Multiscale sparse representation offers significant advantages in point cloud geometry compression, delivering state-of-the-art performance compared to both standardized solutions and other learned approaches. A crucial component of this framework is the cross-scale occupancy prediction, which employs the lower-scale reference representation either from the current frame alone or from both the current and temporal reference frames to establish conditional priors for either static or dynamic coding. However, existing works mainly use local computations, e.g., sparse convolutions and kNN attention, to exploit correlations in such a representation; these methods usually fail to adequately capture global coherence. In addition, the fixed configuration of lossless-lossy scales cannot adapt to temporal dynamics, which limits the reconstruction quality of temporal references in dynamic coding. These limitations constrain the generation of more effective priors used for conditional coding. To address these issues, we propose two new techniques. The first is KPA (Key Point-driven Attention), which integrates both local and global characteristics. The second is AdaScale (Adaptive Lossy/Lossless Scale), which decides whether the transitional scale should be in lossless or lossy mode based on temporal displacement, thereby enhancing the reconstruction quality of the temporal reference. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art methods, including rules-based standard codecs like G-PCC and V-PCC, as well as learning-based approaches like Unicorn and TMAP, across both static/dynamic and lossy/lossless coding scenarios.

KPA (Key Point-driven Attention): A novel attention mechanism that identifies salient key points to bridge local geometric details and global structural context. By focusing computation on these representative points, KPA enables long-range dependency modeling while maintaining efficiency—overcoming the limited receptive field of conventional sparse convolutions.

AdaScale (Adaptive Lossy/Lossless Scale): An adaptive strategy that dynamically selects lossless or lossy coding for transitional scales based on inter-frame motion. When significant temporal displacement is detected, the scale is encoded losslessly to produce a high-quality reference frame; otherwise, it uses lossy compression for efficiency. This improves both rate-distortion performance and temporal prediction accuracy.

• By integrating KPA and AdaScale into the multiscale sparse representation framework, this work demonstrates significant improvements over rule-based solutions like G-PCC and V-PCC, as well as learned approaches like Unicorn and TMAP1.
  • In static coding, it surpasses TMAP by 9.1% to 22.3% in lossless mode and by 12.7% to 40.7% in lossy mode. For dynamic coding, it outperforms TMAP by 8.8% in lossless mode and 37.6% in lossy mode. Note that TMAP requires significantly longer processing time, likely due to redundant modules such as point-wise enhancement.
  • Compared to V-PCC, our method achieves BD-BR gains of about 56% in static lossy mode and 91% in dynamic lossy mode, while maintaining comparable decoding latency.
• The impressive compression gains of this work are attributed to the use of KPA and AdaScale. KPA combines both local and global characteristics of the input reference representation, while AdaScale improves the quality of the reference representation itself. Together, these mechanisms strive to generate better conditional priors for occupancy prediction.
  • KPA samples a limited number of key points to perform cross-attention, maintaining global structural coherence while significantly reducing complexity.
  • AdaScale loosens the fixed configuration of lossless-lossy scales, opening new opportunities to optimize the reference frame quality for better performance.