3D mesh animation compression based on adaptive spatio-temporal segmentation

Guoliang Luo, Zhigang Deng, Xiaogang Jin, Xin Zhao, Wei Zeng, Wenqiang Xie, Hyewon Seo

Introduction

3D mesh animations, which represent dynamic objects through thousands of vertices changing coordinates over time, are essential for modern digital media but present significant data storage and streaming challenges. Traditional compression techniques often rely on Principal Component Analysis (PCA) to reduce dimensionality. However, PCA's effectiveness is limited when applied to long, complex animations where motion is not globally coherent. This paper introduces an adaptive framework that automatically detects optimal spatial and temporal boundaries to create segments that are highly compressible.

Adaptive Segmentation Framework

The core contribution of this work is a pipeline that adaptively partitions mesh animations into spatio-temporal blocks to maximize PCA efficiency. The process is divided into several key stages as illustrated in the system flowchart.

The proposed adaptive spatio-temporal segmentation pipeline.

1. Initial Temporal Cut

The system first analyzes short windows of frames using kernelized Canonical Correlation Analysis (kCCA). This technique identifies distinct changes in dynamic behavior, establishing initial temporal boundaries that separate different types of motion.

2. Two-Stage Vertex Clustering

Spatial segmentation is performed to group vertices that exhibit similar movement patterns. As shown in the visualization below, the process begins by identifying "rigid" areas versus those that undergo significant deformation. These regions are then iteratively refined into a user-defined number of clusters ($\omega$) to minimize reconstruction error.

Visualization of the vertex clustering process showing rigidity detection and spatial segmentation.

3. Refined Temporal Segmentation and PCA

Within each spatial cluster, the algorithm performs a final temporal refinement by monitoring the "Principal Component curve." This identifies the exact frames where motion complexity changes, ensuring each segment is as coherent as possible before being compressed via PCA. Because these segments are localized in both space and time, they can be represented with significantly fewer bits than global or fixed-block approaches.

Experimental Results

The researchers evaluated the framework on diverse datasets, including humanoid motions ('March', 'Jump'), animal deformations ('Horse'), and complex dynamic surfaces ('Cloth', 'Flag').

  • Efficiency and Quality: The proposed method achieved lower reconstruction distortion (measured by KGError) at the same bitrate (bits per vertex per frame, or bpvf) compared to standard adapted PCA methods. It was particularly effective at preserving high-frequency details, such as fine wrinkles in cloth, which are often blurred by traditional methods.
  • Computational Performance: By processing spatial vertex groups independently, the framework is highly parallelizable. In tests on the 'Cloth' dataset, the adaptive method completed in approximately 14.5 seconds, whereas global iterative PCA optimization required over 4,400 seconds.

Conclusion

By jointly optimizing spatial and temporal boundaries, this adaptive framework significantly enhances the compressibility of 3D mesh animations without sacrificing visual fidelity. While the current implementation offers substantial speed improvements over global methods, future work will focus on developing a sliding window update scheme to support real-time streaming applications.

Summary updated

Abstract

With the recent advances of data acquisition techniques, the compression of various 3D mesh animation data has become an important topic in computer graphics community. In this paper, we present a new spatio-temporal segmentation-based approach for the compression of 3D mesh animations. Given an input mesh sequence, we first compute an initial temporal cut to obtain a small subsequence by detecting the temporal boundary of dynamic behavior. Then, we apply a two-stage vertex clustering on the resulting subsequence to classify the vertices into groups with optimal intra-affinities. After that, we design a temporal segmentation step based on the variations of the principle components within each vertex group prior to performing a PCA-based compression. Our approach can adaptively determine the temporal and spatial segmentation boundaries in order to exploit both temporal and spatial redundancies. We have conducted many experiments on different types of 3D mesh animations with various segmentation configurations. Our comparative studies show the competitive performance of our approach for the compression of 3D mesh animations.
BibTeX
@inproceedings{luo2019d,
  author = {Guoliang Luo and Zhigang Deng and Xiaogang Jin and Xin Zhao and Wei Zeng and Wenqiang Xie and Hyewon Seo},
  title = {{3D mesh animation compression based on adaptive spatio-temporal segmentation}},
  booktitle = {I3D},
  year = {2019},
  doi = {10.1145/3306131.3317017},
}
Figure