/fantrack

Please refer to original repo. This is for personal learning only.

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

FANTrack: 3D Multi-Object Tracking with Feature Association Network

Requirements

Please refer to this requirement.txt

Environment Setup

Please refer to this FanTrack_Instructions_Github.ipynb

Training

  1. AVOD FANTrack requires trained weights for AVOD. You can find our pre-trained weights avod_cars_fast.zip (138 Mb) and avod_people_fast.zip (163 Mb). Unzip this folder into fan_track/object_detector/avod. It should produce a folder fan_track/object_detector/avod/data. Alternatively, you can train AVOD yourself using the AVOD instructions.

  2. SimNet

Train simnet by running:

$ python3 fan_track/experiments/train_simnet.py

Checkpoints and data will be saved to fan_track/data/simnet

  1. AssocNet

Train assocnet by running:

$ python3 fan_track/experiments/train_simnet.py
  1. Run tracker
$ python3 fan_track/experiments/run_tracker.py