hanjinliu/impy

Speed up coding/extending image analysis in Python.

Hi there 👋

I'm especially interested in image analysis and GUI development.

impy's Issues

Flexible associable slicing of `ImgArray`

Currently, ImgArray has an attribute labels which is "associable". An associable attribute means that when the ImgArray is sliced the associable attribute will also get sliced at the correct position.

In terms of extensibility, other object should also be compatible with this strategy, such as ROI. All the associable objects should be managed by a list and sliced in _view_labels.

Examples

Following code does not work but clearly implies what should be supported.

img = ip.zeros((100, 100, 100), axes="zyx")  # create an array
img.points = [[10, 20, 30], [60, 50, 40]]
img.points  # Out: Points([[10, 20, 30], [60, 50, 40]])
img[5:].points  # Out: Points([[5, 20, 30], [5, 50, 40]])

Methods to convert an image into a RGBA

• PhaseArray needs special consideration.

`Axes` should support "undef" axis.

In some cases we have to deal with undefined axis.
Axes should support this kind of problem like img.axes = "##yx" suppose the undefined axis is represented by "#".

Examples

1. Slicing using a list.

img = ip.zeros((10, 10, 100, 100), axes="tzyx")

# This is OK. Result has "tzyx" axes.
img[[3, 6, 9]]

# This is not supported well. Returns an array without axes.
# Ideally, axes should be "#yx"
img[[3, 6, 9], [2, 5, 8]]

2. Slicing using a binary array

img = ip.zeros((10, 10, 100, 100), axes="tzyx")

# This is OK. Result has "tzyx" axes.
sl = np.random.random((10,)) > 0.5
img[sl]

# This is not supported well. Returns an array without axes.
# Ideally, axes should be "#yx"
sl = np.random.random((10, 10)) > 0.5
img[sl]

3. New axis

img = ip.zeros((10, 10, 100, 100), axes="tzyx")
img[np.newaxis]  # should be "#tzyx"

Refactor arrays

There are many inefficient things in current version.

• Split image source file path into img.dirpath and img.name is very confusing. Should be img.source or something.
• img.history should be completely removed. It usually becomes useless after binary operation, stacking or other functions that need more than one image. Stop recording history will also simplify the code a lot.
• Clearly split metadata related to axes (such as scale and channel names) and image itself (such as source file).
• Most of the custom napari widgets should be removed because it seems that napari plugin development is so active that people should install what they want. Only those axes and scale related things should be retained on impy side.

Maybe I should create a new repo like Julia's AxesArray to keep things extensible enough, and use impy as an practical extension.

MSVS 16 error impy is not subscriptable

Hi, I have just started out trying to import impy as ip in MSVS 16 2019 with the following statement

import impy as ip

I also tried

from impy import ALL as ip

Both statement fail with VS error that impy is not subscriptable.

Could you please address this issue.

My python 3.8 virtual environment is installed under the project repo folders via command line terminal
I have used the command pip install impy-array to install impy in the project's virtual environment.

Thanks for you help

Unnecessary configuration

The folder impy/.vscode/ seems to be unnecessary and includes your personal settings.

I suggest removing this folder from this repository.

Tiled deconvolution

Use map_overlaps to implement tiled Wiener/Lucy deconvolution.

• PSF should be properly cropped.
• Parallelize 4D image stack processing.