Release v16.0.11 - SFSORT

• Integrated SFSORT
• Added SFSORT to CI and Benchmark for automatic metric generation

Release v16.0.10

• single process, batched, multi-sequence detections and embeddings generation. Instead of multi-process, one per sequence
• resume enabled for detection and embeddings generation
• enable tuning for visdrone dataset
• unified inference engine for detection and embeddings generation
  • unified inference engine used by both tracking and evaluation
  • which enables boxmot track timers in boxmot eval

Release v16.0.9 - Visdrone eval support

• Fix visdrone evaluation bug (#2214)
• Improve custom dataset evaluation flow by centralizing classes to eval and distractor classes in configs

Release v16.0.8 - Transformers RTDetr support

Add RTDetr from transformers (https://huggingface.co/PekingU/rtdetr_v2_r50vd) to detectors

Release v16.0.7

• BugFix: GBRC postprocessing

Release v16.0.6

Fix on-the-fly installs in cloned and pip installed repos

Release v16.0.5

• Fix fatal evaluation bug
• Fix fatal tuning bug

Release v16.0.4

• Refactor lost tracks visualization class into several classes:
  • trackers that only expose active tracks
  • trackers that maintain explicit lists for lost and removed tracks
• Activate the lost tracks visualization by --show-lost
• Improved quick examples in README

Release v16.0.3 - Lost tracks plotting

Add lost tracks display for Kalman Filter debugging. Try it by:

boxmot track yolo12x lmbn_n_duke hybridsort --source 0 --classes 0 --verbose --save --show --show-trajectories

stacked_vertical.mp4

Release v16.0.2 - Multi-class evaluation

• Fully integrated multi-class evaluation. You have two options:

  • Create a file called custom.yaml under boxmot/configs/datasets/

    # https://motchallenge.net/data/MOT17/
    download:
      runs_url: ""
      dataset_url: ""
    
    benchmark:
      source: "assets/MOT17-mini"      # relative dataset location to boxmot's root
      split: "train"                   # set the split 
      classes: "person car bicycle"    # set the classes to evaluate on

  • Run this command with the classes to evaluate on, for example --classes 0,2:

    boxmot eval --yolo-model yolox_x_MOT17_ablation.pt --reid-model lmbn_n_duke.pt --tracking-method botsort --source MOT17-ablation --classes 0,2

• Fully integrated multi-class tuning (by optimizing for averaged metrics over all classes)

Big thanks to @blaisefinix for providing guidelines to solve this (#2118)