3DAC: Learning Attribute Compression for Point Clouds

Abstract

We study the problem of attribute compression for largescale unstructured 3D point clouds. Through an in-depth exploration of the relationships between different encoding steps and different attribute channels, we introduce a deep compression network, termed 3DAC, to explicitly compress the attributes of 3D point clouds and reduce storage usage in this paper. Specifically, the point cloud attributes such as color and reflectance are firstly converted to transform coefficients. We then propose a deep entropy model to model the probabilities of these coefficients by considering information hidden in attribute transforms and previous encoded attributes. Finally, the estimated probabilities are used to further compress these transform coefficients to a final attributes bitstream. Extensive experiments conducted on both indoor and outdoor large-scale open point cloud datasets, including ScanNet and SemanticKITTI, demonstrated the superior compression rates and reconstruction quality of the proposed method.

4DAC: Learning Attribute Compression for Dynamic Point Clouds

“…Figure 1: Point cloud attribute compression results on the sequence "Phil" of the MVUB [33] dataset. The state-of-the-art learning-based static attribute compression method, 3DAC [16], is presented with our proposed 4DAC. Bits Per Point (BPP) and Peak Signal-to-Noise Ratio (PSNR) of the luminance component are also reported.…”

“…Alexiou et al [3] and Sheng et al [54] built the transform coding and entropy coding blocks using a 3D auto-encoder with 3D convolution [1] and point convolution [44], respectively. Fang et al [16] presented a learning-based attribute compression framework with RAHT for transform coding and a RAHT-based deep entropy model for entropy coding. In general, these approaches are designed for static point clouds and do not leverage the temporal information within dynamic point clouds.…”

“…According to the hierarchical transform steps, we can construct RAHT trees following [16]. As shown in Fig.…”

“…Furthermore, it is also possible to extend this temporal dependence to two consecutive RAHT trees. In 3DAC [16], transform coding context and inter-channel correlation are explored via the tree structure for static point cloud attribute compression. In this paper, we further build temporal correlations within dynamic point clouds through the RAHT trees in Secs.…”

“…As shown in Fig. 1, by leveraging the useful temporal information, our method achieves better quantitative and qualitative compression performance than the static point cloud compression counterpart, 3DAC [16]. Our key contributions are as follows:…”