Software

CMDM: Color Mesh Distortion Measure

CMDM is the first metric for quality assessment of 3D meshes with vertex colors, which works entirely on the mesh domain, at vertex level. It is a full-reference data-driven multiscale metric, that incorporates perceptually-relevant curvature-based and color-based features. Our framework is as follows: For given distorted M_dist and reference M_ref meshes, we first establish a correspondence between M_dist and M_ref. Then for each scale h_i, we define a spherical neighborhood around each vertex v of M_dist and compute a set of local geometry and color based features over the points belonging to the neighborhood of v and their corresponding points on M_ref. Local single-scale feature values are pooled into global multiscale features. Finally, CMDM is defined as a linear combination of an optimal subset of features determined through logistic regression.

CMDM code is available in the MEPP2 platforms.
| Manual | Publication | bibtex |

An adapted version for colored point clouds was also developed. It is the first Point Cloud Quality Metric (PCQM) that takes into account both geometry and color.

PCQM code is available online.
| Publication | bibtex |

Graphics-LPIPS: Deep Learning-based Quality Metric for 3D Graphics

Graphics-LPIPS metric is an extension of the LPIPS metric originally designed for images and perceptual similarity tasks, which we adapted for 3D graphics and quality assessment tasks. Graphics-LPIPS employs Convolutional Neural Networks (CNN) with learning linear weights on top. It operates on patches of rendered stimuli that are fed to the CNN to extract features. Features are then fused and pooled to predict the quality of the patch. The overall quality of the 3D model is obtained by averaging local patch qualities

Graphics-LPIPS code is available online.
| Publication | bibtex |