scivision

Optical flow repositories I've collected

I've collected a few optical flow repositories that interested us due to their use in papers relevant to our field. There are many, many more algorithms that are likely of greater merit-here's just a few I've looked at over time.

Black Robust Optical Flow uses C code from 1996 originally. I created Python and Matlab interfaces to this. I added CMake and updated the C code to work on modern compilers. There is newer code from Black circa 2010 that is said to be easier to use and more effective.
PyOptFlow I implemented Python and Matlab code for Horn Schunck optical flow
Python Lucas Kanade I improved Python code for Lucas Kanade optical flow
Phantom generator I wrote a Python phantom generator that makes morphing / moving shapes that are roughly like

name	CMake Standards
description	Coding conventions for CMake files
applyTo	*/.cmake,**/CMakeLists.txt

Editing rules

Assume CMake 3.26 or newer is available cmake_minimum_required(VERSION 3.26)
Use cmake_path() instead of get_filename_component()

Minimal CMake install using older CMake

We often find such a platform-independent script handy for CI or for helping end users get a new-enough CMake. It can also be used to install older specific CMake versions for CI testing of minimum CMake versions.

We have a more elaborate and general script that handles older CMake versions at cmakeutils under scripts/install_cmake.cmake. This script here is intentionally very minimal and does little error checking.

CMake version in repository by Linux distribution

Scientific computing libraries with notable CMake minimum versions:

HDF5: CMake 3.26
Scalapack: CMake 3.26

Heterogeneous CPU core count

This Python script uses only built-in stdlib functions and attempts to detect the CPU count of each type of CPU core. Many CPUs have only one core type, while others (especially for laptops) may have 2 or more CPU core types, sometimes called "big-little" or "super-performance-efficiency".

Usage

python cpu_hetero.py

	cmake_minimum_required(VERSION 3.19)

	project(mod_files LANGUAGES Fortran)

	enable_testing()

	include(CheckCompilerFlag)

	set(tgt dummy)
	set(submod define_pi)

	# Activate AOCC compilers by
	# source aocc.sh

	version=6.0.0
	root=/opt/AMD/aocc-compiler-${version}

	[[ ! -d ${root} ]] && echo "ERROR: ${root} not found" && exit 1

	source ${root}/setenv_AOCC.sh

	cmake_minimum_required(VERSION 3.25)

	project(MaxFortran LANGUAGES Fortran)

	set(CMAKE_EXECUTE_PROCESS_COMMAND_ECHO STDOUT)

	if(NOT DEFINED N)
	if(CMAKE_Fortran_COMPILER_ID STREQUAL "NVHPC")
	# tested with NVHPC SDK 26.5
	set(N 1000)

	cmake_minimum_required(VERSION 4.4)

	project(FlangErr LANGUAGES Fortran)

	if(CMAKE_Fortran_COMPILER_ID MATCHES "GNU\|LLVMFlang")
	add_compile_options(-Wunused-variable)
	elseif(CMAKE_Fortran_COMPILER_ID MATCHES "Intel")
	add_compile_options(-warn:unused)
	endif()

	*.json
	__pycache__/