RU2778864C1 - Implicit geometric division based on a quad-tree or binary tree for encoding a point cloud - Google Patents

Abstract

FIELD: decoding.

SUBSTANCE: method for decoding the geometry of a point cloud in a point cloud decoder includes receiving a bitstream including a slice of the encoded frame of the point cloud, and reconstructing the octree representing the geometry of points in the bounding box of the slice, wherein the current node of the octree is divided by a quadtree (QT)-based division or a binary tree (BT)-based division. The reconstruction (S1520) of the octree includes determining the value of a variable designated partitionSkip and indicating the type of division and the direction of division of the current node of the octree. When implicit QT or BT division is applied, the geometry of the point cloud can be implicitly divided by a subset of all dimensions if certain criteria are met, instead of always dividing by all dimensions.

EFFECT: ensured division of a quadtree or a binary tree.

20 cl, 39 tbl, 16 dwg

Description

INCLUSION BY LINK

[0001] The present application claims priority under U.S. Patent Application 16/909,142 "Implicit Quadtree or Binary Tree Based Geometric Partitioning for Point Cloud Encoding", filed June 23, 2020, which claimed priority under U.S. Provisional Applications 62/ 867,063 "Implicit Geometric Division for Point Cloud Encoding", filed June 26, 2019, 62/904,384 "On Geometry Coding for Point Clouds", filed September 23, 2019, and 62/910,387 "Additional Information on Adaptive Geometric Quantization and Implicit Geometric Separation for Point Cloud Encoding, filed October 3, 2019. The contents of previous applications are incorporated herein by reference in their entirety.

FIELD OF TECHNOLOGY TO WHICH THE INVENTION RELATES

[0002] The present invention relates to point cloud coding.

BACKGROUND OF THE INVENTION

[0003] The description of the prior art is provided herein to represent the overall context of the invention. The work of the authors of the invention, to the extent that it is described in this section, as well as aspects of the description that cannot be qualified as state of the art at the time of filing, neither expressly nor impliedly recognized as the state of the art for the present invention.

[0004] The point cloud has been widely used in recent years. For example, a point cloud can be used in autonomous vehicles to detect and localize objects. The point cloud is also used in geographic information systems (GIS) for cartography, as well as for visualization and archiving of objects and collections of cultural heritage, etc.

[0005] A point cloud frame contains a set of high-dimensional points, typically three-dimensional (3D) points, each of which includes three-dimensional position information and additional attributes such as color, reflectance, and so on. They can be captured with multiple cameras and depth sensors or lidar in a variety of setups and can include thousands or billions of points to realistically represent original scenes.

[0006] For faster transmission or less memory, compression techniques are needed to reduce the amount of data to represent a point cloud. The ISO/IEC MPEG group (JTC 1/SC 29/WG 11) created a special group (MPEG-PCC) to standardize compression methods for static or dynamic point clouds.

SUMMARY OF THE INVENTION

[0007] Aspects of the invention provide a method for decoding point cloud geometry in a point cloud decoder. The method may include receiving a bitstream including a slice of an encoded point cloud frame and reconstructing an octree representing the geometry of the points in the bounding box of the slice using implicit geometric partitioning, where the current node of the octree is partitioned by a partition based on a quadtree (QT, quadtree). ) or separation based on a binary tree (BT, binary-tree).

[0008] In an embodiment of the invention, the splitting method of the current octree node using one of QT splitting, BT splitting, or octree based splitting (OT, octree) is determined based on a predetermined condition. In an embodiment of the invention, a method for splitting the current octree node using one of QT split, BT split, or OT split is determined based on one or more parameters. One of said one or more parameters may be signaled in the bitstream or with a locally configured pre-configured value.

[0009] In an embodiment of the invention, occupancy information belonging to the 8-bin occupancy code of the current octree node is received from the bitstream. Each busy bit corresponds to a busy child node of the current octree node. The four bins that belong to the 8-bin busy code are not signaled in the bitstream when the current octree node is split by QT splitting, and the 6 bins that belong to the 8-bin busy code are not signaled in the bitstream when the current octree node is split by VT separation. In the present application, we may refer to the 8-bin occupancy code as an 8-bit occupancy code, or each bin of the 8-bin occupancy code a bit.

[0010] In an embodiment of the invention, one or more syntax elements are received from the bitstream indicating the three-dimensional (3D) dimensions of the encoded point cloud slice bounding box. In an embodiment of the invention, a value is determined for a variable, denoted partitionSkip, that indicates the type of partition and the direction of partition of the current node of the octree. In the example, the partitionSkip variable is represented in binary form by three bits corresponding to the x, y, and z directions, respectively, with each bit indicating whether to partition along the corresponding x, y, or z direction. In the example, the depth in the x, y, or z dimension for a child node of the current octree node is updated based on the partitionSkip variable.

[0011] In an embodiment of the invention, reconstructing the octree also includes receiving a syntax element indicating that the current node of the octree has one occupied child node, receiving 1 bin if the partitionSkip variable indicates a BT split, or 2 bins if the partitionSkip variable points to splitting the QT, and determining an occupancy map identifying occupied child nodes of the current octree node based on the received 1 or 2 bins.

[0012] In an embodiment of the invention, in the process of parsing a bitstream to determine an occupancy map syntax element that identifies busy child nodes of the current octree node, one or more bins of the occupancy map syntax element may be skipped based on the partitionSkip variable.

[0013] In an embodiment of the invention, for a child node of the current node of the direct-coded octree, for each of the x, y, and z directions, a log2 size value, denoted dx, dy, and dz, respectively, is determined for the child node based on the partitionSkip variable. The positions of the dots at the child node are encoded with fixed length codes using (dx, dy, dz) bits, respectively.

[0014] In an embodiment of the invention, a syntax element is received from the bitstream indicating one of the following parameters: the maximum number of implicit QT and BT splits performed before OT splits, the minimum size of the implicit QT and BT splits that prevents implicit node QT and BT splits, when all node sizes are less than or equal to the minimum size, or a priority indicating which of the QT or BT implicit splits to perform first when both QT and BT splits are enabled.

[0015] In an embodiment, when the octree depth of the current node is less than parameter K, or when the smallest log2 size value among the log2 sizes of the current node in the x, y, and z directions is equal to the parameter M, the split type and split direction for splitting the current node can be defined according to the conditions specified in the following table:

where parameter K is an integer in the range 0 ≤ K ≤ max(d _x , dy, d _z ) - min(d _x , d _y , d _z ) and specifies the maximum number of implicit QT and BT splits that are allowed before OT splits, parameter M is an integer in the range 0 ≤ M ≤ min(d _x , d _y , d _z ), specifies the minimum size of implicit QT and BT splits, and prevents implicit QT and BT splits of a node when all node sizes are less than or equal to M, a dx, dy, and dz are the log2 values of the dimensions of the current node in the x, y, and z directions, respectively. The values dx, dy and dz correspond to d _x , d _y and d _z in the above table. These two designations are used interchangeably in this specification.

[0016] In an embodiment of the invention, when the octree depth of the current node is less than the parameter K, or when the smallest log2 size value among the log2 sizes of the current node in the x, y, and z directions is equal to the parameter M, the variable partitionSkip can be defined as follows:

The variable partitionSkip is represented in binary form by three bits and specifies the type of partition and the direction of partition of the current octree node. The parameter K is an integer in the range 0 ≤ K ≤ max(d _x , d _y , d _z ) - min(d _x , dy, d _z ) and specifies the maximum number of implicit QT and BT splits that are allowed before OT splits. The parameter M is an integer in the range 0 ≤ M ≤ min(d _x , d _y , d _z ), specifies the minimum size of implicit QT and BT splits, and prevents implicit QT and BT node splits when all node sizes are less than or equal to M. Values dx, dy, and dz are the log2 values of the current node's dimensions in the x, y, and z directions, respectively. MaxNodeDimLog2 represents the maximum log2 size value from dx, dy, and dz. The |= operator is a bitwise OR compound operation.

[0017] In an embodiment of the invention, a flag may be obtained from the bitstream indicating whether implicit geometric separation is allowed for a sequence of point cloud frames or a slice of an encoded point cloud frame. In an embodiment of the invention, the planar mode is determined not to be applicable in the x, y, or z direction, and no split is performed for the current node. In an embodiment of the invention, a geometry octree occupancy analysis process is performed in which, for a bean having index binIdx in the occupancy code, the variable binIsInferred is set as follows: (a) if one of the following conditions is met, the variable binlslnferred is set to 1: (1) the variable NeighbourPattem is 0 and the number of previously decoded single-valued bins is less than or equal to (binIdx + minOccupied - maxOccupied), or (2) NeighbourPattem is not 0, binldx is maxOccupied-1, and all previous decoded bins are zero, where minOccupied = 2 , and maxOccupied = 8 if OT split is applied, maxOccupied = 4 if QT split is applied, and maxOccupied = 2 if BT split is applied, and (b) otherwise, if none of the above conditions are met, binlslnferred is set equal to 0.

[0018] In an embodiment, the K parameter and the M parameter are used. The K parameter indicates the maximum number of implicit QT and BT splits before OT splits, and the M parameter specifies the minimum size of implicit QT and BT splits that prevents implicit QT and BT splits of a node when all node sizes are less than or equal to the minimum size. The parameters K and M can be updated as follows: (a) if K is greater than the difference of the maximum log2 root size and the minimum log2 slice root size, K is changed to the difference of the maximum log2 root size and the minimum log2 root size. slice node; (b) if M is greater than the minimum value log2 of the slice root node size, M is changed to the minimum log2 value of the slice root node size; (c) if the maximum value log2 of the root node size and the minimum value log2 of the slice root node size are equal, M is changed to 0; and (d) if trisoup mode is enabled, K is changed to the difference between the maximum log2 root node size and the minimum log2 slice root node size, and M is changed to the minimum log2 slice root size. It should be noted that the log2 value of the node size in triangular grid mode should not be greater than the minimum log2 value of the root node size of the slice.

[0019] Aspects of the invention provide an apparatus for decoding point cloud geometry. The device may comprise a circuit configured to receive a bitstream including a slice of an encoded point cloud frame and reconstruct an octree representing the geometry of the points in the slice's bounding box using implicit geometric partitioning, where the current node of the octree is partitioned by quadtree partitioning (QT) or separation based on a binary tree (BT).

[0020] Aspects of the invention provide a computer-readable medium storing instructions that, when executed by a processor, causes the processor to perform a point cloud geometry decoding method.

BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Various embodiments of the invention, which are offered as examples, will be described in detail with reference to the following drawings, in which like numbers designate like elements.

[0022] FIG. 1 shows a recursive partitioning process in accordance with an embodiment of the invention.

[0023] FIG. 2 shows a process for generating a level of detail (LOD, level of detail) in accordance with an embodiment of the invention.

[0024] FIG. 3 shows an example of an encoder according to an embodiment of the invention.

[0025] FIG. 4 shows an example of a decoder according to an embodiment of the invention.

[0026] FIG. 5 shows octree-based partitioning in three-dimensional (3D) space.

[0027] FIG. 6 shows an example of a bilevel octree partition (OT) and corresponding busy codes.

[0028] FIG. 7 shows a point cloud sequence having principal components in the x and y directions.

[0029] FIG. 8 shows examples of quadtree (QT) partitioning in 3D space according to an embodiment of the invention.

[0030] FIG. 9 shows examples of binary tree (BT) partitions in 3D space according to an embodiment of the invention.

[0031] FIG. 10-13 show two-dimensional (2D) blocks to demonstrate four different separation processes based on implicit QT and BT separations.

[0032] FIG. 14 shows 2D blocks for demonstrating the Testing Model 13 (TMC13) circuit, in which OT partitions are performed based on an extended cubic bounding box.

[0033] FIG. 15 shows a flowchart of a geometry decoding method based on QT/BT separation for point cloud decoding in accordance with an embodiment of the present invention.

[0034] FIG. 16 shows a computer system for implementing geometry decoding methods in various embodiments of the invention.

DETAILED DESCRIPTION OF EMBODIMENTS

[0035] The present invention relates to geometry based point cloud compression (G-PCC). Methods for implicit partitioning based on a quadtree (QT) or a binary tree (BT) are described. When implicit QT or BT partitioning is applied, the point cloud geometry can be implicitly partitioned across a subset of all dimensions when certain criteria are met, instead of splitting across all dimensions all the time.

[0036] I. An encoder for encoding a point cloud L and a decoder

[0037] 1. Point cloud data

[0038] Point cloud (or point cloud) data is used to represent a three-dimensional (3D) scene or object in some emerging applications such as immersive virtual reality (VR) / augmented reality (AR) / mixed reality (MR), automotive / robotic navigation, medical imaging, etc. A point cloud may include a collection of individual 3D points. Each point is associated with a set of 3D coordinates indicating the 3D position of the corresponding point and a number of other attributes such as color, surface normal, opacity, reflectivity, and so on. In various embodiments of the invention, point cloud input data may be quantized and subsequently organized into a three-dimensional grid of cubic voxels that may be described using an octree data structure. The resulting voxel octree facilitates traversal, search, and access to quantized point cloud data.

[0039] A point cloud frame may include a set of 3D points at a particular point in time. Point cloud frames can be used to reconstruct an object or scene as a composition of such points. They can be captured with multiple cameras and depth sensors in a variety of settings and can be made up of thousands or even billions of points to realistically represent rendered scenes. A sequence of point cloud frames may be referred to as a point cloud. The set of Cartesian coordinates associated with the 3D points of a point cloud frame may be referred to as the geometry of those 3D points.

[0040] Compression techniques are needed to reduce the amount of data needed to represent a point cloud (eg, a sequence of point cloud frames). Thus, technologies are needed to compress lossy point clouds for use in real-time communications and virtual reality with six degrees of freedom (6 DoF, Degrees of Freedom). In addition, technologies are required for lossless compression of point clouds in the context of dynamic display for autonomous driving and cultural heritage applications, etc. In addition, standards are needed to address geometry and attribute compression (eg, colors and reflectivity), scalable/progressive coding, time-captured point cloud sequence coding, and random access to point cloud subsets.

[0041] 2. Coordinate quantization

[0042] In an embodiment of the invention, the coordinates of the points in the cloud input may be quantized first. For example, real numeric coordinate values can be quantized into integer values. After quantization, several points may have the same position in some voxels. These repeating points can optionally be merged into a single point.

[0043] 3. Octree Geometry Coding

[0044] FIG. 1 shows a recursive splitting process (100) in accordance with an embodiment of the invention. Process (100) may be performed to create an octree structure to represent the positions of a set of points in a point cloud. As shown, a cubic axis-aligned bounding box (101) containing a set of points is first defined. Then, the bounding box (101) is recursively split to construct an octree structure. As shown, at each stage, the current cube can be divided into 8 sub-cubes. An 8-bit code, called an occupancy code, may be generated to indicate whether each of the 8 sub-cubes contains points. For example, each subcube is associated with a one-bit value. If the subcube is busy, it has a bit value of 1; otherwise, the corresponding subcube has a bit value of 0. Occupied subcubes may be split until a predetermined minimum subcube size is reached. The minimum size subcube is a voxel corresponding to the octree structure. Thus, a sequence of busy codes can be generated, which can subsequently be compressed and transmitted from an encoder to a decoder. By decoding the busy codes (eg, performing an octree decoding process), the decoder can obtain the same octree structure as the encoder or an estimate of the octree structure.

[0045] 4. Passing Attributes

[0046] As a result of the encoder-side octree generation or encoding process, the minimum size subcube may contain more than one point. Thus, a position corresponding to a voxel (eg, the center of the corresponding subcube) may correspond to a plurality of attribute sets of a plurality of points. In such a scenario, in an embodiment of the invention, an attribute transfer process may be performed to determine one attribute set based on a plurality of attribute sets for a corresponding voxel. For example, an averaged attribute of a subset of nearest neighbors can be used as an attribute of the corresponding voxel. In various embodiments of the invention, various methods may be used to pass attributes.

[0047] 5. Level of Detail (LOD) Generation

[0048] FIG. 2 shows a process (200) for generating a level of detail (LOD) in accordance with an embodiment of the invention. Process (200) may be performed on quantized positions (eg, voxel positions) ordered according to the octree decoding process. As a result of process (200), the points may be reorganized or reordered into a set of detail levels. Process (200) may be performed identically in the encoder and decoder. The subsequent attribute encoding process may be performed at the encoder or decoder in accordance with the order given by process (200) (referred to as LOD-based order).

[0049] In particular, in FIG. 2 shows three LOD levels: LOD0, LOD1 and LOD2. The Euclidean distance, d0, d1, or d2, can be specified for LOD0, LOD1, and LOD2, respectively. Each LOD level includes a subset of points P0-P9. The distance between each pair of points at the corresponding LOD level is greater than or equal to the corresponding Euclidean distance. Euclidean distances can be organized as d0>d1>d2. When organized this way, a higher level of detail includes fewer points that are farther apart and provides a coarser representation of the point cloud, while a lower level of detail includes more points that are closer together and provides a more accurate representation. point clouds.

[0050] As a result of the above LOD generation process (200), points in the original order (octree decoding order) from P0 to P9 can be reorganized in the LOD-based order: P0, P5, P4, P2, P1, P6, P3, P9, P8 and P7.

[0051] 6. Attribute Prediction

[0052] The attributes associated with the point cloud may be encoded and decoded in the order specified by the LOD generation process. For example, point by point, in LOD based order, the attribute prediction of each current point (the point currently being processed) can be determined by performing the attribute prediction method in the encoder and/or decoder. A similar attribute prediction process may be performed at the encoder and decoder.

[0053] With the attribute prediction obtained, the encoder can generate a residual signal by subtracting the attribute prediction value from the corresponding original current point attribute value. The residual signal can then also be compressed alone or in combination with other residual signals. For example, transformation and/or quantization operations may be performed followed by entropy encoding of the resulting signals. The compressed residual signal may be transmitted to the encoder in a bitstream.

[0054] At the decoder, the residual signal may be recovered by performing the reverse process of the encoding process at the encoder to encode the residual signal. By predicting the received attribute and reconstructing the residual signal corresponding to the current point, a reconstructed attribute of the current point can be obtained. Similarly, this reconstruction operation may occur in the encoder to obtain a reconstructed attribute.

[0055] Various attribute prediction methods may be used to determine attribute prediction in various embodiments of the invention. Typically, current point attribute prediction is performed using previously reconstructed attributes of points adjacent to the current point. When the attribute prediction process is started based on the LOD-based order, the reconstructed point attribute values up to the current point are already available. In addition, positions (3D coordinates) of points in the point cloud are also available from the octree encoding or decoding process. Accordingly, the attribute prediction process can be performed considering the reconstructed attributes and 3D coordinates of neighboring points of the current point.

[0056] For example, the set of neighboring points of the current point may be first determined using various algorithms. In one example, a k-d tree structure search process may be performed to determine the set of points closest to the current point.

[0057] For example, a geometric distance and/or attribute distance approach is used to determine attribute prediction. A prediction attribute may be determined based on a weighted sum (or weighted average) of the reconstructed attributes of a set of determined neighboring points at an encoder or decoder. For example, using a set of certain neighboring points, a weighted sum (or weighted average) of the reconstructed attributes of certain neighboring points may be determined as a prediction attribute at an encoder or decoder. For example, the weight used in the weighted sum method (also referred to as the interpolation based method) may be inverse (or inversely proportional) to geometric distance or attribute distance. Alternatively, the weight may be a two-sided weight derived from a combination of a geometric distance weight (geometric 8) and an attribute distance weight (attribute weight).

[0058] For example, a rate-distortion (RD) approach is used to determine attribute prediction. For example, a candidate index may be associated with each reconstructed neighbor set attribute value at an encoder or decoder. An RD optimization based method may be performed at the encoder to evaluate which of the possible reconstructed neighbor point attribute values is the best choice to use as an attribute prediction. For example, distortion can be measured by the difference between the original (or true) value of the current point attribute and the candidate prediction (reconstructed candidate attribute value). The rate may be the cost of encoding an index of the selected candidate prediction. To determine the best candidate for the prediction signal, a Lagrangian RD cost function can be determined. Thus, the candidate index of the selected candidate prediction may be signaled to the decoder.

[0059] Accordingly, the decoder may first determine the same set of neighboring points for the corresponding current point and associate indices with reconstructed attribute values of the same set of neighboring points in the same manner as on the encoder side. The decoder may then determine an attribute prediction based on the reconstructed attribute values of neighboring points using the signaled candidate index.

[0060] 7. Examples of coding systems for point cloud compression

[0061] FIG. 3 shows an example of an encoder (300) in accordance with an embodiment of the invention. The encoder may be configured to receive point cloud data and compress the point cloud data to create a bit stream carrying the compressed point cloud data. In an embodiment of the invention, the encoder (300) may include a position quantization module (310), a duplicate point removal module (312), an octree encoding module (330), an attribute transmission module (320), a LOD generation module (340), a module (350) attribute prediction, a residual quantization module (360), an arithmetic coding module (370), a residual dequantization module (380), an addition module (381), and a memory (390) for storing the reconstructed attribute values.

[0062] As shown, the encoder (300) may receive an input point cloud (301). The positions (three-dimensional coordinates) of the point cloud (301) are provided to the quantization module (310). The quantization module (310) is configured to quantize coordinates to generate quantized positions. The additional module (312) for removing duplicate points is configured to receive the quantized positions and perform a filtering process to identify and remove duplicate points. The octree encoding module (330) is configured to receive filtered positions from the duplicate dot removal module and perform an octree-based encoding process to generate a sequence of busy codes that describe a three-dimensional grid of voxels. Busy codes are transmitted to the module (370) arithmetic coding.

[0063] The attribute transfer module (320) is configured to receive attributes of the input point cloud and perform an attribute transfer process to determine an attribute value for each voxel when multiple attribute values are associated with a corresponding voxel. The attribute transfer process may be performed on the reordered points output from the octree encoding module (330). The attributes after the transfer operations are provided to the attribute prediction module (350). The LOD generation module (340) is configured to work with the reordered points output from the octree encoding module (330) and reorganize the points into different LOD levels. The LOD information is provided to the attribute prediction module (350).

[0064] The attribute prediction module (350) processes the points in accordance with the LOD-based order indicated by the LOD information from the LOD generation module (340). The attribute prediction module (350) generates an attribute prediction for the current point based on the reconstructed attributes of the set of neighboring points for the current point stored in the memory (390). The prediction residuals can subsequently be obtained based on the original attribute values received from the attribute transmission module (320) and the locally generated attribute predictions. When the candidate indexes are used in the corresponding attribute prediction process, the index corresponding to the selected prediction candidate may be provided to the arithmetic encoding module (370).

[0065] The residual quantization module (360) is configured to receive prediction residuals from the attribute prediction module (350) and perform quantization to generate quantized residuals. The quantized residuals are fed to the arithmetic coding module (370).

[0066] The residual dequantization module (380) is configured to receive quantized residuals from the residual quantization module (360) and generate reconstructed prediction residuals by performing the reverse operations of the quantization operations performed in the residual quantization module (360). The adder (381) is configured to receive the reconstructed prediction residuals from the residual dequantizer (380) and corresponding attribute predictions from the attribute predictor (350). By combining the reconstructed prediction residuals and the attribute predictions, the reconstructed attribute values are generated and stored in memory (390).

[0067] The arithmetic encoding module (370) is configured to receive busy codes, candidate indices (if used), quantized residuals (if generated), and other information, and perform entropy encoding to further compress the received values or information. As a result, a compressed bit stream (302) carrying compressed information can be generated. The bitstream (302) may be transmitted to or otherwise provided to a decoder that decodes the compressed bitstream, or may be stored in a storage device. A bitstream may refer to a sequence of bits that forms a representation of encoded point cloud frames. An encoded point cloud frame may refer to an encoded representation of a point cloud frame.

[0068] FIG. 4 shows an example of a decoder (400) in accordance with an embodiment of the invention. The decoder (400) may be configured to receive the compressed bitstream and perform decompression of the point cloud data to decompress the bitstream to generate decoded point cloud data. In an embodiment, the decoder (400) may comprise an arithmetic decoding module (410), a residual dequantization module (420), an octree decoding module (430), a LOD generation module (440), an attribute prediction module (450), and a memory (460) for storing reconstructed attribute values.

[0069] As shown, the compressed bit stream (401) can be received in the module (410) arithmetic decoding. The arithmetic decoding module (410) is configured to decode the compressed bitstream (401) to obtain quantized residuals (if any) and point cloud occupancy codes. The octree decoding module (430) is configured to determine the reconstructed point positions in the point cloud according to the occupancy codes. The LOD generation module (440) is configured to reorganize the points into different LOD levels based on the reconstructed positions and determine the order based on the LOD. The residual dequantization module (420) is configured to generate reconstructed residuals based on the quantized residuals received from the arithmetic decoding module (410).

[0070] The attribute prediction module (450) is configured to perform an attribute prediction process to determine attribute predictions for points according to LOD-based order. For example, a current point attribute prediction may be determined based on reconstructed neighboring point attribute values for the current point stored in memory (460). The attribute prediction module (450) may combine the attribute prediction with the corresponding reconstructed residual to generate a reconstructed attribute for the current point.

[0071] The sequence of reconstructed attributes generated by the attribute prediction module (450), together with the reconstructed positions generated by the octree decoding module (430), corresponds to the decoded point cloud (402) that is output from the decoder (400) in one example. In addition, the reconstructed attributes are also stored in memory (460) and can later be used to derive attribute predictions for subsequent points.

[0072] In various embodiments of the invention, the encoder (300) and decoder (400) may be implemented using hardware, software, or a combination thereof. For example, the encoder (300) and decoder (400) may be implemented with a processing circuit, such as one or more integrated circuits (ICs), that operate with or without software, such as an application specific integrated circuit (ASIC). ), field programmable gate array (FPGA), etc. In another example, the encoder (300) and decoder (400) may be implemented as software or firmware, including instructions stored on a persistent (or non-volatile) computer-readable storage medium. The instructions, when executed by a processing circuit, such as one or more processors, cause the processing circuit to perform the functions of an encoder (300) and a decoder (400).

[0073] It should be noted that elements in the encoder (300) and decoder (400) configured to implement the various methods disclosed herein may be included in other decoders or encoders, which may have a structure similar to those shown in FIG. . 3 and FIG. 4, or a structure different from them. In addition, the encoder (300) and decoder (400) may be included in the same device or separate devices in different examples.

[0074] II. G-PCC in test model 13 (TMC13)

[0075] The G-PCC considers both Category 1 (static point clouds) and Category 3 (dynamically derived point clouds) point cloud compression. A test model called TMC13 has been developed by the Moving Picture Experts Group (MPEG) as a basis for exploring potential point cloud coding technologies. The TMC13 model separately compresses the geometry and related attributes such as color or reflectivity. The geometry, which is the 3D point cloud coordinates, is encoded by octree-based partitioning. The attributes are then compressed based on the reconstructed geometry using predictive and lifting techniques.

[0076] 1. Octree-based partitioning and coding of busy information in the current version of TMC13

[0077] Dividing a 3D cube uniformly will result in eight sub-cubes and is known as octree (OT) partitioning in Point Cloud Compression (PCC). OT partitioning resembles binary tree (BT) partitioning in 1D space and quadtree partitioning (QT) in 2D space. The OT separation is illustrated in FIG. 5, where the three-dimensional cube (501) shown as a solid line is divided into eight smaller cubes of the same size shown as a dashed line. The left 4 dice are associated with indices 0-3 and the right 4 dice are associated with indices 4-7.

[0078] In TMC13, if an octree geometry codec is used, geometry encoding occurs as follows. First, the axis-aligned cubic bounding box B is defined by two points (0,0,0) and (2d, ^2d , ^2d ), where ^2d specifies the size of B, and ^d is encoded in the bitstream. It is assumed that all points to be compressed are inside the given bounding box B.

[0079] The octree structure is then built by recursively partitioning B. At each step, the cube is divided into 8 sub-cubes. The size of the subcube after iterative splitting k (k≤d) times will be (2 ^dk , 2 ^dk , 2 ^dk ). An 8-bit code, namely the busy code, is then generated by associating a one-bit value with each subcube to indicate whether it contains dots (i.e., full and has a value of 1) or not (i.e., empty and has a value of 0). Only complete subcubes larger than 1 (i.e., not voxels) are further split. The busy code for each cube can then be compressed using an arithmetic encoder.

[0080] The decoding process starts by reading the dimensions of the bounding box B from the bitstream. The same octree structure is then built by dividing B according to the decoded busy codes. An example of a two-level separation of OTs and associated busy codes is shown in FIG. 6, where the dark cubes and nodes indicate that they are occupied by dots.

[0081] As shown, the cube (601) is divided into 8 subcubes. Using the same indexing method as in FIG. 5, each of the 0th and 7th subcubes is further divided into 8 subcubes. The octree (610) corresponding to the partitions of the cube (601) includes a root node (611) at the first level. The root node (611) is divided into 8 child nodes, which can be indexed from 0 to 7. Nodes 0 and 7 (612-613) at the second level are further divided into 16 child nodes. The level of a node in the octree (610) may correspond to the number of hops from the root to the corresponding node, and may be referred to as the depth of the octree (610). Depths 0-2, corresponding to the first through third levels of the octree (610), are indicated in FIG. 6.

[0082] 2. Busy code coding

[0083] The busy code of each node may be compressed using an arithmetic encoder. The busy code may be denoted as S, which is an 8-bin integer, with each bin in S indicating the busy state of each child node. In TMC13, there are two ways to encode the busy code, namely, bit-by-bit encoding and byte-by-byte encoding, with bit encoding enabled by default. In either case, context simulation arithmetic encoding is performed to encode the busy code, with the context state initialized at the beginning of the entire encoding process and updated during the encoding process.

[0084] For bitwise encoding, the eight bins in S are encoded in a particular order, where each bin is encoded by referring to the busy state of neighboring nodes and children of neighboring nodes, with neighboring nodes being at the same level of the current node.

[0085] For byte coding, S is encoded by referring to an adaptive lookup table (A-LUT) that keeps track of the N (e.g., 32) most frequent busy codes, and a cache that keeps track of the last different observed M (for example, 16) busy codes.

[0086] A binary flag indicating whether S is in the A-LUT or not is encoded. If S is in the A-LUT, the index in the A-LUT is encoded with a binary arithmetic encoder. If S is not in the A-LUT, then a binary flag is encoded indicating whether S is in the cache or not. If S is in the cache, then the binary representation of its cache index is encoded using a binary arithmetic encoder. Otherwise, if S is not in the cache, the binary representation of S is encoded using a binary arithmetic encoder.

[0087] The decoding process begins by parsing the dimensions of bounding box B from the bitstream. The same octree structure is then built by dividing B according to the decoded busy codes.

[0088] III Implicit Geometric Separation for Point Cloud Encoding

[0089] 1. Problem

[0090] In the TMC13 design, the bounding box can only be cube B, which is the same size for all dimensions, and splits OT for full subcubes at each node, with subcubes halved in all dimensions. FROM splitting is performed recursively until the size of the subcubes reaches one or until there are no points left in the subcubes. However, this method is not effective in all cases, especially when the points are unevenly distributed in the 3D scene. One extreme case would be a 2D plane in 3D space where all points are in the xy plane and the z-axis deviation is zero. In this case, the OT split performed on cube B as a starting point would waste a lot of bits to represent occupancy information in the z-direction, which is redundant and useless. In real applications, the worst case may not occur often, however, the point cloud usually has an asymmetric bounding box that is shorter in some dimensions.

[0091] As shown in FIG. 7, the point cloud sequence named ford_01_voxlmm used for testing in MPEG-PCC has principal components in the x and y directions. In fact, many point cloud data generated by a lidar sensor have the same characteristics.

[0092] 2. Implicit separation of QT and BT

[0093] Aspects of the invention provide embodiments of the invention regarding partitioning of a cuboid bounding box, where a cube or node at the time of partitioning can be implicitly defined as being partitioned by a QT or BT partition, instead of the always applied OT partition. The busy bits indicating busy information may be stored based on the implicit separations of QT and BT.

[0094] For a bounding box, which may not be a perfect cube, in some cases nodes at different levels may not (or may not) be separated in all directions. If the separation is performed in all three directions, then such a separation is a typical OT separation. If the split is in two of the three directions, then it is a QT split in 3D space. If the separation is performed in only one direction, then the separation is BT separation in 3D space. Examples of QT and BT separations in 3D space are shown in FIG. 8 and FIG. 9, respectively. For purposes of illustration, the drawings show partitions QT and BT of an ideal cube, but it should be noted that the partition can be based on any conventional cuboid forming a bounding box.

[0095] FIG. 8 shows three cubes (801-803) divided along the axes (or directions) x-y, x-z and y-z, respectively. The subnodes in each cube (801-803) are assigned indexes that are a subset of the 8 indexes for indexing the 8 child nodes in the OT split in the example of FIG. 5. With indexes assigned, QT partitions with three different directions can be represented using busy codes in an octree structure. For example, an occupancy code representing the division of QT along the x-y axes to a cube (801) may take the form x0x0x0x0, where the bit at position x may be used to indicate the busy state (eg, may be 1 or 0). Similarly, an occupancy code representing a QT split along the x-z axes to a cube (802) may take the form xx00xx00, while an occupancy code representing a QT split along the y-z axes to a cube (803) may take the form xxxx0000. As shown, these indices assigned to the subcubes also indicate the positions of the bits corresponding to the resulting subnodes in the 8-bit busy code.

[0096] FIG. 9 shows three cubes (901-903) divided along the x, y, and z axes, respectively. As in FIG. 8, subnodes in each cube (901-903) are assigned indices corresponding to positions in the busy code.

[0097] When predetermined conditions are met, separation of QT and BT may be implicitly performed. "Implicit" means that no additional signaling bits are required to indicate that a QT or BT split is used instead of a OT split. The decoder may determine the type (eg, split QT or BT) and split direction in the same manner as the encoder based on predetermined conditions. Moreover, bits can be saved based on implicit QT or BT split versus OT split in signaling the busy information of each subnode. QT requires four bits instead of eight to represent the busy state of four subnodes, while BT requires only two bits. For example, as shown in FIG. 8-9, bits in the busy code corresponding to indices assigned to subnodes may be signaled, while other bits in the busy code not related to subnodes may be omitted (not signaled). Thus, the busy code may include missing bits and signaled bits when QT and BT separations are introduced.

[0098] It should be noted that QT and BT separations can be implemented in the same OT separation structure. Context selection based on neighboring coded cubes and an entropy encoder can be applied in a similar manner. The context modeling of the busy code based on QT or BT can be modified to suit the asymmetric shape of the subnodes.

[0099] The encoding of the QT and BT implicit separation busy codes may be as in the following examples. First, it can be assumed that the OT busy code is encoded in index order, as shown in FIG. 5. Then, as shown in the leftmost graph in FIG. 8 (cube (801)), the QT split occupancy code along the x-y axes can be encoded by omitting the bits at positions 1, 3, 5, and 7, since they can be inferred at the decoder to be 0, with only the bits signaled. at positions 0, 2, 4, and 6. Similarly, for BT along the x-axis, as shown in the leftmost graph of FIG. 9 (cube (901)), the occupancy information at positions 0 and 4 may be signaled, and the remaining six bits may be inferred to be 0.

должен кодироваться в режиме DM, относительные положения каждой точки в подузлах кодируются посредством кодирования кодами фиксированной длины с использованием (d_x, d_y, d_z) битов, соответственно.[00100] In addition, the current version of TMC13 has a special mode for encoding geometry, which is a direct mode (DM, direct mode) and allows you to encode positions (x, y, z) in a sub-assembly directly without additional separations. For example, the positions are relative positions relative to the origin of the current subnode, encoded with fixed length codes, where the length in bits is determined by the size of the current subnode. Since implicit partitioning can result in sub-nodes with unequal sizes in the (x, y, z) dimensions, the DM mode can be changed accordingly. For example, if the subnode is

is to be encoded in DM mode, the relative positions of each point in the sub-nodes are encoded by fixed-length coding using (d _x , d _y , d _z ) bits, respectively.

[00101] 3. Cuboid Bounding Box Signaling

. Значения d_x, d_y и d_z упоминаются как значения log2 размеров ограничивающего бокса. Заметим, что d_x, d_y, d_z не считаются равными и могут сигнализироваться отдельно в заголовке последовательности (например, в наборе параметров последовательности (SPS)) или заголовке слайса битового потока.[00102] First, bounding box B is not limited to being the same size in all directions, instead the bounding box may be a rectangular box of arbitrary size to better fit the shape of the 3D scene or objects. In various embodiments of the invention, the size of the bounding box B can be represented as a power of two, i.e.

. The values of d _x , d _y and d _z are referred to as the log2 values of the bounding box sizes. Note that d _x , d _y , d _z are not considered equal and may be signaled separately in a sequence header (eg, sequence parameter set (SPS)) or bitstream slice header.

[00103] In addition, it is worth noting that the size of bounding box B can be any positive number without being limited to being a power of two. In FIG. 7 shows an example of a cuboid bounding box enclosing the scene where the z-direction is the smallest.

[00104] Section III of the detailed description of embodiments of the invention shows that some embodiments of the invention are modifications of the specifications of the design standard for geometry-based point cloud compression, ISG / IEC 23090-9: 2019 (E), stage WD, ISG / IEC JTC 1 / SC 29 / WG 1 1 W18179 March 2019

[00105] Embodiment A of the invention

[00106] In one embodiment, the dimensions of a bounding box in three dimensions may be signaled in the geometric slice header as log2, as shown in Table 1. The geometric slice header may include syntax elements applied to the slice. Typically, a slice may refer to a sequence of syntax elements representing a portion of an encoded frame or an entire encoded frame of a point cloud. Slice points may be contained in a bounding box corresponding to the slice.

[00107] The geometry slice header syntax in Table 1 has been changed by adding the following syntax elements in lines 10-12:

[00108] gsh_log2_max_nodesize_x specifies the size of the bounding box in the x dimension, i.e. MaxNodesizeX, which is used in the decoding process as follows:

MaxNodeSizeX=2 ^gsh - ^log2 - ^max - ^nodesize - ^x .

MaxNodeSizeLog2X=gsh_log2_max_nodesize_x.

[00109] gsh_log2_max_nodesize_y specifies the size of the bounding box in the y dimension, i.e. MaxNodesizeY, which is used in the decoding process as follows:

MaxNodeSizeY=2 ^gsh - ^log2 - ^max - ^nodesize_y .

MaxNodeSizeLog2Y=gsh_log2_max_nodesize_y.

[00110] gsh_log2_max_nodesize_z specifies the size of the bounding box in the z dimension, i.e. MaxNodesizeZ, which is used in the decoding process as follows:

MaxNodeSizeZ=2 ^{gsh_log2_max_nodesize_z} .

MaxNodeSizeLog2Z=gsh_log2_max_nodesize_z.

[00111] Embodiment B of the invention

[00112] In another embodiment of the invention, dimensions in three dimensions may be signaled in the geometric slice header as log2. Instead of signaling the three values independently, one can signal their difference as follows.

[00113] The geometry slice header syntax in Table 2 has been changed by adding the following syntax elements on lines 10-12:

[00114] gsh_log2_max_nodesize_x specifies the size of the bounding box in the x dimension, i.e. MaxNodesizeX, which is used in the decoding process as follows:

MaxNodeSizeX=2 ^gsh - ^log2 - ^max - ^nodesize - ^x .

MaxNodeSizeLog2X=gsh_log2_max_nodesize_x.

[00115] gsh_log2_max_nodesize_y_minus_x specifies the size of the bounding box in the y dimension, i.e. MaxNodesizeY, which is used in the decoding process as follows:

MaxNodeSizeY=2 ^{gsh_log2_max_nodesize_y_minus_x+gsh_log2_max_nodesize_x} .

MaxNodeSizeLog2Y=gsh_log2_max_nodesize_y_minus_x+gsh_log2_max_nodesize_x.

[00116] gsh_iog2_max_nodesize_z_minus_y specifies the size of the bounding box in the z dimension, i.e. MaxNodesizeZ, which is used in the decoding process as follows:

MaxNodeSizeZ=2 ^{gsh_log2_max_nodesize_z_minus_y+MaxNodeSizeLog2Y} .

MaxNodeSizeLog2Z=gsh_log2_max_nodesize_z_minus_y+MaxNodeSizeLog2Y.

[00117] Embodiment C

[00118] In another embodiment of the invention, dimensions in three dimensions may be signaled in the geometric slice header using their Cartesian positions as follows.

[00119] The geometry slice header syntax in Table 3 has been changed by adding the following syntax elements on lines 10-12:

[00120] gsh_max_nodesize_x specifies the size of the bounding box in the x dimension, i.e. MaxNodeSizeX, which is used in the decoding process as follows:

MaxNodeSizeX=gsh_max_nodesize_x,

MaxNodeSizeLog2X=ilog2(MaxNodeSizeX),

where ilog2(v) evaluates to the smallest integer greater than or equal to log2(v).

[00121] gsh_max_nodesize_y specifies the size of the bounding box in the y dimension, i.e. MaxNodeSizeY, which is used in the decoding process as follows:

MaxNodeSizeY=gsh_max_nodesize_y.

MaxNodeSizeLog2Y=ilog2(MaxNodeSizeY).

[00122] gsh_max_nodesize_z specifies the size of the bounding box in the z dimension, i.e. MaxNodeSizeZ, which is used in the decoding process as follows:

MaxNodeSizeZ=gsh_max_nodesize_z.

MaxNodeSizeLog2Z=ilog2(MaxNodeSizeZ).

[00123] Embodiment D

[00124] In another embodiment of the invention, dimensions in three dimensions may be signaled in the geometric slice header with their Cartesian positions minus one, as follows.

[00125] The geometry slice header syntax in Table 4 has been changed by adding the following syntax elements on lines 10-12:

[00126] gsh_max_nodesize_x_minus_one specifies the size of the bounding box in the x dimension, i.e. MaxNodeSizeX, which is used in the decoding process as follows:

MaxNodeSizeX=gsh_max_nodesize_x_minus_one + 1.

MaxNodeSizeLog2X=ilog2(MaxNodeSizeX).

[00127] gsh_max_nodesize_y_mimis_one specifies the size of the bounding box in the y dimension, i.e. MaxNodeSizeY, which is used in the decoding process as follows:

MaxNodeSizeY=gsh_max_nodesize_y_minus_one + 1.

MaxNodeSizeLog2Y=ilog2(MaxNodeSizeY).

[00128] gsh_max_nodesize_z_minus_one specifies the size of the bounding box in the z dimension, i.e. MaxNodeSizeZ, which is used in the decoding process as follows:

MaxNodeSizeZ=gsh_max_nodesize_z_minus_one + 1.

MaxNodeSizeLog2Z=ilog2(MaxNodeSizeZ).

[00129] Embodiment E

[00130] In another embodiment of the invention, only one of the three dimensions allows for different dimensions. In this case, the dimensions in the x and y dimensions are the same, but the dimension in the z dimension can be different, so the geometry slice header log2 specifies two values, as shown below.

[00131] The geometry slice header syntax in Table 5 has been changed by adding the following syntax elements in lines 10-11:

[00132] gsh_log2_max_nodesize_x_y specifies the size of the bounding box in x and y dimensions, i.e. MaxNodesizeX and MaxNodesizeY, which are used in the decoding process as follows:

MaxNodeSizeX=MaxNodeSizeY=2 ^gsh - ^log2 - ^max - ^nodesize - ^x_y .

MaxNodeSizeLog2X=MaxNodeSizeLog2Y=gsh_log2_max_nodesize_x_y.

[00133] gsh_log2_max_nodesize_z specifies the size of the bounding box in the z dimension, i.e. MaxNodesizeZ, which is used in the decoding process as follows:

MaxNodeSizeZ=2 ^gsh - ^log2 - ^max - ^nodesize - ^z .

MaxNodeSizeLog2Z=gsh_log2_max_nodesize_z.

[00134] Embodiment F

[00135] In another embodiment of the invention, only one dimension out of three allows for different dimensions. In this case, the dimensions in the x and y dimensions are the same, but the size in the z dimension can be different, so the geometry slice header contains two values in the form log2. Instead of signaling two values independently, one can signal their difference as follows.

[00136] The geometry slice header syntax in Table 6 has been modified by adding the following syntax elements on lines 10-11:

[00137] gsh_log2_max_nodesize_x_y specifies the size of the bounding box in x and y dimensions, that is, MaxNodesizeX and MaxNodesizeY, which are used in the decoding process as follows:

[00138] gsh_log2_max_nodesize_z_minus_xy specifies the size of the bounding box in the z dimension, i.e. MaxNodesizeZ, which is used in the decoding process as follows:

[00139] Embodiment G

[00140] In another embodiment of the invention, only one dimension out of three allows for different dimensions. In this case, the dimensions in the x and y dimensions are the same, but the dimension in the z dimension can be different, so the two values are signaled in the geometry slice header with their Cartesian positions as follows.

[00141] The geometry slice header syntax in Table 7 has been changed by adding the following syntax elements on lines 10-11:

[00142] gsh_max_nodesize_x_y specifies the size of the bounding box in the x and y dimensions, i.e. MaxNodesizeX and MaxNodesizeY, which are used in the decoding process as follows:

MaxNodeSizeX=MaxNodeSizeY=gsh_max_nodesize_x_y.

MaxNodeSizeLog2X=MaxNodeSizeLog2Y=ilog2(gsh_max_nodesize_x_y).

[00143] gshmaxnodesizez specifies the size of the bounding box in the z dimension, i.e. MaxNodeSizeZ, which is used in the decoding process as follows:

MaxNodeSizeZ=gsh_max_nodesize_z.

MaxNodeSizeLog2Z=ilog2(gsh_max_nodesize_z).

[00144] Embodiment H

[00145] In another embodiment of the invention, only one dimension out of three allows for different dimensions. In this case, the dimensions in the x and y dimensions are the same, but the dimension in the z dimension can be different, so the two values are signaled in the geometry slice header by their Cartesian positions minus one, as shown below.

[00146] The geometry slice header syntax in Table 8 has been modified by adding the following syntax elements on lines 10-11:

[00147] gsh_max_nodesize_x_y_minus_one specifies the size of the bounding box in x and y dimensions, that is, MaxNodesizeX and MaxNodesizeY, which are used in the decoding process as follows:

MaxNodeSizeX=MaxNodeSizeY=gsh_max_nodesize_x_y_minus_one + 1.

MaxNodeSizeLog2X=MaxNodeSizeLog2Y=

ilog2(gsh_max_nodesize_x_y_minus_one + 1).

[00148] gsh_max_nodesize_z_minus_one specifies the size of the bounding box in z dimension, i.e. MaxNodeSizeZ, which is used in the decoding process as follows:

MaxNodeSizeZ=gsh_max_nodesize_z_minus_one + 1.

MaxNodeSizeLog2Z=ilog2(gsh_max_nodesize_z_minus_one + 1).

[00149] 4. QT and BT Implicit Separation Signaling

[00150] Embodiment A of the invention

[00151] In one embodiment of the invention, the geometry encoding syntax is as follows.

[00152] The syntax of the geometric slice data in Table 9 has been changed by adding or changing syntax elements in lines 3-14. Line 14 adds new variables depthX, depthY, depthZ, and partitionSkip as input to the geometry node function.

[00153] The variables NodeX[depthX][nodeldx], NodeY[depthY][nodeldx], and NodeZ[depthZ] [nodeldx] represent the x, y, and z coordinates of nodeldx in decoding order at a given depth. The variable NumNodesAtDepth[depth] represents the number of nodes to be decoded at the given depth. The variables depthX, depthY, and depthZ define the depth in the x, y, and z dimensions, respectively.

[00154] The partitionSkip variable indicates the type and direction of the partition in the table below (Table 10).

[00155] The partitionSkip variable is represented in binary form by three bits b _x b _y b _z , which determine whether to skip partitioning in the x, y, and z dimensions, respectively. For example, b _x =1 indicates no split in the x dimension (split is skipped). The type and direction of separation may be determined by specific conditions.

[00156] The syntax of the geometric node in Table 11 has been changed by adding or changing lines 0, 4, 8, 11-12 and 24-33. Line 0 introduces new input variables depthX, depthY, depthZ, and partitionSkip to facilitate the separation of QT and BT. In line 12, the condition for checking whether the minimum node size has been reached is changed based on the evaluation of three dimensions, since node sizes in different dimensions can be different when QT and BT splits are applied.

[00157] The syntax structure in lines 24-33 describes the syntax elements of the 3D coordinates of a point in a child node processed by the DM. The variables ChildNodeSizeLog2X, ChildNodeSizeLog2Y, and ChildNodeSizeLog2Z define the size of the child node for each dimension and can be defined by implicit QT and BT separations as follows:

[00158] In the above process, if a split occurs in one of the dimensions, the log2 value of the child node size in that dimension will be equal to the log2 value of the node size in that dimension minus 1.

[00159] In line 4, the occupancy_idx syntax element identifies the index of the current node's single occupied child in octree geometry child node traversal order. If the OccupancyMap variable is present, it can be retrieved as follows:

[00160] The corresponding occupancy_idx parsing process is described below.

The input to this process is the current node's partitionSkip variable.

The output for this process is the value of the occupancy_idx syntax element constructed as follows:

[00161] In the above method, only the bit(s) associated with the measurement(s) where the split occurred are signaled for the QT or BT split. Thus, fewer bits are signaled than in the OT split to indicate the position of one child node in the busy code. Compound bitwise OR operator | = "set" (set to 1) sets certain bits in the occupancy_idx variable.

[00162] In line 8, the occupancy_map syntax element is a bitmap that identifies the occupied child nodes of the current node. If present, the OccupancyMap variable is set to the output of the occupancy geometric map permutation process when called with the NeighborPattern and occupancy map variables as input. NeighborPattern is a neighbor occupancy pattern for contextual modeling. As described with reference to FIG. 8-9, bins in the busy code may be skipped in signaling for QT or BT separation. Four bins may be omitted for QT separation and 6 bins may be omitted for BT separation. Accordingly, the decoder can infer that the bits at these missing bit positions in the busy code are 0.

[00163] The corresponding geometric octree occupancy analysis process can be described as follows, where some bits are inferred to be zero based on the partitionSkip variable, which indicates the partition type and partition direction using the binIsSkiped[binIdx] variable.

This process reconstructs the occupancy_map syntax element.

The inputs to this process are the NeighbourPattern, binIsSkiped, and binlslnferred of the current node. The output for this process is the value of the syntax element constructed as follows:

where the variable binIsSkiped[binIdx] is set as follows:

the variable bitCodingOrder[i] is defined in the table below:

the inverseMap[i][j] variable is defined in the tables below (see tables 13-14):

[00164] Embodiment B

[00165] In another embodiment of the invention, only one dimension out of three allows for different dimensions. In this case, if this dimension has a larger size than the other two, only implicit BT splits are performed in that dimension. If this dimension is smaller than the other two, only implicit QT splits are performed in the other two dimensions. The implicit BT split or implicit QT split signaling in this embodiment is similar to the signaling described in the previous embodiment.

[00166] Embodiment C

[00167] Parameters can be defined to specify specific conditions for implicit QT and BT separations. These parameters may either be fixed for the encoder and decoder (and take on a locally configured pre-configured value such as a default) or may be signaled in the header of the bitstream to allow optimization at the sequence or slice level. Embodiments C-G of the invention are described below to show how parameters useful for specifying conditions for implicit QT and BT separations are signaled or configured.

[00168] In Embodiment C, the parameters are signaled, which can be either in sequence or in the slice header, as shown below.

[00169] The geometry parameter set syntax in Table 15 has been changed by adding syntax elements for signaling parameters to control QT and BT separations in lines 13-15.

[00170] gps_max_num_implicit_qtbt_before_ot specifies the maximum number of implicit QT and BT splits before OT splits, denoted by K.

[00171] gps_min_size_implicit_qtbt indicates the minimum size of implicit QT and BT splits, which is denoted by M. This parameter prevents implicit QT and BT splits when all sizes of the current node are less than or equal to M.

[00172] gps_implicit_bt_before_implicit_qt_flag sets the priority of implicit QT and BT separations, which is denoted by BTFirst. If BTFirst=1, implicit BT splits are performed before implicit QT splits. If BTFirst=0, implicit QT splits are performed before implicit BT splits.

[00173] Embodiment D

[00174] In another embodiment of the invention, some of the parameters are signaled in either the sequence header or the slice header, while the rest of the parameters are fixed. In the following example, M and BTFirst are fixed and K is signaled with gps_max_num_implicit_qtbt_before_ot.

[00175] The geometry parameter set syntax in Table 16 has been changed by adding a syntax element to signal a parameter to control the QT and BT separations in line 13.

[00176] gps_max_num_implicit_qtbt_before_ot indicates the maximum number of implicit QT and BT splits before OT splits. In this case, the other parameters are fixed and therefore not signaled in the bitstream, eg M is always 0, a. BTFirst is always 1.

[00177] Embodiment E

[00178] In another embodiment of the invention, K and BTFirst are fixed, and M is signaled by gps_min_size_implicit_qtbt, as shown below.

[00179] The syntax of the geometry parameter set in Table 17 has been changed by adding a syntax element to signal a parameter to control the QT and BT separations in line 13.

[00180] gps_min_size_implicit_qtbt specifies the minimum size of the implicit QT and BT splits, i.e. M. This parameter prevents implicit QT and BT splits when all sizes are less than or equal to M. In this case, the other parameters are fixed and therefore not signaled in the bitstream, for example, K is always 0 and BTFirst is always 1.

[00181] Embodiment F

[00182] In another embodiment of the invention, parameter signaling for implicit separation depends on other syntax structures. In the example below, parameter signaling for implicit splitting depends on log2_trisoup_node_size as follows.

[00183] The geometry parameter set syntax in Table 18 is changed by adding a syntax structure for signaling a parameter to control the QT and BT separations in lines 13-15.

[00184] gps_max_num_implicit_qtbt_before_ot indicates the maximum number of implicit QT and BT splits before OT splits, which is equal to K. In this case, gps_max_num_implicit_qtbt_before_ot is signaled only if the value of log2_trisoup_node_size is zero. If log2_trisoup_node_size is non-zero, the K parameter will be set to the default maximum value. Other parameters are fixed and therefore not signaled in the bitstream, eg M is always 0, a. BTFirst is always 1.

[00185] Embodiment G

[00186] In another embodiment of the invention, none of these parameters for implicit separation is signaled, and all parameters can be fixed. For example, parameters K=3, M=0 and BTFirst=1 are fixed.

[00187] 5. Conditions for implicit QT and BT separations

[00188] Embodiments of various conditions for implicit QT and BT separations are described in the following subsections. By using different control parameters and setting different conditions, different separation methods or schemes can be carried out.

[00189] 5.1 Performing implicit QT and BT separations after OT

[00190] In the first splitting scheme, OTs are fully performed until some of the measurements can no longer be split. Therefore, the condition for performing implicit QT and BT separations in this scheme is that one or two measurements achieve the smallest separation unit (ie, one voxel).

[00191] Specifically, the split type and direction may be determined according to Table 19 or Table 20. The priority parameter BTFirst is either fixed or set by gps_implicit_bt_before_implicit_qt_flag. If BTFirst = 1, implicit BT splits take precedence and will be performed before implicit QT splits, in which case Table 19 applies. in this case, table 20 applies. If none of the conditions listed in the tables are met, QT splitting is performed.

. Без потери общности можно предположить, что 0<d_x≤d_y≤d_z. Ниже описаны два варианта осуществления изобретения.[00192] Let bounding box B have size

. Without loss of generality, we can assume that 0<d _x ≤d _y ≤d _z . Two embodiments of the invention are described below.

. Затем неявные разделения ВТ будут выполнены вдоль оси z на следующих глубинах d_z - d_y. После неявных разделений ВТ форма подузлов будет иметь вид

, а затем неявные разделения QT выполняются вдоль осей y-z на последних глубинах d_y - d_x, пока не будут достигнуты листовые узлы.[00193] In one embodiment of the invention, the BTFirst=1 parameter is either fixed or set with gps_implicit_bt_before_implicit_qt_flag, which indicates that implicit BT splits take precedence and will be performed before implicit QT splits, and table 19 applies. In this embodiment, embodiments of the invention, OT separations can be performed at first separation depths d _x . After the FROM splits, the subnodes will be of the form

. Then implicit separations of the BT will be performed along the z-axis at the following depths d _z - d _y . After implicit divisions of BT, the form of subnodes will look like

, and then implicit QT splits are performed along the yz axes at the last depths d _y - d _x until leaf nodes are reached.

. Затем неявные разделения QT могут выполняться вдоль осей y-z на следующих глубинах d_y - d_x. После неявных разделений QT форма подузлов будет

, а затем неявные разделения ВТ выполняются вдоль оси z на последних глубинах d_z - d_y до достижения листовых узлов.[00194] In another embodiment of the invention, the BTFirst=0 parameter is either fixed or set with gps_implicit_bt_before_implicit_qt_flag, which indicates that implicit QT splits have higher priority and will be performed before implicit BT splits, and table 20 applies. In this embodiment, implementation of the invention separation FROM will be performed at the first depth d _x separation. After the FROM splits, the subnodes will be of the form

. Then implicit QT separations can be performed along the yz axes at the next depths d _y - d _x . After QT's implicit splits, the shape of the subnodes will be

, and then implicit divisions of the BT are performed along the z axis at the last depths d _z - d _y until leaf nodes are reached.

[00195] 5.2 Performing implicit QT and BT splits before OT

[00196] The second scheme is to perform implicit QT and BT splits before any OT split so that the subnodes are cubic. Therefore, in this case, the condition is that one or two dimensions have smaller dimensions than the largest dimension.

[00197] Specifically, the split type and direction may be determined according to Table 21 or Table 22. The BTFirst priority parameter is either fixed or set by gps_implicit_bt_before_implicit_qt_flag. If BTFirst = 1, implicit BT splits take precedence and will be performed before implicit QT splits, and table 21 applies. If BTFirst = 0, implicit QT splits take precedence and will be performed before implicit BT splits, and table 22 applies. If none of the conditions listed in the tables is met, OT is split.

Без ограничения общности можно считать, что 0<d_x≤d_y≤d_z. Ниже описаны два варианта осуществления изобретения.[00198] Let bounding box B have size

Without loss of generality, we can assume that 0<d _x ≤d _y ≤d _z . Two embodiments of the invention are described below.

, и разделения ОТ выполняются d_x раз для достижения листовых узлов.[00199] In one embodiment of the invention, the parameter BTFirst = 1 is either fixed or set with gps_implicit_bt_before_implicit_qt_flag, which indicates that implicit BT splits have higher priority and will be performed before implicit QT splits, and table 21 applies. In this embodiment, implementation of the invention, the implicit BT splits will be performed along the z-axis at the first depths d _z - d _y , and then the implicit QT splits will be performed along the yz axes at the next depths d _y - d _x . After the implicit QT and BT splits, the size of all subnodes will be

, and OT splits are performed d _x times to reach leaf nodes.

, и разделения ОТ выполняются d_x раз до достижения листовых узлов.[00200] In another embodiment of the invention, the BTFirst=0 parameter is either fixed or set with gps_implicit_bt_before_implicit_qt_flag, which indicates that implicit QT splits have higher priority and will be performed before implicit BT splits, and table 22 applies. In this embodiment, implementation of the invention, implicit QT separations will be performed along the yz axes at first depths d _y - d _x , and then implicit BT separations will be performed along the z axis at subsequent depths d _z - d _y . After the implicit QT and BT splits, the size of all subnodes will be

, and OT splits are performed d _x times until leaf nodes are reached.

[00201] 5.3 Hybrid implicit QT and BT separation scheme

[00202] The third scheme is a combination of the scheme from sections III.5.1 and III.5.3. In this case, the threshold K (0≤K≤max(d _x , d _y , d _z ) - min(d _x , d _y , d _z )) specifies the maximum number of implicit QT and BT splits that can be performed before OT splits . This scheme is a generalization of the first two schemes, it degenerates into the scheme from section III 5.1 when K=0, and degenerates into the scheme from section III.5.2 when K=max(d _x , d _y , d _z ) - min(d _x , d _y , d _z ).

[00203] In particular, at the first K splitting depths, the split type and direction solutions follow the conditions specified in Table 21 or 22, followed by Table 19 or Table 20. The parameter K is either fixed or set with gps_max_num_implicit_qtbt_before_ot. The BTFirst priority parameter is either fixed or set with gps_implicit_bt_before_impIicit_qt_flag. If BTFirst = 1, implicit BT splits take precedence and will be performed before implicit QT splits, and tables 19 and 21 apply. 20 and 22. If none of the conditions listed in the tables is met, the separation of OT is carried out.

Без ограничения общности можно считать, что 0<d_x≤d_y≤d_z.[00204] Let bounding box B have size

Without loss of generality, we can assume that 0<d _x ≤d _y ≤d _z .

где значения δ_у и δ_z (δ_z ≥ δ_у ≥ 0) зависят от значения K. Затем разделения ОТ выполняются d_x раз, при этом оставшиеся подузлы имеют размер

Наконец, согласно таблице 19, неявные разделения ВТ выполняются δ_z - δ_у раз вдоль оси z, а неявные разделения QT затем выполняются δ_у раз вдоль осей y-z.[00205] In one embodiment of the invention, the parameter K is either fixed or set with gps_max_num_implicit_qtbt_before_ot, which indicates that implicit separations of BT and QT will be performed K times first. The BTFirst = 1 parameter is either fixed or set with gps_implicit_bt_before_implicit_qt_flag, indicating that implicit BT splits have higher priority and will be performed before implicit QT splits, and tables 19 and 21 apply. In this embodiment, at the first K depths (K≤d _z -d _x ) implicit splits of BT along the z-axis are performed, and then implicit splits of QT along the yz-axes are performed according to Table 21. After that, the size of the subnodes is

where the values of δ _y and δ _z (δ _z ≥ δ _y ≥ 0) depend on the value of K. Then the OT splits are performed d _x times, with the remaining subnodes having the size

Finally, according to Table 19, implicit BT splits are performed δ _z - δ _y times along the z-axis, and implicit QT splits are then performed δ _y times along the yz axes.

, где значения δ_у и δ_z (δ_z≥δ_у≥0) зависят от значения K. Затем разделения ОТ выполняются d_x раз, при этом оставшиеся подузлы имеют размер

Наконец, согласно таблице 20, неявные разделения QT выполняются δ_у раз вдоль осей y-z, а неявные разделения ВТ затем выполняются δ_z - δ_у раз вдоль оси z.[00206] In another embodiment of the invention, the parameter K is either fixed or set using gps_max_num_implicit_qtbt_before_ot, which indicates that the implicit separation of BT and QT will be performed first K times. The parameter BTFirst=0 is either fixed or set with gps_implicit_bt_before_implicit_qt_flag, which indicates that implicit QT splits take precedence and will be performed before implicit BT splits, and tables 20 and 22 apply. In this embodiment, at the first K depths (K≤d _z -d _x ) implicit QT splits along the yz axes are performed, followed by implicit BT splits along the z axis according to Table 22. After that, the size of the subnodes is

, where the values of δ _y and δ _z (δ _z ≥ δ _y ≥ 0) depend on the value of K. Then the divisions of OT are performed d _x times, while the remaining subnodes have the size

Finally, according to Table 20, QT implicit splits are performed δ _y times along the yz axis, and BT implicit splits are then performed δ _z - δ _y times along the z axis.

[00207] 5.4 Hybrid scheme with minimum size of QT and BT implicit separations

[00208] The fourth scheme imposes additional restrictions on the previous schemes. In this case, the threshold K (0 ≤ K ≤ max(d _x , d _y , d _z ) - min(d _x , d _y , d _z )) specifies the maximum number of implicit QT and BT splits that can be performed before QT splits . Another parameter M (0 ≤ M ≤ min(d _x , d _y , d _z )) specifies the minimum size of implicit QT and BT splits and prevents implicit QT and BT splits when all sizes are less than or equal to M. The fourth scheme is a generalization of the first three schemes. The fourth scheme degenerates into the scheme from Section III. 5.1, when K=M=0, into the circuit from section III.5.2, when K=max(d _x , d _y , d _z ) - min(d _x , d _yj d _z ), M=0, and into the circuit from Section III.5.3 when 0<K<max(d _x , d _y , d _z ) - min(d _x , d _y , d _z ), M=0.

[00209] Specifically, in the first K levels of splitting, the split type and direction decisions follow the conditions given in Table 21 or 22, followed by Table 23 or Table 24. Tables 23 and 24 resemble Tables 19 and 20 when 0 is replaced by M. The K parameter is either fixed or set with gps_max_num_implicit_qtbt_before_ot. The parameter M is either fixed or set using gps_min_size_implicit_qtbt. The BTFirst priority parameter is either fixed or set using gps_implicit_bt_before_implicit_qt_flag. If BTFirst=1, implicit BT splits take precedence and will be performed before implicit QT splits, and tables 21 and 23 apply. If BTFirst=0, implicit QT splits take precedence and will occur before implicit BT splits, and tables 22 and 24. If none of the conditions specified in the tables is met, the division of OT is carried out.

. Без ограничения общности можно считать, что 0<d_x≤d_y≤d_z.[00210] Let bounding box B have size

. Without loss of generality, we can assume that 0<d _x ≤d _y ≤d _z .

, где значения δ_у и δ_z (δ_z ≥ δ_у ≥ 0) зависят от значения K. Затем разделения ОТ выполняются d_x - М раз, при этом оставшиеся подузлы имеют размер

. Затем, согласно таблице 23, неявные разделения ВТ выполняются δ_z - δ_у раз вдоль оси z, а неявные разделения QT выполняются δ_у раз вдоль осей y-z. Остальные узлы имеют размер 2^(М,М,М), поэтому разделения ОТ выполняются М раз для достижения наименьших единиц.[00211] In one embodiment of the invention, the parameter K is either fixed or set using gps_max_num_implicit_qtbt_before_ot, which indicates that the implicit separation of BT and QT will be performed first K times. The parameter M is either fixed or given with gps_min_size_implicit_qtbt, which indicates the minimum size of the implicit QT and BT separations. The BTFirst=1 parameter is either fixed or set with gps_implicit_bt_before_implicit_qt_flag, which indicates that implicit BT splits take precedence and will be performed before implicit QT splits, and tables 21 and 22 apply. In this embodiment, at the first K depths (K≤d _z -d _x ) BT implicit splits are performed along the z-axis, and then QT implicit splits are performed along the yz-axes based on table 21. Then the subnode size becomes

, where the values of δ _y and δ _z (δ _z ≥ δ _y ≥ 0) depend on the value of K. Then the OT divisions are performed d _x - M times, while the remaining subnodes have the size

. Then, according to Table 23, BT implicit splits are performed δ _z - δ _y times along the z-axis, and QT implicit splits are performed δ _y times along the yz axes. The remaining nodes have size 2 ^(M,M,M) , so OT splits are performed M times to reach the smallest units.

, где значения δ_у и δ_z (δ_z>δ_у>0) зависят от значения K. Затем разделения ОТ выполняются d_x - М раз, в результате чего размер оставшихся подузлов становится

. Затем, согласно таблице 24, неявные разделения QT выполняются δ_у раз вдоль осей y-z, а неявные разделения ВТ затем выполняются δ_z - δ_у раз вдоль оси z. Остальные узлы имеют размер 2^(M,M,M), поэтому разделения ОТ выполняются М раз, чтобы достичь наименьших единиц.[00212] In another embodiment of the invention, the parameter K is either fixed or set using gps_max_num_implicit_qtbt_before_ot, which indicates that implicit separations of BT and QT will be performed first K times. The parameter M is either fixed or given with gps_min_size_implicit_qtbt, which indicates the minimum size of the implicit QT and BT separations. The BTFirst=0 parameter is either fixed or set with gps_implicit_bt_before_implicit_qt_flag, which indicates that implicit QT splits take precedence and will be performed before implicit BT splits, and tables 22 and 24 apply. In this embodiment, at the first K depths (K≤d _z -d _x ) BT implicit splits are performed along the z-axis, and then QT implicit splits are performed along the yz-axes based on table 22. The size of the subnodes then becomes

, where the values of δ _y and δ _z (δ _z > δ _y > 0) depend on the value of K. Then the OT splits are performed d _x - M times, as a result of which the size of the remaining subnodes becomes

. Then, according to Table 24, QT implicit splits are performed δ _y times along the yz axes, and BT implicit splits are then performed δ _z - δ _y times along the z axis. The remaining nodes are of size 2 ^(M,M,M) , so OT splits are performed M times to reach the smallest ones.

[00213] 5.5. 2D block-based demonstration for various QT and BT partitioning schemes

[00214] The above four partitioning schemes can be demonstrated using a 2D block as shown in FIG. 10-13, respectively, where the 16x4 rectangular bounding box B is split using 4-level iterative splits to the smallest unit. In addition, a TMC13 scheme in which OT divisions are performed based on an expanded cubic bounding box is illustrated in FIG. 10 for two-dimensional cases. In 2D illustrations, BT resembles BT or QT in 3D, and QT resembles OT in 3D.

[00215] Example circuit from section III.5.1

[00216] FIG. 10, BT is performed after QT, which is equivalent to performing QT and BT after OT in 3D. As shown, B is first divided into four 4x1 sub-blocks by QT bi-level separation, and the remaining sub-block is then divided by BT bi-level separation along the x-axis. The implicit split order is QT, QT, BT, BT.

[00217] An example circuit from section III.5.2

[00218] In Fig. 11, BT is performed before QT, which is equivalent to performing QT and BT before OT in 3D. As shown, box B is first divided by 2-level BT divisions along the x-axis to form four 4×4 sub-blocks, and then each sub-block is divided by QT divisions. Implicit order of divisions BT, BT, QT, QT.

[00219] Example circuit from section III.5.3

[00220] FIG. 12, the separation is performed in the order of BT, QT, BT, which is equivalent to a hybrid scheme in 3D. A hybrid circuit with K=1 is shown in FIG. 12, where box B is once divided by BT splitting to produce two 8x4 subblocks, and each subblock is then split by QT 2-level splits to give several smaller 2x1 subblocks, which are finally subdivided into smaller units by another BT split along x axis. The implicit order of divisions is BT, QT, QT, BT.

[00221] Example circuit from section III.5.4

[00222] FIG. 13, the separation is performed in the order of BT, QT, BT, QT, which is equivalent to a 3D minimum size constraint hybrid scheme. A hybrid circuit with minimum BT size limitation is shown in FIG. 13 where K=M=1. First, BT splitting is performed before QT splitting because K=1, then QT splitting is performed until the smaller size reaches 2^M=2, that is, 4×2 sub-blocks. Then, BT splitting is required due to the minimum BT size limitation, resulting in the remaining sub-blocks having a size of 2×2, and finally, QT splitting is performed. The implicit order of divisions is BT, QT, BT, QT.

[00223] TMC13 design example

[00224] FIG. 14 shows the QT scheme based on the extended bounding box in the 2D case, which is equivalent to the OT separation scheme of the TMC13 design in the 3D case. As shown, the QT recursive splits start with an expanded 16x16 rectangular bounding box, which is shown in dotted lines in the drawing.

[00225] Assuming that the dots are located at the positions marked with crosses in FIG. 10-14, the bits needed for each circuit can be calculated. It can be seen from the first column of Table 25 that four schemes cost fewer bits than the TMC13 scheme. The schemes in sections III 5.3 and III.5.4 work best. The second column of table 25 calculates the required bits in the worst case when all positions are occupied. From this simulation, it can be seen that the smallest number of bits is achieved in the scheme from Section III.5.2, and the schemes from Sections III.5.2 and III.5.4 perform better than the TMC13 scheme. Thus, by a suitable selection of the control parameters (K and M) and conditions, favorable geometry encoding results can be achieved.

[00226] IV. Interaction with the plenary mode

[00227] 1. Plenary mode

[00228] In some embodiments of the invention, planar modes are introduced in the three x, y, and z directions to encode point cloud geometry. The planar modes may be enabled at the applicable node level with a planar mode enable flag encoded in the bitstream. In addition, additional syntax is added to the bitstream to indicate the position of the plane associated with the activated planar modes.

[00229] Flag and plane position prediction is also introduced to ensure good compression of the new syntax element. Finally, local applicability criteria can be used to avoid the use of planar modes in unfavorable areas of the point cloud and thus avoid degradation of compression performance, especially in dense point clouds.

[00230] 2. Problems

[00231] The QT/BT implicit split and planar modes are based on different observations and have similar coding advantages for sparse point clouds. In general, implicit geometric separation has little complexity, while planar mode is more complex but more adaptive. In terms of performance, these two encoding methods provide the same performance for sparse data, while planar mode can better handle point clouds with many plane patterns.

[00232] First of all, the two coding methods have some similar concepts and coding schemes in occupancy coding, both coding methods use a modified occupancy coding scheme where, under certain conditions, only part of the occupancy code is encoded, and the rest of the occupancy code can be skipped, and o it can be inferred on the decoder side. For implicit geometric separation, the condition is implicitly defined by some predefined parameters. While for planar mode the condition is determined by the encoder, the flag and index are explicitly signaled to indicate which part of the busy code is skipped.

[00233] In addition, implicit geometric partitioning can start partitioning the octree from a non-cubic bounding box, and the shape of the octree nodes can also be asymmetric, while planar mode always assumes symmetrical octree node shapes.

[00234] These two methods do not contradict each other in concept. They have similarities/coincidences and differences. In various embodiments of the invention, these two methods can be combined.

[00235] The sections below describe embodiments of the combination of implicit geometric separation and planar mode. The following description shows some embodiments of the invention, which are modifications to the standard design specifications for point cloud compression based on geometry ISG/IEC 23090-9:2019(E), stage CD, ISO/IEC JTC 1/SC 29/WG 11 W18478, June 2019

[00236] 3. Only implicit geometric separation or only planar mode

[00237] In some embodiments, one of the two methods is enabled when encoding the point cloud geometry, i.e. either implicit geometric separation is enabled (when planar mode is disabled) or planar mode is enabled (when implicit geometric partitioning is disabled).

[00238] 3.1 Control Flag Signaling in High Level Syntax for Implicit Geometric Separation

[00239] A single flag can be signaled in the high-level syntax to enable and disable implicit geometric separation. The flag may be specified in a sequence parameter set, a slice header, or a bitstream geometry parameter set.

[00240] Embodiment A of the invention

[00241] In one embodiment of the invention, the implicit geometric separation flag is set in the geometry parameter set as follows.

[00242] The syntax of the geometry parameter set in Table 26 has been changed by adding a syntax element in line 17.

[00243] A gps_implicit_geom_partition_flag value of 1 indicates that implicit geometric partitioning is enabled for the sequence or slice. A gps_implicit_geom_partition_flag value of 0 indicates that implicit geometric partitioning is disabled for the sequence or slice.

[00244] Embodiment B

[00245] In another embodiment of the invention, the flag is indicated when the triangular mesh scheme is disabled, in which case the geometry parameter set changes are shown in lines 6-8 in Table 27.

[00246] When the value of log2_trisoup_node_size is greater than 0, it can be concluded that the value of gps_implicit_geom_partition_flag is 0, without explicit signaling.

[00247] Embodiment C

[00248] In another embodiment of the invention, in addition to the flag, other parameters that relate to implicit geometric partitioning may be specified in the high-level syntax when the gps_implicit_geom_partition_flag flag is 1. The geometry parameter set syntax in Table 28 is changed in lines 6-10 .

[00249] gps_max_num_implicit_qtbt_before_ot indicates the maximum number of implicit QT and BT splits before OT splits. gps_min_size_implicit_qtbt sets the minimum size for implicit QT and BT separations.

[00250] Embodiment D

[00251] In another embodiment of the invention, the geometry bounding box signaling may depend on the value of gps_implicit_geom_partition_flag as follows. The geometry slice header syntax in Table 29 has been modified by adding a syntax structure in lines 10-16.

[00252] gsh_log2_max_nodesize_x, gsh_log2_max_nodesize_y, and gsh_log2_max_nodesize_z define the dimensions of the bounding box in x, y, z dimensions in log2 scale, respectively. They are only specified when gps_implicit_geom_partition_flag is 1. When gps_impIicit_geom_partition_flag is 0, only one size is specified using gsh_log2_max_nodesize, in which case the three dimensions are assumed to be the same size.

[00253] Embodiment E

[00254] In another embodiment of the invention, the dimensions of the bounding box are given by their differences. The geometry slice header syntax in Table 30 is changed in lines 10-16.

[00255] 3.2 Control Flag Signaling in High-Level Syntax Considering Combination Schema

[00256] In this subsection, several embodiments of the invention are described as examples to show how to signal control flags in high-level syntax when considering a combination of implicit geometric separation and planar mode.

[00257] Embodiment A of the invention

[00258] In one embodiment of the invention, the two control flags are set independently in the geometry parameter set as follows. The syntax of the geometry parameter set in Table 31 is changed in lines 6-7.

[00259] A gps_planar_mode_flag value of 1 indicates that planar mode is enabled for the sequence or slice. A gps_planar_mode_flag value of 0 indicates that planar mode is disabled for the sequence or slice.

[00260] However, in this combination scheme, it is assumed that only one of the two methods is included. For example, gps_implicit_geom_partition_flag and gps_planar_mode_flag cannot both be 1.

[00261] Embodiment B

[00262] In another embodiment of the invention, the control flag gps_planar_mode_flag is specified depending on the value of gps_implicit_geom_partition_flag in the geometry parameter set as follows. The syntax of the geometry parameter set in Table 32 is changed in lines 6-8.

[00263] In this case, gps_planar_mode_flag is only indicated when the value of gps_implicit_geom_partition_flag is 0. When the value of gps_implicit_geom_partition_flag is 1, it can be concluded that the value of gps planar mode flag is 0.

[00264] Embodiment C

[00265] In another embodiment of the invention, the control flag gps_impIicit_geom_partition_flag is specified depending on the value of gps_planar_mode_flag in the geometry parameter set as follows. The syntax of the geometry parameter set in Table 33 is changed in lines 6-8.

[00266] In this case, gps_implicit_geom_partition_flag is only indicated when the value of gps_planar_mode_flag is 0. When the value of gps_planar_mode_flag is 1, it can be concluded that the value of gps implicit geom partition flag is 0.

[00267] Embodiment D

[00268] In another embodiment of the invention, the control flag gps_impIicit_geom_partition_flag is specified depending on the value of gps_planar_mode_flag and the value of log2_trisoup_node_size in the geometry parameter set as follows. The syntax of the geometry parameter set in Table 34 is changed in line 6-8.

[00269] In this case, gps_implicit_geom_partition_flag is only indicated when gps_planar_mode_flag and log2_trisoup_node_size are 0. Otherwise, it can be inferred that the value of gps_implicit_geom_partition_flag is 0. In other words, implicit geometric partitioning can only be enabled when disabled as triangular mode. grids and planar mode.

[00270] 4. Planar mode is only available when the current octree node is a cube

[00271] The second combination scheme may simultaneously include an implicit geometric separation and a planar mode. But for the planar mode, there are some limitations. If implicit geometric separation is enabled, the bounding box and octree nodes may not be a cube. In this scheme, the limitation for applying planar mode is that planar mode is only appropriate when the current octree node is a cube, which indicates d _x =d _y =d _z .

[00272] In one embodiment of the invention, the geometry encoding syntax is shown in Table 35 and may be similar to the syntax in Table 9.

[00273] The variables ChildNodeSizeXLog2, ChildNodeSizeYLog2, and ChildNodeSizeZLog2 indicate the size of the child node for each dimension, and can be determined using implicit QT and BT separations as follows:

[00274] In this embodiment, planar mode is not applicable if NodeSizeLog2X, NodeSizeLog2Y and NodeSizeLog2Z are not equal, as shown below,

where planarModeEligibilityX, planarModeEligibilityY and planarModeEligibilityZ indicate whether the planar mode is appropriate for the X, Y, Z dimensions, respectively, at the current encoded node.

[00275] 5. Planar mode is not applicable for some measurements that are not separated

[00276] The third combination scheme further relaxes the planar mode limitation, where planar mode is applicable even if the current octree node is not a cube. If one or two measurements are not separated at the same depth, planar mode is not applicable for these measurements, while it is applicable for the remaining measurements. For example, if you decide to perform an implicit geometric split in the x and y dimensions, but not in the z dimension at a certain split depth, then planar mode is only suitable for the x and y dimensions.

[00277] In this embodiment, planar mode is not applicable for certain measurements where separation is omitted in favor of an implicit geometric separation method as follows,

[00278] It should be noted that in other examples, the planar mode may be modified in other aspects and in a more complex manner to accommodate the fact that the octree node may now be a cuboid.

[00279] V. Additional Embodiments of Implicit Geometric Separation

[00280] Embodiment A

[00281] Embodiment A provides improved implicit geometric separation and saves more bits when encoding a busy code with implicit geometric separation. In particular, the geometric octree occupancy analysis process described in section III can be modified by introducing a process for defining the variable binlslnferred at the end of the geometry octree occupancy analysis process.

This process reconstructs the occupancymap syntax element.

The input to this process is the current node's NeighbourPattern, binlsSkiped, and binlslnferred.

The output for this process is the value of the syntax element constructed as follows:

where the variable binIsSkiped[binIdx] is set as follows:

and for each bean, the binIsInferred variable is set as follows:

- If one of the following conditions is met, the binIsInferred variable is set to 1:

- the variable NeighbourPattem is equal to 0, and the number of previously decoded single-valued bins is less than or equal to (binIdx + minOccupied - maxOccupied).

- the variable NeighbourPattem is not equal to 0, the value of binldx is equal to maxOccupied-1, and the values of all previous decoded bins are equal to zero,

while minOccupied = 2, and maxOccupied = 8 (if OT is applied), maxOccupied = 4 (if QT is applied) and maxOccupied = 2 (if BT is applied).

- Otherwise, if none of the above conditions is met, the binIsInferred variable is set to 0.

[00282] Embodiment B of the invention

[00283] In one embodiment of the invention, the control flag gps_implicit_geom_partition_flag and the two parameters K and M are set in the geometry parameter set as follows.

[00284] A gps_implicit_geom_partition_flag value of 1 indicates that implicit geometric partitioning is enabled for the sequence or slice. A gps_implicit_geom_partition_flag value of 0 indicates that implicit geometric partitioning is disabled for the sequence or slice. If the value of gps_implicit_geom_partition_flag is 1, the following two parameters are signaled:

[00285] (1) The gps_max_num_implicit_qtbt_before_ot value specifies the maximum number of implicit QT and BT splits before OT splits, that is, K=gps_max_num_implicit_qtbt_before_ot.

[00286] (2) The gps_min_size_implicit_qtbt value specifies the minimum size of the implicit QT and BT separations, i.e., M=gps_min_size_implicit_qtbt. This M option prevents QT and BT from being implicitly separated when all dimensions are less than or equal to M.

[00287] Embodiment C

[00288] In one embodiment of the invention, if implicit QTBT separation is enabled, the size of the bounding box is given by three values in the geometry slice header as follows:

[00289] gsh_log2_max_nodesize_x indicates the size of the bounding box in the x dimension, i.e. MaxNodesizeXLog2, which is used in the decoding process as follows:

MaxNodeSizeXLog2=gsh_log2_max_nodesize_x.

MaxNodeSizeX=1<<MaxNodeSizeXLog2.

[00290] gsh_log2_max_nodesize_y_minus_x specifies the size of the bounding box in the y dimension, i.e. MaxNodesizeYLog2, which is used in the decoding process as follows:

MaxNodeSizeYLog2=gsh_log2_max_nodesize_y_minus_x+MaxNodeSizeXLog2.

MaxNodeSizeY=1<<MaxNodeSizeYLog2.

[00291] gsh_log2_max_nodesize_z_minus_y specifies the size of the bounding box in the z dimension, i.e. MaxNodesizeZLog2, which is used in the decoding process as follows:

MaxNodeSizeZLog2=gsh_log2_max_nodesize_z_minus_y + MaxNodeSizeYLog2.

MaxNodeSizeZ=1<<MaxNodeSizeZLog2.

[00292] The K and M parameters are then updated as follows:

In an embodiment of the invention, when triangular grid mode is enabled (log2_trisoup_node_size != 0 is true), M is changed to the minimum log2 value of the slice root node size. It should be noted that the log2 value of the node size in triangular grid mode should not be greater than the minimum log2 value of the root node size of the slice.

[00293] Embodiment D

[00294] In an embodiment, the geometric slice data syntax in Table 37 is used in place of the syntax structure in Table 35 or Table 9. The QT/BT implicit decision function is reintroduced in row 3 of Table 37. The QT/BT implicit decision function syntax is shown in Table 38, where the variable partitionSkip is defined on lines 9-16 and the depths in the x, y, and z directions are updated on lines 17-23.

[00295] VI. Geometry Decoding Process Examples

[00296] FIG. 15 shows a flowchart describing a method (1500) according to an embodiment of the invention. For example, method (1500) may be used to decode a point cloud. Method (1500) may be used to create an octree structure for representing the geometry of points in a slice bounding box in a point cloud decoder. In various embodiments, the method (1500) may be performed by a processing circuit, such as a processing circuit performing the functions of an arithmetic decoding module (410) and an octree decoding module (430) in the decoder (400). In some embodiments, method (1500) is implemented as program instructions such that when the program instructions are executed by the processing circuit, the processing circuit executes method (1500). The method starts from step (S1501) and proceeds to step (S1510).

[00297] In step (S1510), a bitstream including a slice of an encoded point cloud frame may be received. For example, a slice may include a sequence of syntax elements and bins representing part of an encoded frame or an entire encoded frame of a point cloud.

[00298] In step (S1520), an octree representing the geometry of the points in the bounding box of the slice can be reconstructed. In the octree construction process, the current node of the octree may be partitioned by quadtree partitioning (QT) or binary tree partitioning (BT). For example, a method for splitting the current octree node by one of QT splitting, BT splitting, or octree based splitting (QT) can be determined based on predetermined conditions. The method of splitting the current octree node by one of QT splitting, BT splitting, or OT splitting may also be determined based on one or more parameters, each of which may be signaled in the bitstream or may have a pre-locally configured value.

[00299] During construction of the octree, the value of a variable denoted partitionSkip and indicating the type of partition and direction of partition of the current node of the octree can be determined. For example, the partitionSkip variable can be represented in binary form with three bits corresponding to the x, y, and z directions, respectively, with each bit indicating whether the partition is performed along the corresponding x, y, or z direction. Based on the partitionSkip variable, the depth in the x, y, or z dimension for a child node of the current octree node can be updated.

[00300] During octree construction, busy bits belonging to the 8-bin busy code of the current octree node may be received (or parsed) from the bitstream. Each busy bit corresponds to a busy child node of the current octree node. The four bins that belong to the 8-bin busy code are not signaled in the bitstream when the current octree node is split by QT splitting, and the six bins that belong to the 8-bin busy code are not signaled in the bitstream when the current octree node is split by VT separation.

[00301] In addition, one or more syntax elements may be received from the bitstream indicating the three-dimensional (3D) dimensions of the encoded point cloud slice bounding box. For example, a slice's bounding box may be in the shape of a cuboid.

[00302] During octree construction, after receiving a syntax element indicating that the current octree node has one occupied child node, 1 bin can be obtained if the partitionSkip variable indicates a BT split, or 2 bins can be obtained if the variable partitionSkip points to QT partitioning. An occupancy map may be defined identifying occupied child nodes of the current octree node, based on received 1 or 2 bins.

[00303] During octree construction, a bitstream parsing process may be performed to determine an occupancy map syntax element identifying occupied child nodes of the current octree node. During parsing, you can determine the busy map syntax element bean based on the partitionSkip variable.

[00304] During octree construction, a child node of the current node of the direct-coded octree, for each of the x, y, and z directions, can be assigned a log2 size value, denoted by dx, dy, and dz, respectively, for the child node based on the variable partitionSkip. It may be determined that the positions of points at a child node are encoded by fixed length encoding using (dx, dy, dz) bits, respectively. Accordingly, bins corresponding to the coordinates of points at the child node can be received based on known encoding lengths.

[00305] In an example, for octree reconstruction, a syntax element may be received from the bitstream indicating one of the following parameters: the maximum number of implicit QT and BT splits performed before OT splits, the minimum size of implicit QT and BT splits that prevents implicit QT splits and node BT when all node sizes are less than or equal to the minimum size, or a priority indicating which of the implicit QT or BT splits is performed first when both QT and BT splits are enabled.

[00306] In the example, at the time of constructing the octree, when the octree depth of the current node is less than the parameter K, or when the smallest log2 size value among the log2 sizes of the current node in the x, y, and z directions is equal to the parameter M, the split type and the split direction to split the current node can be defined according to the conditions specified in the following table.

[00307] The parameter K is an integer in the range 0 ≤ K ≤ max(d _x , d _y , d _z ) - min(d _x , d _y , d _z ) and specifies the maximum number of implicit QT and BT splits that are allowed up to divisions OT. The parameter M is an integer in the range 0<M<min(d _x , d _y , d _z ) and specifies the minimum size of implicit QT and BT splits, which prevents implicit QT and BT node splits when all node sizes are less than or equal to M. The dimensions dx, dy, and dz can be log2 values of the dimensions of the current node in the x, y, and z directions, respectively. In the example, the above partition type and partition direction for partitioning the current node can be represented by the partitionSkip variable.

[00308] In an example, each of the parameters K and M may be signaled to the decoder or take on pre-configured value(s) and then updated, for example, based on the 3D dimensions of the slice root node or based on whether triangular mode is enabled. grids.

[00309] It should be noted that the parameters K or M, or other parameters used to control or implement implicit QT/BT separation, are described simply as examples to illustrate some embodiments of the invention. There may be various forms of parameters signaled in the bitstream and used to perform QT/BT separations at the decoder. These parameters may or may not be similar to K or M, but can be similarly used to control QT/BT splits so that point cloud compression performance can be improved.

[00310] After step (S1520), the process (1500) may end at step (S1599).

[00311] VII. computer system

[00312] The above methods may be implemented as computer software using computer readable instructions and physically stored on one or more computer readable media. For example, in FIG. 16 shows a computer system (1600) suitable for practicing some embodiments of the invention.

[00313] Computer software may be coded using any suitable machine code or computer language that may be assembled, compiled, linked, or similarly processed to produce code containing instructions that can be executed directly or through interpretation, execution of microcode, etc. etc., one or more computer central processing units (CPU, central processing units), graphic processors (GPU, Graphics Processing Units), etc.

[00314] The instructions may be executed on various types of computers or components thereof, including, for example, personal computers, tablet computers, servers, smartphones, gaming devices, IoT devices, and the like.

[00315] The computer system components (1600) shown in FIG. 16 are illustrative in nature and are not intended to impose any limitation on the scope of application or functionality of computer software implementing embodiments of the present invention. The component configuration should also not be interpreted as having any dependency or requirement in connection with any component or combination of components shown in the exemplary computer system (1600).

[00316] The computer system (1600) may include some human-machine interface input devices. Such an input device may be responsible for input by one or more users through, for example, tactile input (e.g., keystrokes, waving, glove movements), audio input (e.g., voice, claps), visual input (e.g., gestures), olfactory input (not shown). Interface devices may also be used to capture certain media not necessarily directly related to conscious human input, such as sound (e.g. speech, music, external sound), images (e.g. scanned images, photographic images received from a still camera), video (eg 2D video, 3D video including stereoscopic video).

[00317] The input human-machine interface of the device may include one or more of the following (shown one at a time): keyboard (1601), mouse (1602), touch pad (1603), touch screen (1610), information glove (not shown) , joystick (1605), microphone (1606), scanner (1607) and camera (1608).

[00318] The computer system (1600) may also include some human interface output devices. Such output devices may stimulate the senses of one or more users through, for example, tactile output, sound, light, and smell/taste. Such output devices may include tactile output devices (e.g., tactile feedback via touch screen (1610), data glove (not shown) or joystick (1605), as well as tactile feedback devices that do not serve as input devices), audio output devices (for example: speakers (1609), headphones (not shown)), visual output devices (for example, screens (1610), including CRT screens, LCD screens, plasma screens, OLED screens, each with touch screen input capabilities or without, each with or without haptic feedback capabilities, some of which are capable of 2D visual output or more than 3D output via, for example, stereographic output, virtual reality goggles (not shown), holographic displays, smoke tanks (not shown). ) and printers (not shown).

[00319] The computer system (1600) may also include human accessible storage devices and associated media, such as optical media including CD/DVD ROM/RW (1620) with CD/DVD media (1621) and the like, flash memory cards (1622), removable hard drive or solid state drive (1623), traditional magnetic media such as tape and magnetic disk (not shown), dedicated ROM/ASIC/PLD based devices such as security dongles (not shown). shown), etc.

[00320] Those skilled in the art should also understand that the term "computer-readable media" as used in connection with the present invention does not encompass transmission media, carrier waves, or other time signals.

[00321] The computer system (1600) may also include an interface (1654) to one or more communication networks (1655). Networks can be, for example, wireless, wired, optical. Networks can also be local, global, urban, transport and industrial, real-time, latency, etc. Examples of networks include local area networks, such as Ethernet, wireless LANs, cellular networks, including GSM, 3G, 4G, 5G, LTE, etc., digital wide area networks of wired or wireless television, including cable television, satellite television and terrestrial broadcasting, transport and industrial networks, including CANBus, etc. Some networks typically require external network interface adapters that are connected to some general purpose data ports or peripheral buses (1649) (for example, computer system USB ports (1600)); others are typically built into the core of the computer system (1600) by connecting to a system bus as described below (eg, an Ethernet interface to a PC computer system or a cellular network interface to a smartphone computer system). Using any of these networks, the computer system (1600) can communicate with other entities. Such communication may be unidirectional with receive-only capability (eg broadcast), unidirectional with transmit-only capability (eg CANbus to some CANbus devices), or bidirectional, eg to other computer systems using a local area network or a wide area digital network. Several protocols and protocol stacks may be used on each of these networks and the network interfaces described above.

[00322] The aforementioned human-machine interface devices, human-accessible storage devices, and network interfaces can be connected to the core (1640) of the computer system (1600).

[00323] The core (1640) may include one or more central processing units (CPU) (1641), graphics processing units (GPU) (1642), specialized programmable processing modules in the form of field programmable gate arrays (FPGA, Field Programmable Gate Arrays) ( 1643), hardware accelerators (1644) for some tasks, graphics adapters (1650), etc. These devices, together with read-only memory (ROM) (1645), random access memory (1646), high-capacity internal storage such as user-inaccessible internal hard drives, SSDs, etc. (1647), can be connected via the system bus (1648). In some computer systems, the system bus (1648) may be available in the form of one or more physical connectors to allow for expansions with additional CPUs, GPUs, or the like. Peripherals can be connected either directly to the system bus (1648) of the kernel, or via a peripheral bus (1649). For example, a screen (1610) may be connected to a graphics adapter (1650). Peripheral bus architectures include PCI, USB, and the like.

[00324] CPU (1641), GPU (1642), FPGA (1643), and accelerator (1644) devices may execute some instructions that together may constitute the aforementioned computer code. This computer code may be stored in ROM (1645) or RAM (1646). Temporary data may also be stored in a RAM memory (1646), while permanent data may be stored in, for example, an internal large data storage (1647). Fast storage and retrieval from any storage device can be achieved through the use of cache memory, which can be closely associated with one or more processors CPU (1641), GPU (1642), mass storage (1647), ROM memory (1645) , RAM (1646), etc.

[00325] Computer-readable media may store computer code for performing various computer-executable operations. The media and computer code may be specially created for the purposes of the present invention or may be well known and available to those skilled in the art of computer software.

[00326] By way of example, and not limitation of the invention, a computer system having an architecture (1600), and in particular a core (1640), can provide functionality by being executed by the processor(s) (including(and) CPU, GPU , FPGAs, accelerators, etc.) software embodied in one or more tangible computer-readable media. Such computer-readable media may be media associated with the user-accessible mass storage described above, as well as some persistent kernel storage (1640), such as internal mass storage (1647) or ROM (1645). Software implementing various embodiments of the present invention may be stored on such devices and executed by the kernel (1640). The computer-readable medium may include one or more storage devices or microcircuits, in accordance with specific needs. The software may cause the core (1640) and, in particular, its processors (including CPUs, GPUs, FPGAs, and the like) to perform specific methods or specific parts of the specific methods described here, including setting data structures stored in RAM memory (1646 ), and modifying such data structures according to the methods specified by the software. Additionally or alternatively, the computer system may provide functionality through logic hardwired or otherwise embodied in circuitry (e.g., an accelerator (1644)), which may operate in place of, or in conjunction with software, to perform specific methods or specific portions of the specific methods described herein. ways. A reference to software may cover logic, and vice versa when appropriate. Reference to computer-readable media may cover a circuit (eg, an integrated circuit (IC, integrated circuit)), where software is stored for execution, a circuit embodying logic for execution, or both circuits, when appropriate. The present invention encompasses any suitable combination of hardware and software.

[00327] Although aspects of the present disclosure have been described in connection with its specific embodiments, which are offered as examples, alternatives, modifications, and variations of the examples can be made. Accordingly, the embodiments set forth herein are intended to be illustrative and not restrictive. Changes may be made without departing from the scope of the claims set forth below.

Claims (60)

1. A method for decoding the geometry of a point cloud in a point cloud decoder, including: receiving (S1510) a bitstream including a slice of the encoded point cloud frame, and reconstructing (S1520) an octree representing the geometry of points in a slice bounding box, where the current node of the octree is split by quadtree based splitting (QT) or binary tree based splitting (BT), wherein the reconstruction (S1520) of the octree includes: defining the value of a variable, denoted partitionSkip, that specifies the type of partition and direction of partition of the current octree node. 2. The method of claim 1, wherein reconstructing (S1520) the octree includes: determining how to split the current octree node using one of QT split, BT split, or octree based split (OT) based on a predetermined condition. 3. The method of claim 1, wherein reconstructing (S1520) the octree includes: determining how to split the current octree node using one of QT split, BT split, or OT split based on one or more parameters, wherein one of said one or more parameters are signaled in the bitstream or using a locally configured pre-configured value. 4. The method of claim 1, wherein reconstructing (S1520) the octree includes: receiving busy bins belonging to an 8-bin busy code of the current octree node from a bitstream, each busy bin corresponds to a busy child node of the current octree node, The 4 bins that belong to the 8-bin busy code do not signal in the bitstream when the current octree node is split by QT splitting, and The 6 bins that belong to the 8-bin busy code do not signal in the bitstream when the current octree node is divided by BT splitting. 5. The method of claim 1, wherein reconstructing (S1520) the octree includes: receiving one or more syntax elements indicating three-dimensional (3D) dimensions of a point cloud encoded frame slice bounding box from a bitstream. 6. The method of claim 1, wherein the partitionSkip variable is represented in binary form by three bits corresponding to the x, y, and z directions, respectively, with each bit indicating whether to partition along the respective x, y, or z directions. 7. The method of claim 1, wherein reconstructing the octree also includes: updating the depth in the x, y, or z dimension for a child node of the current octree node based on the partitionSkip variable. 8. The method of claim 1, wherein reconstructing the octree also includes: receiving a syntax element indicating that the current octree node has a single occupied child node; getting 1 bin if the partitionSkip variable points to a BT partition, or 2 bins if the partitionSkip variable points to a QT partition; and determining an occupancy map identifying occupied child nodes of the current octree node based on the received 1 or 2 bins. 9. The method of claim 1, wherein reconstructing the octree also includes: in the process of parsing the bitstream to determine the busy map syntax element that identifies busy child nodes of the current octree node, omitting the definition of the busy map syntax element bean based on the partitionSkip variable. 10. The method of claim 1, wherein reconstructing the octree also includes: for a child node of the current node of the direct-coded octree, determining for each of the x, y, and z directions a log2 value of size, denoted dx, dy, and dz, respectively, for the child node based on the variable partitionSkip, with the positions of the points in the child node being coded fixed length codes using (dx, dy, dz) bits, respectively. 11. The method of claim 1, wherein reconstructing (S1520) the octree also includes: receiving from the bitstream a syntax element indicating one of the following: the maximum number of implicit QT and BT splits performed before OT splits, the minimum size of implicit QT and BT splits, which prevents implicit QT and BT splits of a node when all node sizes are less than or equal to the minimum size, or a priority indicating which of the implicit QT or BT splits to perform first when both QT and BT splits are enabled. 12. The method of claim 1, wherein reconstructing (S1520) the octree also includes: in response to the current node's octree depth being less than the parameter K, or the smallest log2 size value among the log2 sizes of the current node in the x, y, and z directions is equal to the M parameter, specifying the split type and split direction to split the current node according to the conditions specified in the following table:

where parameter K is an integer in the range 0≤K≤max(d _x , d _y , d _z ) - min(d _x , d _y , d _z ) and specifies the maximum number of implicit QT and BT splits that are allowed before OT splits , parameter M is an integer in the range 0≤M≤min(d _x , d _y , d _z ), specifies the minimum size of implicit QT and BT splits, and prevents implicit splits QT and BT of a node when all node sizes are less than or equal to M, and d _x , d _y , and d _z are the log2 values of the sizes of the current node in the x, y, and z directions, respectively. 13. The method of claim 1, wherein reconstructing (S1520) the octree also includes: in response to the current node's octree depth being less than the K parameter, or the smallest log2 size value among the log2 sizes of the current node in the x, y, and z directions is equal to the M parameter, the definition of the partitionSkip variable is as follows:

where the partitionSkip variable is represented in binary form by three bits and indicates the type of partition and the direction of partition of the current node of the octree, parameter K is an integer in the range 0≤K≤max(d _x , d _y , d _z ) - min(d _x , d _y , d _z ) and specifies the maximum number of implicit QT and BT splits that are allowed before OT splits, parameter M is an integer in the range 0≤M≤min(d _x , d _y , d _z ), specifies the minimum size of the implicit splits QT and BT of the node, and prevents implicit splits QT and BT of the node when all node sizes are less than or equal to M, dx, dy and dz are the log2 values of the sizes of the current node in the x, y and z directions, respectively, MaxNodeDimLog2 represents the maximum log2 size among dx, dy and dz, and the |= operator is a compound bitwise OR operation. 14. The method of claim 1, wherein reconstructing (S1520) the octree also includes: receiving from the bitstream a flag indicating whether implicit geometric separation is allowed for a sequence of point cloud frames or for a slice of an encoded point cloud frame. 15. The method of claim 1, wherein reconstructing (S1520) the octree includes: determining that the planar mode is not applicable in the x, y, or z direction, however, no split is performed for the current node. 16. The method of claim 1, wherein reconstructing (S1520) the octree includes: performing a geometric octree occupancy analysis process, where, for a bin having index binIdx in the occupancy code, the binIsInferred variable is set as follows: if any of the following conditions are met, the binIsInferred variable is set to 1: (1) NeighbourPattem is 0 and the number of previously decoded single bins is less than or equal to (binIdx + minOccupied - maxOccupied), or (2) NeighborPattem is not 0, binIdx is maxOccupied-1, and all previous decoded bins are zero, where minOccupied = 2 and maxOccupied = 8 if OT split is used, maxOccupied = 4 if QT split is used, and maxOccupied = 2 if BT split is used, and otherwise, if none of the above conditions are met, the binlslnferred variable is set to 0. 17. The method of claim 1, wherein reconstructing (S1520) the octree also includes: updating the parameter K indicating the maximum number of implicit QT and BT splits before OT splits and the parameter M specifying the minimum size of implicit QT and BT splits that prevents implicit QT and BT splits of a node when all node sizes are less than or equal to the minimum size, as follows : if K is greater than the difference of the maximum value log2 of the root node size and the minimum value of the log2 size of the root node of the slice, K is changed to the difference of the maximum value of the log2 size of the root node and the minimum value of the log2 size of the root node of the slice; if M is greater than the minimum log2 value of the slice root node size, M is changed to the minimum log2 value of the slice root node size; if the maximum value log2 of the root node size and the minimum value log2 of the slice root node size are equal, M is changed to 0; and if the triangular grid mode is enabled, K is changed to the difference between the maximum value log2 of the root node size and the minimum value of the log2 size of the root node of the slice, and M is changed to the minimum value of the log2 size of the root node of the slice. 18. The method of claim 1, wherein reconstructing (S1520) the octree also includes: reconstructing an octree representing the geometry of points in a slice bounding box, determining a split type and a split direction for the current node of the octree that is split by a split based on a quadtree ( QT) or binary tree partitioning (BT). 19. A device for decoding the geometry of a point cloud, containing a circuit configured to implement the method according to any one of paragraphs. 1-18. 20. A computer-readable medium that stores instructions that, when executed by the processor, cause the processor to carry out the method according to any one of paragraphs. 1-18.

Images