fastlbp-baseline-imbg 0.0.7

A wrapper for Ben's LBP bio pipeline

ImageTextureFinder

A project to create an easy-to-use way of finding areas of common patterns and structures within an image. Should work on any image, designed for use on any biological images including DAPI, IMC and H&E.

See sample_run.sh for details.

• Branch baseline is the most stable. It is ready for pip packaging.
• Branch pip is stale at the moment
• Branch dev is unstable and for dev purposes only.

Container

Image tag is mkrooted/imbg-fastlbp. Hosted on Docker Hub (https://hub.docker.com/repository/docker/mkrooted/imbg-fastlbp/general). See https://github.com/imbg-ua/fastLBP-sandbox for details

---

Guides

How to build and deploy a pip package

Src: https://packaging.python.org/en/latest/tutorials/packaging-projects/

• Add your access token to .pypirc

  # ~/.pypirc 
  [pypi]
    username = __token__
    password = pypi-TOKEN_FROM_YOUR_PYPI_SETTINGS_GOES_HERE
  

• Ensure that your Python is 3.8 because the package targets Python 3.8 and thus requires to be build using this Python version

  python --version
  # Should show Python 3.8.something
  

• Install prerequisites (twine and build)

  pip install --upgrade twine build
  

• Edit project version in pyproject.toml
• Build and upload the project

  # while in root project directory
  python -m build      # .whl and .gz output will be at ./dist directory
  python3 -m twine upload dist/*   # note that this can accidentally upload unneeded builds
  

---

Algorithm notes

Step 1 performs an LBP and creates histograms for each method.

Method is a combination of the following parameters:

• image name
• image channel
• LBP radius
• LBP number of points

Every method's result got saved into the separate .npy file. There is a correspondence betweeen a method and a computational job.

Step 2 collects all the results and concatenate them along the features dimension. That means that feature vector of a patch is a concatenation of all LBP codes from all channels and all LBP radii.