Point Cloud Attribute Prediction with Partitioned Reference Buffers

The patent segments the set of all reference points into two distinct groups: a first reference point set containing points at the same partition depth as the current node, and a second reference point set containing points at different partition depths. This segmentation allows the system to manage reference points in an organized manner, processing them in specific orders (same-depth first, then different-depth) to balance prediction accuracy with computational efficiency.

The patent applies partial action by selectively using reference points from the first reference point set (same partition depth) before resorting to the second reference point set (different partition depths). This partial approach ensures that the most relevant reference points are used first for prediction, reducing the need to process all available reference points and thus lowering overall processing complexity while maintaining adequate prediction accuracy.

The patent segments reference points into two sets based on their partition depth relative to the current node. The first set contains reference points at the same partition depth, while the second set contains reference points at different partition depths. This segmentation enables the system to buffer and process reference points in a structured manner, optimizing memory usage by loading only the necessary reference points for prediction rather than all available points.

The patent performs preliminary action by pre-establishing the partition tree structure and pre-organizing reference points into the two sets before attribute prediction is needed. This preliminary organization allows the decoding device to efficiently access and buffer only the required reference points during prediction, reducing real-time memory requirements while maintaining prediction performance.

The patent segments the attribute prediction process into two stages: first using reference points from the same partition depth (first reference point set), then using reference points from different partition depths (second reference point set). This segmented approach enables efficient compression by processing reference points in a structured manner, achieving good compression ratios while maintaining decoding performance through systematic prediction.

The patent introduces dynamic adaptability in the prediction process by allowing the decoding device to flexibly select which reference point sets to use based on the specific prediction requirements. This dynamic approach enables optimization of the compression-decoding performance trade-off by adapting the reference point selection strategy to different encoding scenarios.