COOLER

[GCPR'23 Oral] COOLer: Class-Incremental Learning for Appearance-Based Multiple Object Tracking

Installation

python3 -m pip install -r requirements.txt
python3 setup.py develop

Note that this project depends on other projects:

• scalabel
• bdd100k

The training framework in based on Pytorch and PytorchLightning. And the detection models are based on mmdet. Please create a python environment and install specific versions of these packages suitable for your hardware.

Download the BDD100k tracking dataset to data/bdd100k and the SHIFT tracking dataset to data/shift_dataset.

Running CL Experiments

Run BDD experiments:

bash run_scripts/run_bdd.sh  run_scripts/bdd_cfgs/most_to_least.toml --config configs/experiments/bdd/PL_GT_contrast/cooler_ct_loss.toml --exp_name most_to_least ## most to least setting
bash run_scripts/run_bdd.sh  run_scripts/bdd_cfgs/general_to_specific.toml --config configs/experiments/bdd/PL_GT_contrast/cooler_ct_loss.toml --exp_name general_to_specific ## general to specific setting
bash run_scripts/run_bdd.sh  run_scripts/bdd_cfgs/vehicle_bike_person.toml --config configs/experiments/bdd/PL_GT_contrast/cooler_ct_loss.toml --exp_name vehicle_bike_person ## vehicle to bike to human setting

Run SHIFT experiments:

bash run_scripts/run_shift.sh  run_scripts/shift_cfgs/most_to_least.toml --config configs/experiments/clear_daytime_shift/PL_GT_contrast/cooler_ct_loss.toml --exp_name most_to_least ## most to least setting
bash run_scripts/run_shift.sh  run_scripts/shift_cfgs/general_to_specific.toml --config configs/experiments/clear_daytime_shift/PL_GT_contrast/cooler_ct_loss.toml --exp_name general_to_sepcific ## general to specific setting
bash run_scripts/run_shift.sh  run_scripts/shift_cfgs/vehicle_bike_person.toml --config configs/experiments/clear_daytime_shift/PL_GT_contrast/cooler_ct_loss.toml --exp_name vehicle_bike_person ## vehicle to bike to human setting

Usage

Training

python3 -m ilmot.trainer train --config <config_path> <maybe other arguments>

Testing

python3 -m ilmot.trainer test --config <config_path> <maybe other arguments>

Prediction

python3 -m ilmot.trainer predict --config <config_path> <maybe other arguments>

Visualization

python3 -m ilmot.trainer visualize --config <config_path> <maybe other arguments>