point to a points layer with ndim == 2, you will get an error. Adding 3D points to a 2D layer through the GUI doesn’t raise an error because it is taking the “displayed dims” of the coordinate (i.e., 0 and 1 axis). I am not sure what the correct behavior should be, but we can disable this for now.

If you try examples/nD_points.py the point addition “works” in 3D for when the points ndim > 2. However, it is adding the points to the invisible camera plane, which I can see how that is not ideal. We could add a flag to disable the “add” button on the layer controls when rendering in 3D.

I don’t love the plane idea because it’s totally invisible. As a user there should be some kind of visual intuition about where the click is going to go. What do you think about the parallax idea? (Click to make a ray, rotate, click to turn the ray to a point.)?

I like this idea. I would be up to try to make this in 0.4.16

[kevinyamauchi]

I merged #4184, which fixes the main issue here. Thus, I will close this issue and open a new one for improving the 3D points adding UI.