Release v10.0.81

tracking per class decorator moved into basetracker

Release v10.0.80

max_time_lost now equal to buffer_size in BoTSORT and ImprAssoc, following their original implementations (#1608)

Release v10.0.79

Add missing virtual trajectory tracking over occlusion periods for OCSORT, HybridSORT and DeepOCSORT (#1601)

Release v10.0.78

plot_results now takes in thickness and font_scale by @LanluZ

Release v10.0.77

Make it possible to pass embs externally to trackers:

from boxmot.tracker_zoo import create_tracker
import numpy as np

model = create_tracker(
      'botsort',
      tracker_config=config_path,
      reid_weights=Path('osnet_x0_25_msmt17.pt'),
      device='cpu',
      half=False,
      per_class=False
)

dets = np.random.rand(5, 6)
rgb = np.random.randint(255, size=(640, 640, 3), dtype=np.uint8)
embs = np.random.rand(5, 512)

outputs = model.update(
    dets,
    img,
    embs
)

Release v10.0.76

Merge pull request #1556 from mikel-brostrom/pass-embs

pass external embs to trackers

Release v10.0.75

• Noise covariance search enabled for OCSORT and DeepOCSROT by @edblu1 in (#1484)
• Added Improved Association Pipeline Tracker by @rolson24 in (#1527)
• Refactor requirements checker
• Complete exporters refactor
• Complete auto-backend refactor
• Complete Kalman Filters refactor
• Enabled TFLite ReID export and inference
• Implemented benchmarking in CI for all trackers on very small subset of MOT17

Release v10.0.73

• Implemented dynamic deque lenghts based on max_age (fixes #1480)
• CPU torch & torchvision CI installation by simple sed command on pyproject.toml file in CI pipeline
• Enable MPS / CUDA detection and embedding generation by @johnnynunez (#1476)
• Fix MacOS detection and embedding generation by @johnnynunez (#1475)

Release v10.0.72

CUDA/MPS enabled torch and torchvision installed by default

Release v10.0.69

Fix pkg version fetch in boxmot