openptv2

Particle Tracking Velocimetry, single Cython 3 pure-Python engine

Overview

openptv2 is a single-engine PTV library: the same src/openptv2/algorithms/ modules run interpreted in development and compiled through Cython 3 when built, with no separate C core, no Cython bindings to an external library, and no runtime engine selector.

• Algorithms (src/openptv2/algorithms/) - Cython 3 pure-Python particle tracking, correspondence, and calibration code
• Plugins (src/openptv2/plugins/) - pluggable tracker implementations (fast, MyPTV/ProPTV-informed trackers)
• GUI (src/openptv2/gui/) - TraitsUI/Chaco desktop application
• Batch pipeline (openptv2-batch) - headless sequence/tracking runner for scripting and cloud use
• Tutorials & Case Studies - Lid-Driven Cavity Flow Tutorial | Aortic Pulsatile Flow Tutorial

What's Inside (and Where the Tracking Concepts Come From)

openptv2 is a self-contained PTV engine: detection, camera calibration, correspondence/stereo-matching, and the classic forward tracking loop are all implemented from scratch here in src/openptv2/algorithms/.

On top of that native engine, openptv2 ships plugins that bring in tracking concepts from two external open-source projects:

• MyPTV — MIT licensed (© 2022 Ron Shnapp). openptv2's myptv_2d_tracking / nearest_hungarian_3d plugins adapt MyPTV's algorithm ideas (2D per-camera image-space tracking with multi-camera consensus, and 3D kinematic prediction + linear-assignment matching) onto openptv2's own data structures and assignment machinery.

• proPTV — MIT licensed (© 2023 DLR, Robin Barta). openptv2 vendors the small pure-numpy core of proPTV (src/openptv2/plugins/proptv/): the Gaussian-Mixture-Model / basis approximation and Savitzky-Golay smoothing routines used in proPTV's track prediction. The predictive_gmm_3d plugin wires those routines into openptv2's tracker.

Important: openptv2 incorporates and adapts parts of these projects as plugins — not their full frameworks. The complete tracking capabilities of each project — e.g. proPTV's 2D-image triangulation pipeline, Soloff calibration, backtracking/repair and Eulerian field estimation, or MyPTV's full photogrammetry, trajectory smoothing and anisotropic-particle tools — live in and are available only from their own repositories:

• MyPTV: https://github.com/ronshnapp/MyPTV
• proPTV: https://github.com/RobinBarta/proPTV

If you need those advanced features, please use those projects directly. Both are permissively MIT-licensed (see License).

Key Features

• Single runtime: one codebase, no C/Cython vs. Python fallback split to keep in sync
• Pluggable trackers: swap tracking algorithms via the plugin architecture
• Easy installation: pip install openptv2 for headless/batch use, openptv2[gui] to add the desktop GUI

---

Installation

There are three install profiles:

Command	For	Includes
pip install openptv2	scripting + openptv2-batch (default)	headless algorithms/API + sequence/tracking pipeline
pip install openptv2[gui]	desktop users	default plus the TraitsUI/Chaco/PySide6 desktop GUI
pip install openptv2[dev]	contributors	everything: GUI, tests, lint, type-check, notebooks, docs

Default (headless / batch)

The bare install is the headless batch runtime — the library, the openptv2-batch sequence + tracking pipeline, no GUI:

uv pip install openptv2
# or
pip install openptv2

Don't have uv? curl -LsSf https://astral.sh/uv/install.sh | sh

GUI (desktop users)

uv pip install openptv2[gui]
# or
pip install openptv2[gui]

Verify Installation

python -c "import openptv2; print(f'openptv2 version: {openptv2.__version__}')"
python -c "from openptv2 import Tracker; print('Tracker imported successfully')"

Note: launching the GUI (openptv2-gui) requires the [gui] extra. A default install that then launches the GUI will fail with ModuleNotFoundError: No module named 'traitsui' (and chaco/PySide6) — that is expected; install openptv2[gui].

Docker (no Python setup required)

A single image serves both the GUI and batch. It bakes in a trimmed test_cavity demo at /demo/test_cavity so the first run works with no data.

# Build once
docker build -t openptv2 -f docker/Dockerfile .

# GUI on the host X display (Linux/X11), current folder mounted at /data
./docker/run-gui.sh
#   in the GUI, open /demo/test_cavity to try the baked demo

# Headless batch on your own data
docker run --rm -v "$PWD:/data" openptv2 \
  openptv2-batch /data/<experiment>/parameters_Run1.yaml <first> <last>

X11 notes: run-gui.sh handles xhost and mounts /tmp/.X11-unix. On Wayland run xhost +local:root in an XWayland session; on macOS/Windows use an X server (XQuartz / VcXsrv) and set DISPLAY accordingly.

Headless cloud batch: docker/Dockerfile.cloud is a slim, no-GUI, free-threaded 3.14t image for servers/Cloud Run. See docs/cloud-batch.md for the one-command install, openptv2-batch usage, and measured timings.

Zarr + HDF5 Cloud Storage: OpenPTV2 includes a native, high-performance Zarr storage engine (res/run.zarr) replacing thousands of per-frame text files with a cloud-native chunked format. See docs/zarr-hdf5-storage.md for usage, terminal inspection, and Flowtracks HDF5 export.

---

For Developers (Build from Source)

Prerequisites:

• Python 3.11–3.14 (free-threaded 3.14t works too, see docker/Dockerfile.cloud)
• C compiler (gcc on Linux, clang on macOS, MSVC Build Tools on Windows) — needed to build the Cython 3 extensions, no CMake involved
• uv (recommended) or pip

System Dependencies

Linux (Debian/Ubuntu):

sudo apt-get update
sudo apt-get install -y build-essential python3-dev

Linux (Fedora/RHEL):

sudo dnf install -y gcc gcc-c++ python3-devel

macOS:

xcode-select --install

Windows:

• Install Microsoft Visual C++ Build Tools

Step 1: Clone the Repository

git clone https://github.com/alexlib/openptv2.git
cd openptv2

Step 2: Install Dependencies and Build

Using uv (recommended):

# Sync all dependencies and build the package
uv sync --extra dev

Using pip:

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install build dependencies
pip install setuptools cython numpy

# Install in development mode
pip install -e ".[dev]"

The default editable install builds the Cython 3 pure-Python modules:

uv pip install -e .

Step 3: Verify Build

Always run verification inside your virtual environment to ensure the compiled libraries are correctly found:

# Test imports
uv run python -c "import openptv2; print(f'openptv2 version: {openptv2.__version__}')"
uv run python -c "from openptv2 import Tracker; print('Tracker imported successfully')"

# Run core tests
uv run pytest tests/unit/ -v

---

GUI Dependencies

The [gui] extra includes:

• traitsui, enable, chaco (Enthought framework + visualization)
• PySide6 (Qt bindings)
• matplotlib, pandas, flowtracks (analysis)

(scikit-image and numpy are core dependencies, installed either way.)

---

Installing from Binary Wheels

Pre-built wheels are published to PyPI:

# Using pip
pip install openptv2

# Using uv
uv pip install openptv2

---

Troubleshooting Installation

Common Issues

1. "C compiler not found"

# Linux: sudo apt-get install build-essential
# macOS: xcode-select --install
# Windows: Install MSVC Build Tools

2. "Cython not found"

pip install cython>=3.0.0

3. "NumPy version mismatch"

pip install numpy>=2.0.0

4. Cython extensions not rebuilt after editing src/openptv2/algorithms/

uv run python setup.py build_ext --inplace

---

Usage

Basic Tracking

The quickest way to run a full detection → correspondence → tracking pipeline is the batch CLI (see Batch Processing below) or the GUI. For scripting against the library directly — loading calibrations/parameters and driving openptv2.Tracker — see docs/tutorials/.

Runtime

import openptv2

print(openptv2.get_runtime_info())
# {'engine': 'cython3-pure-python', 'compiled': True/False, 'package': 'openptv2'}

GUI

Launch the graphical interface using the unified console scripts. Ensure your virtual environment is activated (source .venv/bin/activate) or prefix the commands with uv run:

# Launch the GUI using the standard openptv2-gui or shorter pyptv_gui shortcut
uv run pyptv_gui

# Launch the GUI in the single-engine runtime
uv run pyptv_gui --workdir=./test_data/test_cavity

Batch Processing

Run high-throughput processing sequences using command-line batch utilities (ensure your virtual environment is activated or use uv run):

# Run batch processing with the single runtime
uv run pyptv_batch --workdir=./test_data/test_cavity --first=10000 --last=10005

# Run with legacy positional arguments (for backward compatibility)
uv run pyptv_batch ./test_data/test_cavity/parameters_Run1.yaml 10000 10004

# Parallel batch processing (distributes frame chunks to multiple cores)
uv run python -m openptv2.batch.pyptv_batch_parallel test_data/test_cavity/parameters_Run1.yaml 10000 10004 4 --mode sequence

### Parallel Processing Config
OpenPTV2 supports multi-core parallel processing during image preprocessing and target detection (Approach C). This is highly effective for accelerating execution on multi-core systems.

* **GUI Configuration**: Open the **Main Parameters** dialog, navigate to the **Sequence** tab, check the **Parallel Pre-processing** box, and set the **Number of workers** (e.g., `4` or `0` for automatic core detection).
* **CLI/Environment Configuration**: Set environment variables before running any tracking or batch sequence:
  ```bash
  export OPENPTV_PARALLEL_PREPROCESS=True
  export OPENPTV_NUM_WORKERS=4  # Set worker processes (omitting uses all CPU cores)

Command-line Shortcuts and Running Without uv

You can run these command-line tools without prefixing them with uv run using any of the following approaches:

1. Activate the Virtual Environment (Standard Workflow)

By activating the project's virtual environment, the environment's bin/ directory is added to your shell's PATH. This registers all entry points (like pyptv_gui, pyptv, pyptv_batch) directly in your terminal:

source .venv/bin/activate

# Now run directly without uv
pyptv_gui -w ./test_data/test_cavity

2. Execute via Direct Path

If you do not wish to activate the virtual environment, you can run the built executable directly from the local .venv folder:

./.venv/bin/pyptv_gui -w ./test_data/test_cavity

3. Define Shell Aliases (Global Access)

To run these shortcuts cleanly from any directory without manual paths, add alias entries to your shell profile (e.g., ~/.bashrc or ~/.zshrc):

# Add these lines to ~/.bashrc or ~/.zshrc
alias pyptv_gui='/home/user/Documents/GitHub/openptv2/.venv/bin/pyptv_gui'
alias pyptv_batch='/home/user/Documents/GitHub/openptv2/.venv/bin/pyptv_batch'

# Reload shell profile
source ~/.bashrc

# Now launch cleanly from any folder
pyptv_gui -w ./test_data/test_cavity

---

Single Runtime Behavior

OpenPTV2 now ships a single runtime:

1. One codebase: algorithms/*.py is the only implementation path.
2. Two execution modes: the same modules run interpreted during development and compiled when built through Cython 3.
3. No engine switching: OPENPTV_ENGINE, set_engine(), and optv dispatching are no longer part of the runtime model.

---

Short vs. Long Options

Both styles are fully supported across all command-line scripts. Choose based on your context:

• Short Options (-e, -w, -f, -l): Best for manual terminal use, quick debugging, and live interactive typing.
• Long Options (--engine, --workdir, --first, --last): Best for automation scripts, documentation, and config files to maximize readability.

---

Documentation

• Installation Guide
• Quick Start
• API Reference
• Algorithm Documentation
• Developer Guide

OpenPTV³ Auto-Research Dashboard

Live, zero-install demo of the differentiable PTV pipeline (docs/plans/differentiable_ptv_nextgen_plan.md): drag the Stage-1 intensity threshold and watch the gradient flow through to the Lagrangian physics loss, live acceleration PDF, and velocity power spectrum.

---

Repository Structure

openptv2/
├── src/openptv2/
│   ├── algorithms/    # Cython 3 pure-Python engine: calibration,
│   │                  # correspondences, orientation, tracking, etc.
│   │                  # (the only algorithm implementation path)
│   ├── plugins/       # Pluggable tracker/sequence implementations
│   │                  # (fast, MyPTV/ProPTV-inspired trackers, rembg, ...)
│   ├── batch/         # openptv2-batch / pyptv_batch headless pipeline
│   ├── storage/       # Zarr frame store (res/run.zarr)
│   ├── gui/           # TraitsUI/Chaco desktop application
│   ├── tracker.py, calibration.py, correspondences.py, ...  # public API
│   └── __init__.py
├── tests/             # Test suite (unit, parity, perf, integration, gui)
├── docs/              # Documentation, tutorials, developer guide
├── scripts/           # Build/analysis/benchmark helper scripts
├── docker/            # Dockerfiles for GUI and cloud batch images
├── test_data/         # Calibration files, parameter files, fixtures
├── pyproject.toml     # Python project config (build, deps, entry points)
└── README.md

---

Testing

# All tests
uv run pytest

# By marker
uv run pytest -m unit
uv run pytest -m "not slow"

# GUI tests (headless)
uv run pytest tests/gui/ -v

# Hot-path smoke test (tracking + correspondences)
uv run pytest tests/unit/test_track.py tests/unit/test_track3d.py tests/unit/test_correspondences.py -v --tb=short

---

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Run tests: pytest
5. Submit a pull request

Development Workflow

# Set up development environment
uv sync --extra dev

# Make changes to source code

# Run tests
pytest tests/ -v

# Build documentation (optional)
cd docs && make html

---

License

MIT. See LICENSE for details.

As of 0.3.2, openptv2 is relicensed from LGPL-3.0 to MIT. The C/Cython core that carried the LGPL license now lives in the separate openptv repo; this codebase is a from-scratch pure-Python/Cython PTV engine plus a pluggable tracker architecture.

Third-party tracking code

openptv2 includes and adapts tracking code from two MIT-licensed projects, in accordance with their licenses:

• MyPTV (MIT, © 2022 Ron Shnapp) — algorithm concepts adapted into the myptv_2d_tracking / nearest_hungarian_3d plugins. Their MIT notice is incorporated; see https://github.com/ronshnapp/MyPTV.

• proPTV (MIT, © 2023 DLR / Robin Barta) — the GMM / Savitzky-Golay routines in src/openptv2/plugins/proptv/ are vendored from proPTV. proPTV's license requires that its copyright/permission notice be included in copies and that its underlying publication be cited. Accordingly:

  MIT License, Copyright (c) 2023 DLR (Project owner: Robin Barta). GMM / Savitzky-Golay routines adapted from Barta, Robin, et al. "proPTV – A probabilistic particle tracking velocimetry framework." Journal of Computational Physics (2024), https://doi.org/10.1016/j.jcp.2024.113212.

  The vendored modules at src/openptv2/plugins/proptv/ carry the MIT copyright/permission notice in their headers.

Using either project's advanced, project-specific features is out of scope here — refer to and use those projects directly.

---

Acknowledgments

openptv2 combines work from:

• openptv - C library and bindings
• pyptv - Python GUI
• openptv-python - Python/Numba engine
• MyPTV - 2D/3D tracking algorithm concepts adapted into the MyPTV plugins (MIT)
• proPTV - GMM / Savitzky-Golay track-prediction routines vendored into src/openptv2/plugins/proptv/ (MIT, © 2023 DLR / Robin Barta)

See License for the licensing and citation details.

---

Contact

• Mailing list: openptv@googlegroups.com
• GitHub: https://github.com/alexlib/openptv2
• Issues: https://github.com/alexlib/openptv2/issues

---

Helper Scripts

The project includes scripts for building and testing:

Script	Purpose
scripts/build_wheel.sh	Build binary wheel from source
scripts/install_wheel.sh	Install wheel in clean test environment
scripts/run_tests.sh	Run test suite in test environment
docker/Dockerfile.slim	Slim Docker image for testing