Visual features and motion model are entangled

quantumjot / btrack

Bayesian multi-object tracking

https://btrack.readthedocs.io

MIT License

311 stars 50 forks source link

Closed quantumjot closed 1 year ago

quantumjot commented 1 year ago

After taking a look at the tracking with visual features only we realised a few things:

Motion updates happen to dummy objects even if we're not using the motion model
The bit masks to determine whether to use the motion model/visual model might be incorrect

Anything I'm missing @lupinthief?

lupinthief commented 1 year ago

Don't think so