NOTICE

LOFOP (Looking For Object Point)
Copyright 2026 DURGAMANI SASIKUMAR and Nishanandhini A.

Licensed under the Apache License, Version 2.0. See LICENSE for the full
text.

================================================================================
Third-party code
================================================================================

LOFOP contains no third-party source code.

Every component in this repository -- the LOFOP-Detect model family
(RidgeNet, SwiftNet, DeltaFusion, ApexHead, StencilHead, VertexHead,
SlateHead), the native C++ and CUDA kernels, the assignment solver, the
tracking subsystem (motion model, association, tracker), the data subsystem,
the training engine, and the deployment path -- is an original implementation
written for this project.

Where LOFOP implements a published algorithm or a standard mathematical
method, it does so from the mathematics or the published description, not by
adapting another project's source. Such methods are not themselves
copyrightable; the implementations here are LOFOP's own expression of them.
This applies in particular to:

  * the recursive linear estimator ("Kalman filter") in lofop/tracking/motion.py,
    implemented from the standard predict/correct equations with LOFOP's own
    state parameterisation, naming, and noise model;
  * the assignment problem solver in lofop/ops/assignment.py, implemented from
    the shortest-augmenting-path formulation;
  * standard convolutional patterns (depthwise separable convolution,
    inverted residual blocks, squeeze-and-excitation gating, feature pyramid
    fusion, focal loss, GIoU), which are published techniques implemented
    here directly.

================================================================================
Optional runtime dependencies
================================================================================

LOFOP does not bundle, vendor, or redistribute any of the following. They are
optional extras installed by the user, and remain under their own licences:

  PyYAML          MIT
  Pillow          MIT-CMU
  PyTorch         BSD-3-Clause          (extra: models)
  numpy           BSD-3-Clause          (extras: tracking, supervision)
  onnx            Apache-2.0            (extra: deploy)
  onnxruntime     MIT                   (extra: deploy)
  tensorboard     Apache-2.0            (extra: tensorboard)
  TensorRT        NVIDIA proprietary    (extra: tensorrt)
  rich            MIT                   (extra: rich)
  supervision     MIT                   (extra: supervision)

The supervision integration in lofop/tracking/supervision_bridge.py converts
between LOFOP's own data structures and supervision's, using only that
library's documented public API. No supervision source code is included.