Skip to content

Getting started

CPAS Toolbox is a package for evaluation of categorical pose and shape estimation methods. It contains metrics, and wrappers for datasets and methods.

Installation

Run

pip install cpas_toolbox
to install the latest release of the toolbox. There is no need to download any additional weights or datasets. Upon first usage the evaluation script will ask to download the weights if they are not available at the expected path.

Evaluation of baseline methods

To reproduce the REAL275 benchmark run:

python -m cpas_toolbox.evaluate --config real275.yaml all_baselines.yaml --out_dir ./results/
To reproduce the REDWOOD75 benchmark run:
python -m cpas_toolbox.evaluate --config redwood75.yaml all_baselines.yaml --out_dir ./results/

We can overwrite settings of the configuration via the command-line. For example,

python -m cpas_toolbox.evaluate --config redwood75.yaml all_baselines.yaml --out_dir ./results/ --visualize_gt True --visualize_prediction True
enables interactive visualization of ground truth and predictions. Alternatively, you could specify --store_visualization True to save the visualization of every prediction in the results directory.

Citation

If you find this library useful in your research, consider citing our publication:

@article{bruns2022evaluation,
  title={On the Evaluation of {RGB-D}-based Categorical Pose and Shape Estimation},
  author={Bruns, Leonard and Jensfelt, Patric},
  journal={arXiv preprint arXiv:2202.10346},
  year={2022}
}