fastlbp 0.3.1

Parallel multiradial LBP features

fastLBP

Highly parallel LBP implementation

Important pre-release warning: If aborted mid-execution, this software sometimes create a lot of orphan processes that needs to be killed manually. Please, note down the name of your python script, search for Python in your task manager and look for the processes that correspond to your python script.

Requirements

FastLBP is tested with Python 3.11 on Windows 10, Debian 11, and Ubuntu 22.04

Python requirements are:

• numpy >= 1.26.0
• Cython (to build the binary modules, will be optional in the future)
• scikit-image >= 0.22.0 (mostly for testing, we plan making this requirement optional in the future)
• pandas >= 2.1.1
• psutil

Installation

• Activate or create a Python 3.11 environment (e.g. using conda create -y -n p11 python=3.11 && conda activate p11)
• Verify you are using the right env
  • python --version and pip --version
• Install a stable version from PyPI
  pip install fastlbp
• Or build the latest version from sources

  git clone [e-mail]:imbg-ua/fastLBP.git
  cd fastLBP
  # git checkout <branchname> # if you need a specific branch
  pip install . # this will install the fastlbp package in the current env
  

• You can use import fastlbp as fastlbp now

GPU (CUDA) optional install

• CPU-only install (default):

  pip install fastlbp
  

• GPU build (opt-in): ensure CUDA is discoverable by setting CUDA_HOME (or having nvcc on PATH) and install the extra:

  export CUDA_HOME=/usr/local/cuda  # adjust to your CUDA install
  pip install fastlbp
  

• To hard-require CUDA and fail if not found, set:

  FORCE_CUDA=1 pip install fastlbp
  

If CUDA isn’t detected, the build will proceed with CPU-only features. At runtime, you can check availability:

import fastlbp
fastlbp.fastlbp.is_cuda_available()  # -> True/False

Testing

# in repo root
conda activate fastlbp
pip install -e .
python -m unittest

Bug reporting

You can report a bug or suggest an improvement using our github issues

Implemented modules

run_fastlbp

Computes multiradial LBP of a single multichannel image in a parallel fashion.

Features:

• Powered by fastlbp.lbp, our implementation of skimage.feature.local_binary_pattern
• Concurrency is managed by Python's multiprocessing module
• Parallel computation via multiprocessing.Pool of size ncpus
• Efficient memory usage via multiprocessing.shared_memory to make sure processes do not create redundant copies of data
• If save_intermediate_results=False then computes everything in RAM, no filesystem usage

TODO:

• Use max_ram parameter to estimate optimal number of sub-processes and collect memory stats. Now max_ram is ignored.

Planned modules

run_chunked_skimage

Similar to 1. run_fastlbp, but each subprocess should compute LBP for its image chunk, not the whole image.

run_dask and run_chunked_dask

Similar to 1. run_fastlbp, but use Dask and dask.array.map_overlap for parallelisation instead of multiprocessing and manual data wrangling