Quality assessment for point cloud compression

Point clouds denote a viable alternative for immersive media. Considering the huge amount of data that is required for such content representations, efficient compression algorithms are inevitable. In this study [1], the emerging MPEG point cloud codecs (V-PCC and G-PCC variants) are assessed, and best practices for rate allocation are investigated. For this purpose, three experiments are conducted. In the first experiment, a rigorous evaluation of the codecs is performed, adopting test conditions dictated by experts of the group on a carefully selected set of models, using both subjective and objective quality assessment methodologies. In the other two experiments, different rate allocation schemes for geometry-only and geometry-plus-color encoding are subjectively evaluated, in order to draw conclusions on the best-performing approaches in terms of perceived quality for a given bit rate.

In this dataset, we make publicly available quality scores associated with the stimuli under assessment for each experiment. For purposes of reproducibility, a content that was used while not being part of established point cloud repositories adopted by standardisation bodies, is re-distributed. Moreover, scripts are provided in order to generate the reference models and the rendering-related meta-data that were used in this study.

Download

In this URL link you will find the dataset, containing three (3) folders with the related material:

  • In folder contents, you will find the re-distributed model amphoriskos, as prepared for our study, along with a README file with instructions on where to find and how to prepare the reference versions of the rest of the contents.
  • In folder scores, you will find subjective and objective scores from the first experiment, and subjective votes from the second and the third experiments.
  • In folder scripts, you will find the MATLAB functions that were used for voxelization (i.e., fun_voxelize), and computation of the size of each point per stimulus (i.e., fun_pointsize). The former script was used to prepare the reference models of the study (where needed), while the latter script was used to produce meta-data per stimulus that are required by our renderer.

Our rendering software that was developed and used for subjective and objective quality evaluation, is also made publicly available in this GitHub repository.

For more details, the reader can refer to [1].

Conditions of use

If you wish to use any of the provided material in your research, we kindly ask you to cite [1].

Permission is hereby granted, without written agreement and without license or royalty fees, to use, copy, modify, and distribute the data provided and its documentation for research purpose only. The data provided may not be commercially distributed. In no event shall the École Polytechnique Fédérale de Lausanne (EPFL) be liable to any party for direct, indirect, special, incidental, or consequential damages arising out of the use of the data and its documentation. The École Polytechnique Fédérale de Lausanne (EPFL) specifically disclaims any warranties. The data provided hereunder is on an “as is” basis and the École Polytechnique Fédérale de Lausanne (EPFL) has no obligation to provide maintenance, support, updates, enhancements, or modifications.

References

  1. E. Alexiou, E., Viola, I., Borges, T., Fonseca, T., De Queiroz, R., & Ebrahimi, T. (2019). “A comprehensive study of the rate-distortion performance in MPEG point cloud compression. APSIPA Transactions on Signal and Information Processing, 8, E27. doi:10.1017/ATSIP.2019.20