Samba Dancing

The original animation was obtained from Adobe’s Mixamo animation library.

Reconstruction Performance

The 3D viewer presents the original animation (right-most, with orange indicator overhead) along with a series of animations constructed using Salient Poses. See the table below for further information.

Animation Total Frames Keyframes Compression Maximum Error (mm) Average Error (mm)
Compression 4 1092 27 97.53% 472.94 152.72
Compression 3 1092 55 94.96% 245.96 38.61
Compression 2 1092 109 90.02% 141.66 6.16
Compression 1 1092 218 80.04% 23.42 1.16
Original 1092 1092 00.00% 0.00 0.00


Approximation Error: Greedy vs Optimal

A comparison between Salient Poses and a greedy approximation algorithm showing how approximation error reduces as keyframes are added.

Reconstruction Accuracy: SP vs Maya vs PCA

A comparison of how effectively animations can be reconstructed over a range of compression levels using Salient Poses, PCA (over the poses, not the curves), and Maya’s Simplify Curves algorithm.


The time required for the algorithm to obtain a selection of N keyframes.