# `TrackForgex.Trackers.ByteTrack`
[🔗](https://github.com/monoflow-ayvu/track_forgex/blob/v0.3.0/lib/trackers/byte_track.ex#L1)

ByteTracker is a state-of-the-art multi-object tracking (MOT) system that achieves high performance
by associating every detection box across video frames, including low-confidence ones.
Instead of discarding uncertain detections, ByteTracker uses them to improve tracking accuracy,
especially for occluded or partially visible objects.

The system works in a tracking-by-detection paradigm: an object detector first identifies objects
in each frame, producing bounding boxes with confidence scores. ByteTracker then applies a two-phase
association process. In the first phase, high-confidence detections are associated with existing tracks
using motion prediction (Kalman filter) and appearance similarity (ReID features).
In the second phase, low-confidence detections are associated with unconfirmed or lost tracks to
recover missed objects.

This approach allows ByteTracker to maintain consistent tracks even during occlusions and achieves
strong performance metrics, such as 80.3 MOTA and 77.3 IDF1 on the MOT17 benchmark while
running at 30 FPS on a single GPU.

# `new`

```elixir
@spec new(TrackForgex.Trackers.ByteTrack.Settings.t()) :: reference()
```

Create a new ByteTracker instance.

## Examples

```elixir
iex> settings = %TrackForgex.Trackers.ByteTrack.Settings{track_thresh: 0.5, track_buffer: 30, match_thresh: 0.8, det_thresh: 0.6}
iex> byte_track = TrackForgex.Trackers.ByteTrack.new(settings)
iex> is_reference(byte_track) == true
```

# `update`

```elixir
@spec update(reference(), [TrackForgex.Utils.Detection.t()]) :: [
  TrackForgex.Trackers.ByteTrack.DetectionResult.t()
]
```

Update the ByteTracker instance with new detections.

## Examples

```elixir
iex> settings = %TrackForgex.Trackers.ByteTrack.Settings{track_thresh: 0.5, track_buffer: 30, match_thresh: 0.8, det_thresh: 0.6}
iex> byte_track = TrackForgex.Trackers.ByteTrack.new(settings)
iex> detections = [%TrackForgex.Utils.Detection{bbox: %TrackForgex.Utils.BBox{x: 0.0, y: 1.0, w: 0.0, h: 0.0}, score: 0.9, class_id: 0}]
iex> results = TrackForgex.Trackers.ByteTrack.update(byte_track, detections)
iex> results # list of DetectionResult structs
```

---

*Consult [api-reference.md](api-reference.md) for complete listing*