OCTproEngine

High-performance Optical Coherence Tomography (OCT) processing library with GPU acceleration.

Performance

Preliminary results: Performance Benchmark

Requirements

• CMake ≥ 3.18
• CUDA Toolkit ≥ 11.0 (optional, for CUDA backend)
• FFTW3 (optional, for CPU backend)
• OpenCL (optional, for OpenCL backend)
  • VkFFT
• Vulkan SDK (optional, for Vulkan backend)
  • Windows: Download from LunarG Vulkan SDK
  • VkFFT (auto-downloaded if enabled)
• Python ≥ 3.8 (optional, for Python bindings)
  • pybind11, NumPy
• C++ Compiler
  • Win: VS 2019+ (C++14, must be compatible with your cuda version)

Build

Windows

For a complete walkthrough from installing the prerequisites to running the Python examples, see the Getting Started on Windows guide. The short version:

Open Developer Command Prompt for VS 2022 (or your Visual Studio version), navigate to the project root directory:

cd path\to\octproengine # Replace with your actual path

then run:

build_windows.bat

The bat script builds the C++ library as well as the Python bindings.

Alternative: build manually
For default build without python bindings, run:

cd path\to\octproengine # Replace with your actual path
mkdir build
cd build
cmake .. cmake --build . --config Release

or you can enable python bindings with:

cmake .. -DBUILD_PYTHON=ON
cmake --build . --config Release

Here is a list of all available build options:

Build Options:

Option	Default	Description
BUILD_CUDA	ON	Build with CUDA backend support
BUILD_CPU	ON	Build with CPU backend (requires FFTW3)
BUILD_OPENCL	ON	Build with OpenCL backend (requires VkFFT)
BUILD_VULKAN	ON	Build with Vulkan backend (requires Vulkan SDK, VkFFT)
BUILD_PYTHON	OFF	Build Python bindings (requires pybind11, NumPy)
BUILD_TESTS	ON	Build test suite
BUILD_EXAMPLES	ON	Build example applications
BUILD_TOOLS	ON	Build optional ProcessorTools (Recorder, etc.)
BUILD_OCT_VIEWER	OFF	Build interactive OCTproViewer with ImGui (auto-downloads GLFW & ImGui)
FFTW3_AUTO_DOWNLOAD	ON	Auto-download FFTW3 if not found (Windows only)
VKFFT_AUTO_DOWNLOAD	ON	Auto-download VkFFT if not found

Note: At least one backend (BUILD_CUDA, BUILD_CPU, BUILD_OPENCL, or BUILD_VULKAN) must be enabled.

Linux / Jetson

See Jetson Build Instructions for NVIDIA Jetson platforms.

Running Tests

C++ Tests

Functional tests (run from the repository root):

ctest --test-dir build -C Release -LE perf

Individual tests can also be run directly from build/tests/Release/.

C++ Performance Benchmark

The benchmark prints its result table to the console and saves it as CSV, so run it directly:

cd build/tests/Release
test_performance_benchmark

Performance tests carry the ctest label perf, which is what excludes them from the functional test run above.

Python Tests

After building, the Python module is in the build directory but not yet on your Python path.

Set PYTHONPATH (temporary, per-session):

The build script outputs the exact command you need. After build_windows.bat completes, it will show:

set PYTHONPATH=C:\your\actual\path\octproengine\build\python\Release;%PYTHONPATH%

Copy and paste that exact command into your command prompt.

Then you can run Python tests:

cd python/tests
python run_all_tests.py

Manual Verification

You can use the OCTproViewer app to visually verify the processing and run a basic performance benchmark. You can find the app in build/examples/Release/octproviewer.exe if you built it.

Quick Start - C++

#include"processor.h"
#include<iostream>intmain() {
// Create processor (VULKAN, CUDA, CPU, or OPENCL)
ope::Processor processor(ope::Backend::CUDA);
// Configure
processor.setInputParameters(2048, 512, 1, ope::DataType::UINT16);
processor.enableResampling(true);
processor.enableWindowing(true);
processor.enableLogScaling(true);
// Initialize
processor.initialize();
// Set callback
processor.addOutputCallback([](const ope::IOBuffer& output) {
std::cout << "Processed " << output.getSizeInBytes() << " bytes" << std::endl;
});
// Get buffer, fill with data, process
ope::IOBuffer& buffer = processor.getNextAvailableInputBuffer();
// ... fill buffer with your OCT data ...
processor.process(buffer);
return0;
}

Quick Start - Python

importoctproengineasopeimportnumpyasnp# Create processor (VULKAN, CUDA, CPU, or OPENCL)proc=ope.Processor(ope.Backend.CUDA)
# Configureproc.set_input_parameters(2048, 512, 1, ope.DataType.UINT16)
proc.enable_resampling(True)
proc.enable_windowing(True)
proc.enable_log_scaling(True)
proc.initialize()
# Set callbackdefon_output(output_array, buffer_id):
print(f"Processed buffer {buffer_id}: {output_array.shape}, dtype={output_array.dtype}")
proc.add_output_callback(on_output)
# Process datadata=np.random.randint(0, 65535, size=2048*512, dtype=np.uint16)
buffer=proc.get_next_available_buffer()
buffer[:] =dataproc.process(buffer)