Release v10.0.23

Added MANIFEST.in for including yamls and the CLIP vocab .txt.gz

Release v10.0.22

Fix fatal YoloX and YoloNAS bug: ValueError: tile cannot extend outside image

Release v10.0.21

Fix several fatal import bugs

Release v10.0.20

lower coverage for pass

Release v10.0.19

• GSI (https://arxiv.org/abs/2202.13514) implemented
• CLIPReID (https://arxiv.org/abs/2211.13977) implemented
• Yolo detectors refactor
• Delete releases with fatal bugs
  • Fix KF import bug in 10.0.17
  • Fix postprocessing import bug in 10.0.18
• Fix DeepOCSORT association bug when linear association fails
• Enable ECC camera motion compensation by default in StrongSORT
• Fixed eval bug by which the eval process could not be parallelized for each sequence
• CMC methods refactored instead of single class for better code maintainability
• StrongSORT CMC redundancy fixed by @chanwutk in #1010
• More tests added by @henriksod in #1001
• pre-commit added by @henriksod in #1001
• KF adapters added by @henriksod in #1001

Release v10.0.16

• MOT results now saved under EXAMPLES/exp*
• Class filtering for YoloX and Yolo-NAS provided by @saurabheights in #979
• Added YOLOX ByteTrack people models for paper reprodusibility
• Major BoxMOT refactor by @henriksod in #986
• Applied strategy design pattern to multi yolo for easier code mantainability
• Tracking per class enabled for OCSORT and DeepOCSORT. Mostly based on @SceneraPieter work here: #997. Usage example:
• • python3 examples/track.py --tracking-method ocsort --per-class
• • python3 examples/track.py --tracking-method deepocsort --per-class

Release v10.0.15

• Added more in depth custom model example
• Assert inputs types (numpy.array)
• Assert input shapes to trackers
• • dets: (N, 6)
• • im: (h, w, 3)
• enable save mot or dets results only. Previously both dets and mot results had to be saved
• crops per class and id by: python3 examples/track.py --save-id-crop

Release v10.0.14

• All prints substituted by LOGGER
• Always print environment info, followed by args
• Fix passing multiple classes to val.py and evolve.py for multi-class evaluation

Release v10.0.13

• Inputs (det, im0) to all trackers are now numpy.ndarray
• Fix README custom model example
• Clean up classifiers in setup.py
• Add trackers instantiation pytest tests
• Lower CI time by ~2 minutes by using a more CPU friendly architecture

Release v10.0.12

• Fix google colab example
• Less verbose evaluation
• The evaluation process stops if a single subprocess fails.