/sixd_toolkit

Python scripts to facilitate participation in the SIXD Challenge.

Primary LanguagePythonMIT LicenseMIT

SIXD Toolkit

Python scripts to facilitate participation in the SIXD Challenge: http://cmp.felk.cvut.cz/sixd/challenge_2017

  • conversion - Scripts used to convert the datasets from the original format to the SIXD standard format.
  • doc - Documentation and conventions.
  • params - Parameters (paths to datasets etc.) used by other scripts.
  • pysixd - Core library that takes care of i/o operations, rendering, calculation of pose errors etc.
  • tools - Scripts for evaluation, rendering of training images, visualization of 6D object poses etc.

Installation

Dependencies

To install the required python packages, run:

pip install -r requirements.txt

In the case of problems, try to run pip install --upgrade pip setuptools first.

Rendering

Rendering is implemented using the Glumpy library and was tested with the GLFW library as the window backend. In Linux, you can install the GLFW library with:

apt-get install libglfw3

To use a different backend library, see the first lines of pysixd/renderer.py.

Evaluation

  1. Run your method on the SIXD datasets and prepare the results in this format.
  2. In params/dataset_params.py set common_base_path to the path of the SIXD datasets. For T-LESS, you will also need to set tless_tk_path to the path of the T-LESS Toolkit.
  3. Run tools/eval_calc_errors.py to calculate errors of the pose estimates (fill list result_paths with paths to the results first).
  4. Run tools/eval_loc.py to calculate performance scores in the 6D localization task (fill list error_paths with paths to the calculated errors first).

Author

Tomas Hodan