v2.2.0

Release notes

We release Norfair 2.2.0, which includes a major improvement to distances performance as well as many minor fixes and improvements. Including a refactoring of drawing

This version also drops support for Python 3.6.

Changelog

Features

• New vectorized distance functions. (#211) @facundo-lezama
• Added optimized IoU function (#226) @facundo-lezama
• Added the flag draw_scores to the function draw_boxes. (#219) @moooises
• Enable users to draw custom objects (#229) @javiber
• Add estimate_velocity property to the TrackedObject class (#238) @DiegoFernandezC

Demos

• Introduce Norfair ROS in the README (#231) @DiegoFernandezC

Documentation

• Fixed typehint of draw_points and draw_boxes (#230) @javiber
• update sahi version (#217) @fcakyon

Other features & fixes

• Drop support for Python 3.6. (#234) @facundo-lezama
• Drawing refactor (#209) @javiber
• Fix reid_hit_counter initialization. (#224) @facundo-lezama
• Automatically deploy new doc versions (#239) @javiber
• Make numpy a direct dependency of Norfair. (#233) @facundo-lezama
• New issue templates (#206) @javiber
• Set Ubuntu version to 20.04 in CI. (#222) @facundo-lezama
• Fixed draw_box and removed infinite distance check on Tracker (#220) @javiber
• Fixed a problem with the hex color parsing (#215) @javiber

v2.1.1

Release notes

We release this patch release to fix a missing dependency on Python 3.6 and 3.7. A new demo for panoptic perception and some updates to sahi demo and documentation are included as well.

Changelog

Fixes

• Fix importlib-metadata dependency bug affecting python 3.6 and 3.7 (#212) @javiber

Demos

• Add panoptic driving perception demo using YOLOPv2 (#204) @DiegoFernandezC
• Update sahi demo to use their latest version (#214) @fcakyon

Docs

• Add Getting started guide (#208) @javiber

v2.1.0

Release notes

We release Norfair 2.1, with support for Python 3.10. The rest is mostly small features and fixes, and a new demo.

Changelog

Features

• Support Python 3.10 (#195) @dekked
• Support for object count when using multiple Tracker instances (#196) @javiber
• Added optional quality_level field to MotionEstimator class (#200) @facundo-lezama

Demos

• Add small object tracking demo using SAHI (#188) @JuanFKurucz

Documentation

• Added Contributing guide (#205) @javiber
• Documentation site for different Norfair versions (#201) @javiber
• Added documentation for drawers related to camera motion (#203) @javiber

Other features & fixes

• Update MOT17 metrics and add MOT20 results (#202) @facundo-lezama
• Version has single source of truth in pyproject.toml (#187) @dekked
• Fixed bug when transforming coordinates of 1-rank detections (#198) @javiber
• Fixed the formatting of warning strings (#194) @javiber
• Fix Python 3.6 dependency in PyPI & upgrades in pyproject.toml (#191) @dekked
• Fix broken link to documentation in error message (#192) @dekked
• tox works again for running tests locally (#193) @dekked

v2.0.0

Release notes

We are excited to announce the release of Norfair 2.0, the biggest upgrade to Tryolabs’ open-source multi-object tracking library since its first release two years ago 🙌

Read announcement blog post.

Changelog

New features

• Support re-identification with appearance embeddings (#118) @facundo-lezama
• Estimate camera motion using the mode of the Optical Flow (#139) @aguscas
• Support n-dimensional tracking (#138) @aguscas
• Add new documentation using MkDocs (#154) @javiber
• Added option to control output video extension (#176) @javiber
• Added predefined distance functions for typical cases (#135) @javiber

Demos

• Add demo in Hugging Face Spaces (#178) @DiegoFernandezC @dekked
• Add official demo in Google Colab (#184) @DiegoFernandezC
• Revamp demos and dockerize them for reproducibility (#146) @aguscas @dekked
• Add YOLOv7 demo (#147) @aguscas
• Added example with MMDetection (#134) @javiber
• Update OpenPose demo (#81) @rocioxl
• Added IoU function to the YOLOv5 demo (#90) @ffedee7

Other features & fixes

• Set optimized filter as default (#145) @aguscas
• Add version info in __init__.py (#123) @fcakyon
• Blackify the whole repo (#141) @javiber
• Fixed reid video (#179, #180) @javiber
• Avoid drawing paths of dead objects (#175) @javiber
• Fixed dependency on GitHub Actions (#174) @javiber
• Fixes in YOLO demos when tracking bounding boxes (#161) @javiber
• Fix dependency on older version of rich (#160) @javiber
• Fixed draw_tracked_boxes when draw_box is False (#150) @javiber
• Fix many initializations in the trackers (#142) @aguscas
• Fix profiling demo (#137) @aguscas
• Unit tests refactor (#162) @javiber
• Remove redundant CI lines (#116) @joaqo
• Replace gifs for videos on demos (#177) @javiber

v1.0.0

Release notes

One year and 8 months after the first public release, we got the API stable enough to be comfortable releasing Norfair 1.0! 🥂

Changelog

• Simplify API (#106) @aguscas
  • hit_inertia_min -> Removed
  • hit_inertia_max -> hit_counter_max
  • point_transience -> pointwise_hit_counter_max
• Re-add support for Python 3.6 (#113) @joaqo
• Add 'No filter' and 'Optimized KF' setups (#108) @aguscas
• Add mot metrics GitHub action (#114) @joaqo
• Create scripts to compare norfair and ByteTrack on MOT17 (#100) @aguscas
• Profiling demo (#110) @aguscas

v0.4.3

Changelog

• Add demo tracking keypoints and bounding boxes using OpenPose and YOLOv5 (#99) @facundo-lezama
• Add documentation for draw_boxes and draw_tracked_boxes (#101) @aguscas
• Fix bug by removing reference to 'points_of_interest' (#95) @aguscas
• Add support for path drawing in video (#89) @aguscas
• Added labels to the drawing functions (#93) @facundo-lezama
• Added feature for tracking objects with different classes (#91) @facundo-lezama

v0.4.0

Changelog

• Fix yolov4 demo (#87) @joaqo
• Update metrics docs (#86) @joaqo
• Improve docs (#85) @joaqo
• Update dependencies due to Pillow<0.9 containing vulnerabilities (#83) @joaqo
• ✏️ Fixed argument parsing (#78) @Huh-David
• Add support for storing several past detections on TrackedObjects (#74) @joaqo

v0.3.1

Changelog

• Bug fix in index for prediction score in YOLOv5 demo (#58) @fcakyon
• fix typo in docs (#57) @fcakyon

v0.3.0

Changelog

New features

• [Backwards incompatible] Support custom list of colors in draw_tracked_boxes (renames line_color to border_colors and line_width to border_width). (#54) @aguscas
• Support receiving custom Kalman filters on Tracker (#53) @aguscas

Documentation & demos

• Add demo of tracking cars and pedestrians using YOLOv5 (#52) @fcakyon
• Add table with MOT17 metrics to Readme (#43) @aguscas

v0.2.0

Changelog

• Make OpenCV and motmetrics optional dependencies (#40) @joaqo
• Small refactor to metrics (#33) @aguscas
• Add support for evaluating trackers on the MOT Challenge dataset (#17) @aguscas