Skip to content

Migration Guide: tryolabs/norfair → norfair-enough#

norfair-enough is a fork of tryolabs/norfair, based on upstream v2.3.0. This guide describes what you need to change when moving an existing project from the upstream package to norfair-enough.

This fork is no longer maintained

v2.4.3 (August 2026) is the final release and the repository is archived. The package remains installable and working, but nothing further will be released. This guide is kept for anyone migrating onto that final version, or forking it to carry it forward.

TL;DR#

  • Uninstall norfair, install norfair-enough.
  • The import path stays the same: from norfair import ....
  • Make sure you are on Python 3.10 or newer.
  • Most code continues to work unchanged; a few APIs have been deprecated or behavior has been tightened.

Installation#

Replace the upstream package with the fork:

1
2
pip uninstall norfair
pip install norfair-enough

Do not install both packages at the same time

Both norfair and norfair-enough ship the same top-level Python module (norfair). Installing both into the same environment will lead to undefined import behavior. Always uninstall one before installing the other.

Imports do not change — norfair-enough keeps the original module name:

1
from norfair import Detection, Tracker, draw_points

If you install from source, use the fork's repository URL:

1
pip install "git+https://github.com/akator-de/norfair-enough.git@main#egg=norfair-enough"

Python version#

norfair-enough requires Python 3.10 or newer.

Upstream norfair historically supported earlier Python versions (3.8/3.9). If your project is still on Python 3.8 or 3.9 you will need to upgrade your interpreter before migrating. The fork follows the upstream decision (v2.4.0) to drop older Python versions — see CHANGELOG.md.

The CI matrix for norfair-enough covers Python 3.10 through 3.13.

Dependency changes#

  • filterpy is no longer a dependency. The fork ships an internal Kalman filter implementation derived from the original FilterPy KalmanFilter class (MIT, Roger R. Labbe Jr.). FilterPyKalmanFilterFactory still exists for backwards compatibility but now uses the internal implementation — you can remove filterpy from your own requirements.
  • NumPy 2.x is supported alongside NumPy 1.23+. No code change required on your side.

Deprecated drawing APIs#

The following drawing helpers are deprecated in norfair-enough but still work. They now emit a deprecation warning on first use. Migrate at your convenience:

Deprecated Replacement
draw_tracked_objects(...) draw_points(...)
draw_tracked_boxes(...) draw_boxes(...)

Both replacements accept tracked objects directly, so most call sites only need a rename.

Additional deprecated parameters on draw_points:

  • color_by_label=True → use color="by_label"
  • detections=... → use drawables=...
  • label_size=... → use text_size=...

Additional deprecated parameters on draw_boxes:

  • color_by_label=True → use color="by_label"
  • detections=... → use drawables=...
  • label_size=... → use text_size=...
  • random_color=True → use color="random"
  • line_color=... → use color=...
  • line_width=... → use thickness=...

See the [Drawing reference][norfair.drawing] for the full current API.

Behavior changes#

Tracker.update() may mutate Detection objects -- behavior now documented#

In both upstream norfair and norfair-enough, Tracker.update() may mutate attributes on the Detection instances you pass in (for example detection.absolute_points when coord_transformations is provided, and detection.age when detections are mapped to tracked objects).

The difference in norfair-enough is that this mutation behavior is now documented explicitly in the docstrings for Tracker.update() and Detection. If your code reuses Detection objects after calling Tracker.update() -- for drawing, logging, or further processing -- be aware of these mutations and review your code accordingly. In most real-world pipelines this is invisible.

Stricter input validation#

  • Tracker and related APIs now raise ValueError for invalid input that previously triggered assert statements. If you run Python with -O, the previous assertions were silently skipped; now the checks always run.
  • The iou distance function validates both candidates and objects (previously only one side was checked).

Other notable fixes#

These are bug fixes relative to upstream v2.3.0. If your code accidentally depended on the buggy behavior you may see different results:

  • draw_tracked_objects now returns the frame (previously returned None).
  • distance_matrix.any() no longer incorrectly skips zero distances during matching.
  • hex_to_bgr now accepts uppercase hex color strings.
  • PredictionsTextFile fixes file-handle leaks.
  • AbsolutePaths fixes memory leaks for destroyed objects.
  • Accumulators.update fixes quadratic array growth.
  • Video output filenames now handle paths and file extensions portably; it also supports the context manager protocol (with Video(...) as v:) and exposes a close() method for reliable resource cleanup.
  • TrackedObject.global_id counter is now thread-safe.

For the full list of fixes and changes, see CHANGELOG.md.

Newly exported symbols#

The following symbols are now exported directly from the top-level norfair package and no longer need to be imported from submodules:

  • TrackedObject
  • Distance
  • FilterFactory
  • ColorLike
  • camera motion classes
  • metrics classes

Old imports continue to work.

Repository differences#

  • Demos: the fork ships modernized demos under demos/. Some have been updated to newer model versions and dependencies.
  • Tooling: norfair-enough uses ruff for linting and pytest for tests, with a CI matrix covering Python 3.10–3.13.
  • Packaging: published on PyPI as norfair-enough.

Staying on upstream tryolabs/norfair#

Upstream tryolabs/norfair has had no commits since April 2025. If you prefer to stay on it, you can continue to install it from its repository or PyPI — but be aware that bug fixes and new Python/NumPy compatibility work will not flow back.

Both projects are now dormant. norfair-enough v2.4.3 carries the fixes listed above and requires Python 3.10+ (tested through 3.13), NumPy 2.x and OpenCV 5, so it is the further-along of the two — but if you need something under active development, the alternatives listed in the README are the better starting point.