trackpy is a Python package providing tools for particle tracking. Read the walkthrough to skim or study an example project from start to finish.
Then browse a list of more examples, or download the full repository of sample code and data to try them yourself.
There are many similar projects. (See table below.) Our implementation is distinguished by succinct and flexible usage, a thorough testing framework ensuring code stability and accuracy, scalability, and complete documentation.
Several researchers have merged their independent efforts into this code. We would like to see others in the community adopt it and potentially contribute code to it.
If you use trackpy in published research, please read the section Citing Trackpy.
- The widely-used particle locating algorithm originally implemented by John Crocker and Eric Weeks in IDL is reimplemented in Python. Wherever possible, existing tools from widely-used Python modules are employed.
- This reimplemention is full-featured, including subpixel precision verified with test cases.
- The module is actively used and tested on Windows, Mac OSX, and Linux, and it uses only free, open-source software.
- Frames of video can be loaded from a video file (AVI, MOV, etc.), a multi-frame TIFF, or a directory of sequential images (TIFF, PNG, JPG, etc.).
- Results are given as DataFrames, high-performance spreadsheet-like objects from Python pandas which can easily be saved to a CSV file, Excel spreadsheet, SQL database, HDF5 file, and more.
- Particle trajectories can be characterized, grouped, and plotted using a suite of convenient functions.
- To verify correctness and stability, a suite of 150+ tests reproduces basic results.
- Both feature-finding and trajectory-linking can be performed on arbitrarily long videos using a fixed, modest amount of memory. (Results can be read and saved to disk throughout.)
- A prediction framework helps track particles in fluid flows, or other scenarios where velocity is correlated between time steps.
- Feature-finding and trajectory-linking works on images with any number of dimensions, making possible some creative applications.
- Uncertainty is estimated following a method described in this paper by Savin and Doyle.
- High-performance components (numba acceleration and FFTW support) are used only if if available. Since these can be tricky to install on some machines, the code will automatically fall back on slower pure Python implementations as needed.
The examples linked to above are the best place to start. To try them out on your own computer, you will want to have the sample data as well; you can download all of the examples and data from the examples repository. There is also complete documentation for every function in the package.
Installation is simple on Windows, OSX, and Linux, even for Python novices.
To get started with Python on any platform, download and install Anaconda. It comes with the common scientific Python packages built in.
If you are using Windows, I recommend 32-bit Anaconda even if your system is 64-bit. (One of the optional dependencies is not yet compatible with 64-bit Python.)
Open a command prompt. That's "Terminal" on a Mac, and "Start > Applications > Command Prompt" on Windows. Type or paste these lines to make certain that Anaconda will work well with trackpy:
conda update conda
conda install numpy=1.8 scipy=0.14.0 matplotlib=1.3 pandas=0.13.0 scikit-image=0.10.1 pyyaml numba=0.12.2
conda install pip
Then, to install trackpy:
pip install trackpy
Finally, to try it out, type
ipython notebook
This will automatically open a browser tab, ready to interpret Python code. To get started, check out the links to tutorials at the top of this document.
You can install any of the dependencies using pip, but we recommend starting with Anaconda, which comes with several of the essential dependencies included. Canopy is another distribution that makes a good starting point.
Essential Dependencies:
You will also need the image- and video-reader PIMS, which is, like trackpy itself, part of the github.com/soft-matter organization.
You can install PIMS and trackpy using pip:
pip install pims
pip install trackpy
Or, if you plan to edit the code, you can install them manually:
git clone https://github.com/soft-matter/pims
pip install -e pims
git clone https://github.com/soft-matter/trackpy
pip install -e trackpy
Optional Dependencies:
pyFFTWto speed up the band pass, which is one of the slower steps in feature-findingPyTablesfor saving results in an HDF5 file. This is included with Anaconda.numbafor accelerated feature-finding and linking. This is included with Anaconda and Canopy. Installing it any other way is difficult; we recommend sticking with one of these. Note thatnumbav0.12.0 (included with Anaconda 1.9.0) has a bug and will not work at all; if you have this version, you should update Anaconda. We support numba versions 0.11 and 0.12.2.
PIMS has its own optional dependencies for reading various formats. You can read what you need for each format here on PIMS` README.
The code is under active development. To update to the current development version, run this in the command prompt:
pip install --upgrade http://github.com/soft-matter/trackpy/zipball/master
See the releases page for details.
The original release is tagged Version 0.1. Although there have been major changes to the code, v0.2.x maintains complete reverse compatibility with v0.1 and can be used as drop-in replacement. We recommend all users upgrade.
The master branch on github contains the latest tested development code.
Changes are thoroughly tested before being merged. If you want to use the
latest features it should be safe to rely on the master branch.
(The primary contributors do.)
Roadmap:
- expansion of data structures to simplify sharing frame-wise data between research groups
- interactive filtering and visualization tools
- continued performance improvments and benchmarking for a range of use cases (frame size, particle density, etc.)
- tests that compare results again "battle-tested" Crocker-Grier code
- Daniel Allan feature-finding, uncertainty estimation, motion characterization and discrimination, plotting tools, tests
- Thomas Caswell multiple implementations of sophisticated trajectory-linking, tests
- Nathan Keim alternative trajectory-linking implementations, major speed-ups, prediction
If you use trackpy for published research, please cite this repository,
including the primary contributors' names -- Daniel B. Allan, Thomas A. Caswell,
and Nathan C. Keim -- and doi:10.5281/zenodo.9971.
If your citation style also allows for a URL,
please include github.com/soft-matter/trackpy to help other
researchers discover trackpy. Our
DOI record page
provides more detail and citations in various formats.
| Author(s) | Project URL | Programming Language |
|---|---|---|
| Crocker and Grier | http://physics.nyu.edu/grierlab/software.html | IDL |
| Crocker and Weeks | http://www.physics.emory.edu/~weeks/idl/ | IDL |
| Blair and Dufresne | http://physics.georgetown.edu/matlab/ | MATLAB |
| Maria Kilfoil | http://people.umass.edu/kilfoil/downloads.html | MATLAB and Python |
| Graham Milne | http://zone.ni.com/devzone/cda/epd/p/id/948 | LabVIEW |
| Ryan Smith and Gabe Spalding | http://titan.iwu.edu/~gspaldin/rytrack.html | stand alone/IDL GUI |
| Peter J Lu | https://github.com/peterlu/PLuTARC_centerfind2D | C++ (identification only) |
| Thomas A Caswell | https://github.com/tacaswell/tracking | C++ |
This package was developed in part by Daniel Allan, as part of his PhD thesis work on microrheology in Robert L. Leheny's group at Johns Hopkins University in Baltimore, MD. The work was supported by the National Science Foundation under grant number CBET-1033985. Dan can be reached at dallan@pha.jhu.edu.
This package was developed in part by Thomas A Caswell as part of his PhD thesis work in Sidney R Nagel's and Margaret L Gardel's groups at the University of Chicago, Chicago IL. This work was supported in part by NSF Grant DMR-1105145 and NSF-MRSEC DMR-0820054. Tom can be reached at tcaswell@gmail.com.
This package was developed in part by Nathan C. Keim, as part of his postdoctoral research in Paulo Arratia's group at the University of Pennsylvania, Philadelphia. This work was supported by NSF-MRSEC DMR-1120901.