PointSSIM: Point cloud structural similarity metric

Accurate quality predictors are essential for every type of imaging modality. In the case of point clouds, several objective quality metrics have been recently proposed. In this study [1], we extend previous efforts by generalizing the quality prediction process, and exploring a higher dimensional feature space including geometry, normals, curvatures and color attributes. In some aspect, we explore the applicability of the well-known SSIM in a higher-dimensional, irregular space (volumetric content), incorporating geometrical and textural information.

The proposed methodology is full-reference and provides structural similarity scores for a point cloud under evaluation, when compared to a reference. A structural similarity score is obtained per attribute. It is computed by pooling across the complement of 1 to an error map, which indicates the relative difference of associated feature maps extracted from the point clouds under comparison. The feature maps are computed using statistical dispersion estimators, applied on quantities that reflect attribute properties in local neighborhoods.

Outline of point cloud similarity working principle.
The depicted point cloud is part of the
 PointXR dataset (original model source: link, creator: Thomas Flynn, license: CC Attribution)

Auxiliary scripts that are used for pre-processing are also part of the provided material. For example, the implementation of voxelization, which can be optionally enabled prior to feature extraction similarly to downsampling in multi-scale quality metrics for 2D imaging (e.g., MS-SSIM), leading to improvements in prediction accuracy under certain conditions.

The algorithmic steps that are employed in the proposed algorithm allow for different configurations, and facilitate the integration of new quantities, attributes and estimators.

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


Scripts with prototype implementations for the computation of structural similarity scores and the voxelization of point clouds, can be downloaded from the following GitHub repository.

Conditions of use

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

  1. E. Alexiou and T. Ebrahimi, “Towards a Point Cloud Structural Similarity Metric,” 2020 IEEE International Conference on Multimedia & Expo Workshops (ICMEW), London, United Kingdom, 2020, pp. 1-6. doi: 10.1109/ICMEW46912.2020.9106005