WO2019115867A1 - An apparatus, a method and a computer program for volumetric video - Google Patents

Abstract

There are disclosed various methods, apparatuses and computer program products for video encoding.The method comprises inputting a point cloud frame in an encoder; projecting a3D object represented by the point cloud frame onto a first arbitrary shaped 2D patch for generation of at least a geometry image and a texture image;subdividing the first 2D patch into image blocks of a predetermined size along a predetermined block grid;maintaining, in the first 2D patch, only fully filled image blocks where each pixel has a projected value;removing the 3D object points having a corresponding 2D projection point in the fully filled image blocks of the first 2D patch from the 3D object; encoding the first 2D patch of the geometry image or texture image having only fully filled image blocks of the predetermined size into a bitstream of encoding video frames; and subjecting remaining 3D object points iteratively for one or more new projections into subsequent 2D patches.

Description

AN APPARATUS, A METHOD AND A COMPUTER PROGRAM FOR

VOLUMETRIC VIDEO

TECHNICAL FIELD

[0001] The present invention relates to an apparatus, a method and a computer program for content dependent projection for volumetric video coding and decoding.

BACKGROUND

[0002] Volumetric video data represents a three-dimensional scene or object and can be used as input for virtual reality (VR), augmented reality (AR) and mixed reality (MR) applications. Such data describes the geometry, e.g. shape, size, position in three-dimensional (3D) space, and respective attributes, e.g. colour, opacity, reflectance and any possible temporal changes of the geometry and attributes at given time instances. Volumetric video is either generated from 3D models through computer-generated imagery (CGI), or captured from real-world scenes using a variety of capture solutions, e.g. multi-camera, laser scan, combination of video and dedicated depth sensors, and more. Also, a combination of CGI and real-world data is possible.

[0003] Typical representation formats for such volumetric data are triangle meshes, point clouds (PCs), or voxel arrays. Representation of the 3D data depends on how the 3D data is used. Dense Voxel arrays have been used to represent volumetric medical data. In 3D graphics, polygonal meshes are extensively used. Point clouds on the other hand are well suited for applications such as capturing real world 3D scenes where the topology is not necessarily a 2D manifold.

[0004] In dense point clouds or voxel arrays, the reconstructed 3D scene may contain tens or even hundreds of millions of points. One way to compress a time- varying volumetric scene/object is to project 3D surfaces on to some number of pre-defined 2D planes. Regular 2D video compression algorithms can then be used to compress various aspects of the projected surfaces. For e.g. a time-varying 3D point cloud, with spatial and texture coordinates, can be mapped into a sequence of at least three sets of planes, where a first set carries the temporal motion image data, a second set carries the texture data and a third set carries the depth data, i.e. the distance of the mapped 3D surface points from the projection surfaces.

[0005] For the MPEG standard, there has been developed a test model for point cloud compression. MPEG W17248 discloses a projection-based approach for a test model for standardisation of dynamic point cloud compression. In intra coding of this test model, projected texture and geometry data is accompanied with an additional binary occupancy map to signal if a 2D pixel should be reconstructed to 3D space at the decoder side.

[0006] However, transmitting the additional occupancy map is costly in terms of the bit rate budget. In IPPP coding structure (intra frame followed by three uni-predicted inter frames), the additional occupancy map is only transmitted in every fourth frame (I-frame). Thus, 3D resampling and additional motion information, possibly including padding and/or interpolation, is required to align the I-frame occupancy map to the three P-frames without transmitted occupancy map.

SUMMARY

[0007] Now, an improved method and technical equipment implementing the method has been invented, by which the above problems are alleviated. Various aspects of the invention include a method, an apparatus and a computer readable medium comprising a computer program or a signal stored therein, which are characterized by what is stated in the

independent claims. Various details of the invention are disclosed in the dependent claims and in the corresponding images and description.

[0008] According to a first aspect, there is provided a method comprising: inputting a point cloud frame in an encoder; projecting a 3D object represented by the point cloud frame onto a first arbitrary shaped 2D patch for generation of a geometry image and a texture image;

subdividing the first 2D patch into image blocks of a predetermined size along a

predetermined block grid; maintaining, in the first 2D patch, only fully filled image blocks where each pixel has a projected value; removing the 3D object points having a corresponding 2D projection point in the fully filled image blocks of the first 2D patch from the 3D object; encoding the first 2D patch of the geometry image or texture image having only fully filled image blocks of the predetermined size into a bitstream of encoding video frames; and subjecting remaining 3D object points iteratively for one or more new projections into subsequent 2D patches.

[0009] According to an embodiment, the remaining 3D object points are subjected for a new projection until one of the following criteria is fulfilled: complete shape of the 3D object is projected, there are no more suitable patch projections, or a predefined stopping condition is met.

[0010] According to an embodiment, the method further comprises analysing and processing said arbitrary shaped 2D patches prior to packing into the respective 2D images.

[0011] According to an embodiment, the analysing and processing for each of said arbitrary shaped 2D patches further comprises checking whether the patch fills at least one predetermined image block fully; dividing the patch into image blocks of the predetermined size; checking, for each image block, if the image block is fully filled; removing any partially filled image block from the patch and returning the 2D projection points contained by said image block are to the 3D object; and packing the block-aligned 2D patch containing only fully filled blocks into a block-aligned 2D image for encoding.

[0012] According to an embodiment, the stopping condition is one the following:

the 3D object has been fully projected;

a percentage of remaining 3D points is below a certain threshold compared to original 3D points;

no further block-aligned patches are possible;

a threshold number of patches has been projected;

no space available in the projected 2D images;

time limit for the processing is met.

[0013] According to an embodiment, remaining non-projected 3D data is processed according to one of the following:

- the remaining 3D points are dropped from the encoded representation;

- the remaining 3D points are packed without block alignment in additional individual images with accompanying occupancy maps;

- the remaining 3D points are packed without block alignment in the same pictures as the block-aligned patches.

[0014] According to an embodiment, upon packing the remaining 3D points without block alignment in additional individual images with accompanying occupancy maps, the method further comprises decomposing the 3D object into a set of block-aligned patches and arbitrary-shaped patches; and using a reduced signalling of occupancy information only for the arbitrary-shaped patches.

[0015] According to an embodiment, upon packing the remaining 3D points without block alignment in the same pictures as the block-aligned patches, the method further comprises using block-aligned signaling for occupancy map compression, wherein the occupancy map comprises information about the 2D projection points to be reconstructed into 3D space at the decoder side.

[0016] According to an embodiment, the method further comprises signaling per-pixel occupancy information within a block as a quadtree.

[0017] An apparatus according to a second aspect comprises at least one processor and at least one memory, said at least one memory stored with code thereon, which when executed by said at least one processor, causes the apparatus to perform the above method.

[0018] A computer readable storage medium according to a third aspect comprises code for use by an apparatus, which when executed by a processor, causes the apparatus to perform the above method.

[0019] Further aspects relate at least to an apparatus and computer readable storage medium with computer program code comprising means for performing the above method and embodiments related thereto.

BRIEF DESCRIPTION OF THE DRAWINGS

[0020] For a more complete understanding of example embodiments of the present invention, reference is now made to the following descriptions taken in connection with the accompanying drawings in which:

[0021] Fig. 1 shows a system for capturing, encoding, decoding, reconstructing and viewing a three-dimensional scheme;

[0022] Figs. 2a and 2b show a capture device and a viewing device;

[0023] Figs. 3a and 3b show an encoder and decoder for encoding and decoding texture pictures, geometry pictures and/or auxiliary pictures;

[0024] Figs. 4a, 4b, 4c and 4d show a setup for forming a stereo image of a scene to a user;

[0025] Figs. 5a illustrates projection of source volumes in a scene and parts of an object to projection surfaces, as well as determining depth information;

[0026] Fig. 5b shows an example of projecting an object using a cube map projection format;

[0027] Fig. 6 shows a projection of a source volume to a projection surface, and inpainting of a sparse projection;

[0028] Fig. 7 shows an examples of occlusion of surfaces;

[0029] Fig. 8 shows a known test model for standardized validating of an intra point cloud frame encoding; [0030] Fig. 9 shows a flow chart for the intra point cloud frame coding according to an embodiment;

[0031] Figure 10 illustrates an example of a block-aligned patch trimming process, where an original patch and a block-aligned patch are shown in parallel;

[0032] Figure 11 shows a flow chart of the block-alignment process according to an embodiment;

[0033] Figure 12 shows a block chart of an encoder arranged to carry out reduced occupancy signalling according to an embodiment; and

[0034] Figure 13 shows an example of quadtree occupancy coding according to an embodiment.

DETAILED DESCRIPTON OF SOME EXAMPLE EMBODIMENTS

[0035] In the following, several embodiments of the invention will be described in the context of point cloud, voxel or mesh scene models for three-dimensional volumetric video and pixel and picture based two-dimensional video coding. It is to be noted, however, that the invention is not limited to specific scene models or specific coding technologies. In fact, the different embodiments have applications in any environment where coding of volumetric scene data is required.

[0036] It has been noticed here that identifying correspondences for motion-compensation in three-dimensional space is an ill-defined problem, as both the geometry and the respective attributes of the objects to be coded may change. For example, temporal successive“frames” do not necessarily have the same number of meshes, points or voxel. Therefore, compression of dynamic 3D scenes is inefficient.

[0037] “Voxel” of a three-dimensional world corresponds to a pixel of a two-dimensional world. Voxels exist in a three-dimensional grid layout. An octree is a tree data structure used to partition a three-dimensional space. Octrees are the three-dimensional analog of quadtrees. A sparse voxel octree (SVO) describes a volume of a space containing a set of solid voxels of varying sizes. Empty areas within the volume are absent from the tree, which is why it is called“sparse”.

[0038] A three-dimensional volumetric representation of a scene is determined as a plurality of voxels on the basis of input streams of at least one multicamera device. Thus, at least one but preferably a plurality (i.e. 2, 3, 4, 5 or more) of multicamera devices are used to capture 3D video representation of a scene. The multicamera devices are distributed in different locations in respect to the scene, and therefore each multicamera device captures a different 3D video representation of the scene. The 3D video representations captured by each multicamera device may be used as input streams for creating a 3D volumetric representation of the scene, said 3D volumetric representation comprising a plurality of voxels. Voxels may be formed from the captured 3D points e.g. by merging the 3D points into voxels comprising a plurality of 3D points such that for a selected 3D point, all neighbouring 3D points within a predefined threshold from the selected 3D point are merged into a voxel without exceeding a maximum number of 3D points in a voxel.

[0039] Voxels may also be formed through the construction of the sparse voxel octree. Each leaf of such a tree represents a solid voxel in world space; the root node of the tree represents the bounds of the world. The sparse voxel octree construction may have the following steps: 1) map each input depth map to a world space point cloud, where each pixel of the depth map is mapped to one or more 3D points; 2) determine voxel attributes such as colour and surface normal vector by examining the neighbourhood of the source pixel(s) in the camera images and the depth map; 3) determine the size of the voxel based on the depth value from the depth map and the resolution of the depth map; 4) determine the SVO level for the solid voxel as a function of its size relative to the world bounds; 5) determine the voxel coordinates on that level relative to the world bounds; 6) create new and/or traversing existing SVO nodes until arriving at the determined voxel coordinates; 7) insert the solid voxel as a leaf of the tree, possibly replacing or merging attributes from a previously existing voxel at those coordinates. Nevertheless, the size of voxel within the 3D volumetric representation of the scene may differ from each other. The voxels of the 3D volumetric representation thus represent the spatial locations within the scene.

[0040] A volumetric video frame is a complete sparse voxel octree that models the world at a specific point in time in a video sequence. Voxel attributes contain information like colour, opacity, surface normal vectors, and surface material properties. These are referenced in the sparse voxel octrees (e.g. colour of a solid voxel), but can also be stored separately.

[0041] Point clouds are commonly used data structures for storing volumetric content. Compared to point clouds, sparse voxel octrees describe a recursive subdivision of a finite volume with solid voxels of varying sizes, while point clouds describe an unorganized set of separate points limited only by the precision of the used coordinate values.

[0042] When encoding a volumetric video, each frame may produce several hundred megabytes or several gigabytes of voxel data which needs to be converted to a format that can be streamed to the viewer, and rendered in real-time. The amount of data depends on the world complexity and volume. The larger impact comes in a multi-device recording setup with a number of separate locations where the cameras are recording. Such a setup produces more information than a camera at a single location.

[0043] Fig. 1 shows a system for capturing, encoding, decoding, reconstructing and viewing a three-dimensional scheme, that is, for 3D video and 3D audio digital creation and playback. The task of the system is that of capturing sufficient visual and auditory

information from a specific scene to be able to create a scene model such that a convincing reproduction of the experience, or presence, of being in that location can be achieved by one or more viewers physically located in different locations and optionally at a time later in the future. Such reproduction requires more information that can be captured by a single camera or microphone, in order that a viewer can determine the distance and location of objects within the scene using their eyes and their ears. To create a pair of images with disparity, two camera sources are used. In a similar manner, for the human auditory system to be able to sense the direction of sound, at least two microphones are used (the commonly known stereo sound is created by recording two audio channels). The human auditory system can detect the cues, e.g. in timing difference of the audio signals to detect the direction of sound.

[0044] The system of Fig. 1 may consist of three main parts: image sources, a server and a rendering device. A video source SRC 1 may comprise multiple cameras CAM 1 , CAM2, ... , CAMN with overlapping field of view so that regions of the view around the video capture device is captured from at least two cameras. The video source SRC1 may comprise multiple microphones to capture the timing and phase differences of audio originating from different directions. The video source SRC1 may comprise a high-resolution orientation sensor so that the orientation (direction of view) of the plurality of cameras CAM1, CAM2, ..., CAMN can be detected and recorded. The cameras or the computers may also comprise or be functionally connected to means for forming distance information corresponding to the captured images, for example so that the pixels have corresponding depth data. Such depth data may be formed by scanning the depth or it may be computed from the different images captured by the cameras. The video source SRC1 comprises or is functionally connected to, or each of the plurality of cameras CAM1, CAM2, ..., CAMN comprises or is functionally connected to a computer processor and memory, the memory comprising computer program code for controlling the source and/or the plurality of cameras. The image stream captured by the video source, i.e. the plurality of the cameras, may be stored on a memory device for use in another device, e.g. a viewer, and/or transmitted to a server using a communication interface. It needs to be understood that although a video source comprising three cameras is described here as part of the system, another amount of camera devices may be used instead as part of the system.

[0045] Alternatively or in addition to the source device SRC1 creating information for forming a scene model, one or more sources SRC2 of synthetic imagery may be present in the system, comprising a scene model. Such sources may be used to create and transmit the scene model and its development over time, e.g. instantaneous states of the model. The model can be created or provided by the source SRC1 and/or SRC2, or by the server SERVER. Such sources may also use the model of the scene to compute various video bitstreams for transmission.

[0046] One or more two-dimensional video bitstreams may be computed at the server SERVER or a device RENDERER used for rendering, or another device at the receiving end. When such computed video streams are used for viewing, the viewer may see a three- dimensional virtual world as described in the context of Figs 4a— 4d. The devices SRC1 and SRC2 may comprise or be functionally connected to one or more computer processors (PROC2 shown) and memory (MEM2 shown), the memory comprising computer program (PROGR2 shown) code for controlling the source device SRC1/SRC2. The image stream captured by the device and the scene model may be stored on a memory device for use in another device, e.g. a viewer, or transmitted to a server or the viewer using a communication interface COMM2. There may be a storage, processing and data stream serving network in addition to the capture device SRC1. For example, there may be a server SERVER or a plurality of servers storing the output from the capture device SRC1 or device SRC2 and/or to form a scene model from the data from devices SRC1, SRC2. The device SERVER comprises or is functionally connected to a computer processor PROC3 and memory MEM3, the memory comprising computer program PROGR3 code for controlling the server. The device SERVER may be connected by a wired or wireless network connection, or both, to sources SRC1 and/or SRC2, as well as the viewer devices VIEWER1 and VIEWER2 over the communication interface COMM3.

[0047] The creation of a three-dimensional scene model may take place at the server SERVER or another device by using the images captured by the devices SRC1. The scene model may be a model created from captured image data (a real-world model), or a synthetic model such as on device SRC2, or a combination of such. As described later, the scene model may be encoded to reduce its size and transmitted to a decoder, for example viewer devices.

[0048] For viewing the captured or created video content, there may be one or more viewer devices VIEWER 1 and VIEWER2. These devices may have a rendering module and a display module, or these functionalities may be combined in a single device. The devices may comprise or be functionally connected to a computer processor PROC4 and memory MEM4, the memory comprising computer program PROG4 code for controlling the viewing devices. The viewer (playback) devices may consist of a data stream receiver for receiving a video data stream and for decoding the video data stream. The video data stream may be received from the server SERVER or from some other entity, such as a proxy server, an edge server of a content delivery network, or a file available locally in the viewer device. The data stream may be received over a network connection through communications interface COMM4, or from a memory device MEM6 like a memory card CARD2. The viewer devices may have a graphics processing unit for processing of the data to a suitable format for viewing. The viewer VIEWER1 may comprise a high-resolution stereo-image head-mounted display for viewing the rendered stereo video sequence. The head-mounted display may have an orientation sensor DET1 and stereo audio headphones. The viewer VIEWER2 may comprise a display (either two-dimensional or a display enabled with 3D technology for displaying stereo video), and the rendering device may have an orientation detector DET2 connected to it. Alternatively, the viewer VIEWER2 may comprise a 2D display, since the volumetric video rendering can be done in 2D by rendering the viewpoint from a single eye instead of a stereo eye pair.

[0049] It needs to be understood that Fig. 1 depicts one SRC1 device and one SRC2 device, but generally the system may comprise more than one SRC1 device and/or SRC2 device.

[0050] Any of the devices (SRC 1 , SRC2, SERVER, RENDERER, VIEWER 1 , VIEWER2) may be a computer or a portable computing device, or be connected to such or configured to be connected to such. Moreover, even if the devices (SRC1, SRC2, SERVER, RENDERER, VIEWER1, VIEWER2) are depicted as a single device in Fig. 1, they may comprise multiple parts or may be comprised of multiple connected devices. For example, it needs to be understood that SERVER may comprise several devices, some of which may be used for editing the content produced by SRC1 and/or SRC2 devices, some others for compressing the edited content, and a third set of devices may be used for transmitting the compressed content. Such devices may have computer program code for carrying out methods according to various examples described in this text.

[0051] Figs. 2a and 2b show a capture device and a viewing device, respectively. Fig. 2a illustrates a camera CAM1. The camera has a camera detector CAMDET1, comprising a plurality of sensor elements for sensing intensity of the light hitting the sensor element. The camera has a lens OBJ1 (or a lens arrangement of a plurality of lenses), the lens being positioned so that the light hitting the sensor elements travels through the lens to the sensor elements. The camera detector CAMDET1 has a nominal centre point CP1 that is a middle point of the plurality of sensor elements, for example for a rectangular sensor the crossing point of diagonals of the rectangular sensor. The lens has a nominal centre point PP1, as well, lying for example on the axis of symmetry of the lens. The direction of orientation of the camera is defined by the line passing through the centre point CP1 of the camera sensor and the centre point PP1 of the lens. The direction of the camera is a vector along this line pointing in the direction from the camera sensor to the lens. The optical axis of the camera is understood to be this line CP1-PP1. However, the optical path from the lens to the camera detector need not always be a straight line but there may be mirrors and/or some other elements which may affect the optical path between the lens and the camera detector.

[0052] Fig. 2b shows a head-mounted display (HMD) for stereo viewing. The head- mounted display comprises two screen sections or two screens DISP1 and DISP2 for displaying the left and right eye images. The displays are close to the eyes, and therefore lenses are used to make the images easily viewable and for spreading the images to cover as much as possible of the eyes' field of view. When the device will be used by a user, the user may put the device on her/his head so that it will be attached to the head of the user so that it stays in place even when the user turns his head. The device may have an orientation detecting module ORDET1 for determining the head movements and direction of the head. The head-mounted display gives a three-dimensional (3D) perception of the

recorded/streamed content to a user.

[0053] The system described above may function as follows. Time-synchronized video and orientation data is first recorded with the capture devices. This can consist of multiple concurrent video streams as described above. One or more time-synchronized audio streams may also be recorded with the capture devices. The different capture devices may form image and geometry information of the scene from different directions. For example, there may be three, four, five, six or more cameras capturing the scene from different sides, like front, back, left and right, and/or at directions between these, as well as from the top or bottom, or any combination of these. The cameras may be at different distances, for example some of the cameras may capture the whole scene and some of the cameras may be capturing one or more objects in the scene. In an arrangement used for capturing volumetric video data, several cameras may be directed towards an object, looking onto the object from different directions, where the object is e.g. in the middle of the cameras. In this manner, the texture and geometry of the scene and the objects within the scene may be captured adequately. As mentioned earlier, the cameras or the system may comprise means for determining geometry

information, e.g. depth data, related to the captured video streams. From these concurrent video and audio streams, a computer model of a scene may be created. Alternatively or additionally, a synthetic computer model of a virtual scene may be used. The models (at successive time instances) are then transmitted immediately or later to the storage and processing network for processing and conversion into a format suitable for subsequent delivery to playback devices. The conversion may involve processing and coding to improve the quality and/or reduce the quantity of the scene model data while preserving the quality at a desired level. Each playback device receives a stream of the data (either computed video data or scene model data) from the network, and renders it into a viewing reproduction of the original location which can be experienced by a user. The reproduction may be two- dimensional or three-dimensional (stereo image pairs).

[0054] Figs. 3a and 3b show an encoder and decoder for encoding and decoding texture pictures, geometry pictures and/or auxiliary pictures. A video codec consists of an encoder that transforms an input video into a compressed representation suited for

storage/transmission and a decoder that can uncompress the compressed video representation back into a viewable form. Typically, the encoder discards and/or loses some information in the original video sequence in order to represent the video in a more compact form (that is, at lower bitrate). An example of an encoding process is illustrated in Figure 3a. Figure 3a illustrates an image to be encoded (P); a predicted representation of an image block (P'ⁿ); a prediction error signal (Dⁿ); a reconstructed prediction error signal (D'ⁿ); a preliminary reconstructed image (I'ⁿ); a final reconstructed image (R'ⁿ); a transform (T) and inverse transform (T ¹); a quantization (Q) and inverse quantization (Q ¹); entropy encoding (E); a reference frame memory (RFM); inter prediction (Pi_nter); intra prediction (Pi_ntra); mode selection (MS) and filtering (F).

[0055] An example of a decoding process is illustrated in Figure 3b. Figure 3b illustrates a predicted representation of an image block (P'ⁿ); a reconstructed prediction error signal (D'ⁿ); a preliminary reconstructed image (I'ⁿ); a final reconstructed image (R'ⁿ); an inverse transform (T ¹); an inverse quantization (Q ¹); an entropy decoding (E ¹); a reference frame memory (RFM); a prediction (either inter or intra) (P); and filtering (F).

[0056] Many hybrid video encoders encode the video information in two phases. Firstly pixel values in a certain picture area (or“block”) are predicted for example by motion compensation means (finding and indicating an area in one of the previously coded video frames that corresponds closely to the block being coded) or by spatial means (using the pixel values around the block to be coded in a specified manner). Secondly the prediction error, i.e. the difference between the predicted block of pixels and the original block of pixels, is coded. This is typically done by transforming the difference in pixel values using a specified transform (e.g. Discrete Cosine Transform (DCT) or a variant of it), quantizing the coefficients and entropy coding the quantized coefficients. By varying the fidelity of the quantization process, encoder can control the balance between the accuracy of the pixel representation (picture quality) and size of the resulting coded video representation (file size or transmission bitrate).

[0057] Many video encoders partition a picture into blocks along a block grid. For example, in the High Efficiency Video Coding (HEVC) standard, the following partitioning and definitions are used. A coding block may be defined as an NxN block of samples for some value of N such that the division of a coding tree block into coding blocks is a partitioning. A coding tree block (CTB) may be defined as an NxN block of samples for some value of N such that the division of a component into coding tree blocks is a partitioning. A coding tree unit (CTU) may be defined as a coding tree block of luma samples, two corresponding coding tree blocks of chroma samples of a picture that has three sample arrays, or a coding tree block of samples of a monochrome picture or a picture that is coded using three separate color planes and syntax structures used to code the samples. A coding unit (CU) may be defined as a coding block of luma samples, two corresponding coding blocks of chroma samples of a picture that has three sample arrays, or a coding block of samples of a monochrome picture or a picture that is coded using three separate color planes and syntax structures used to code the samples. A CU with the maximum allowed size may be named as LCU (largest coding unit) or coding tree unit (CTU) and the video picture is divided into non overlapping LCUs.

[0058] Entropy coding/decoding may be performed in many ways. For example, context- based coding/decoding may be applied, where in both the encoder and the decoder modify the context state of a coding parameter based on previously coded/decoded coding parameters. Context-based coding may for example be context adaptive binary arithmetic coding

(CABAC) or context-based variable length coding (CAVLC) or any similar entropy coding. Entropy coding/decoding may alternatively or additionally be performed using a variable length coding scheme, such as Huffman coding/decoding or Exp-Golomb coding/decoding. Decoding of coding parameters from an entropy-coded bitstream or codewords may be referred to as parsing. [0059] Figs. 4a, 4b, 4c and 4d show a setup for forming a stereo image of a scene to a user, for example a video frame of a 3D video. In Fig. 4a, a situation is shown where a human being is viewing two spheres Al and A2 using both eyes El and E2. The sphere Al is closer to the viewer than the sphere A2, the respective distances to the first eye El being L_{EI ,AI} and L_{EI ,A2}. The different objects reside in space at their respective (x,y,z) coordinates, defined by the coordinate system SZ, SY and SZ. The distance di₂ between the eyes of a human being may be approximately 62-64 mm on average, and varying from person to person between 55 and 74 mm. This distance is referred to as the parallax, on which stereoscopic view of the human vision is based on. The viewing directions (optical axes) DIR1 and DIR2 are typically essentially parallel, possibly having a small deviation from being parallel, and define the field of view for the eyes. The head of the user has an orientation (head orientation) in relation to the surroundings, most easily defined by the common direction of the eyes when the eyes are looking straight ahead. That is, the head orientation tells the yaw, pitch and roll of the head in respect of a coordinate system of the scene where the user is.

[0060] When the viewer's body (thorax) is not moving, the viewer's head orientation is restricted by the normal anatomical ranges of movement of the cervical spine.

[0061] In the setup of Fig. 4a, the spheres Al and A2 are in the field of view of both eyes. The centre-point O₁₂ between the eyes and the spheres are on the same line. That is, from the centre-point, the sphere A2 is behind the sphere Al. However, each eye sees part of sphere A2 from behind Al, because the spheres are not on the same line of view from either of the eyes.

[0062] In Fig. 4b, there is a setup shown, where the eyes have been replaced by cameras Cl and C2, positioned at the location where the eyes were in Fig. 4a. The distances and directions of the setup are otherwise the same. Naturally, the purpose of the setup of Fig. 4b is to be able to take a stereo image of the spheres Al and A2. The two images resulting from image capture are Fci and Fc₂. The "left eye” image Fci shows the image SA2 of the sphere A2 partly visible on the left side of the image S_AI of the sphere Al . The "right eye" image Fc₂ shows the image S_A2 of the sphere A2 partly visible on the right side of the image S_AI of the sphere Al. This difference between the right and left images is called disparity, and this disparity, being the basic mechanism with which the HVS determines depth information and creates a 3D view of the scene, can be used to create an illusion of a 3D image.

[0063] In this setup of Fig. 4b, where the inter-eye distances correspond to those of the eyes in Fig. 4a, the camera pair Cl and C2 has a natural parallax, that is, it has the property of creating natural disparity in the two images of the cameras. Natural disparity may be understood to be created even though the distance between the two cameras forming the stereo camera pair is somewhat smaller or larger than the normal distance (parallax) between the human eyes, e.g. essentially between 40 mm and 100 mm or even 30 mm and 120 mm.

[0064] It needs to be understood here that the images Fci and Fc₂ may be captured by cameras Cl and C2, where the cameras Cl and C2 may be real-world cameras or they may be virtual cameras. In the case of virtual cameras, the images Fci and Fc₂ may be computed from a computer model of a scene by setting the direction, orientation and viewport of the cameras Cl and C2 appropriately such that a stereo image pair suitable for viewing by the human visual system (HVS) is created.

[0065] In Fig. 4c, the creating of this 3D illusion is shown. The images Fci and Fc₂ captured or computed by the cameras Cl and C2 are displayed to the eyes El and E2, using displays Dl and D2, respectively. The disparity between the images is processed by the human visual system so that an understanding of depth is created. That is, when the left eye sees the image S_A2 of the sphere A2 on the left side of the image S_AI of sphere Al, and respectively the right eye sees the image S_A2 of the sphere A2 on the right side, the human visual system creates an understanding that there is a sphere V2 behind the sphere VI in a three-dimensional world. Here, it needs to be understood that the images Fci and Fc₂ can also be synthetic, that is, created by a computer. If they carry the disparity information, synthetic images will also be seen as three-dimensional by the human visual system. That is, a pair of computer-generated images can be formed so that they can be used as a stereo image.

[0066] Fig. 4d illustrates how the principle of displaying stereo images to the eyes can be used to create 3D movies or virtual reality scenes having an illusion of being three- dimensional. The images Fxi and Fx2 are either captured with a stereo camera or computed from a model so that the images have the appropriate disparity. By displaying a large number (e.g. 30) frames per second to both eyes using display Dl and D2 so that the images between the left and the right eye have disparity, the human visual system will create a cognition of a moving, three-dimensional image.

[0067] The field of view represented by the content may be greater than the displayed field of view e.g. in an arrangement depicted in Fig. 4d. Consequently, only a part of the content along the direction of view (a.k.a. viewing orientation) is displayed at a single time. This direction of view, that is, the head orientation, may be determined as a real orientation of the head e.g. by an orientation detector mounted on the head, or as a virtual orientation determined by a control device such as a joystick or mouse that can be used to manipulate the direction of view without the user actually moving his head. That is, the term "head orientation" may be used to refer to the actual, physical orientation of the user's head and changes in the same, or it may be used to refer to the virtual direction of the user's view that is determined by a computer program or a computer input device.

[0068] The content may enable viewing from several viewing positions within the 3D space. The texture picture(s), the geometry picture(s) and the geometry information may be used to synthesize the images Fxi and/or Fx₂ as if the displayed content was captured by camera(s) located at the viewing position.

[0069] The principle illustrated in Figs. 4a-4d may be used to create three-dimensional images to a viewer from a three-dimensional scene model (volumetric video) after the scene model has been encoded at the sender and decoded and reconstructed at the receiver. Because volumetric video describes a 3D scene or object at different (successive) time instances, such data can be viewed from any viewpoint. Therefore, volumetric video is an important format for any augmented reality, virtual reality and mixed reality applications, especially for providing viewing capabilities having six degrees of freedom (so-called 6DOF viewing).

[0070] Fig. 5a illustrates projection of source volumes in a digital scene model SCE and parts of an object model OBJ1, OBJ2, OBJ3, BG4 to projection surfaces Sl, S2, S3, S4, as well as determining depth information for the purpose of encoding volumetric video.

[0071] The projection of source volumes SV1, SV2, SV3, SV4 may result in texture pictures and geometry pictures, and there may be geometry information related to the projection source volumes and/or projection surfaces. Texture pictures, geometry pictures and projection geometry information may be encoded into a bitstream. A texture picture may comprise information on the colour data of the source of the projection. Through the projection, such colour data may result in pixel colour information in the texture picture. Pixels may be coded in groups, e.g. coding units of rectangular shape. The projection geometry information may comprise but is not limited to one or more of the following:

- projection type, such as planar projection or equirectangular projection

- projection surface type, such as a cube, sphere, cylinder, polyhedron

- location of the projection surface in 3D space

- orientation of the projection surface in 3D space

- size of the projection surface in 3D space

- type of a projection centre, such as a projection centre point, axis, or plane, from which a geometry primitive is projected onto the projection surface

- location and/or orientation of a projection centre. [0072] The projection may take place by projecting the geometry primitives (points of a point could, triangles of a triangle mesh or voxels of a voxel array) of a source volume SV1, SV2, SV3, SV4 (or an object OBJ1, OBJ2, OBJ3, BG4) onto a projection surface Sl, S2, S3, S4. The geometry primitives may comprise information on the texture, for example a colour value or values of a point, a triangle or a voxel. The projection surface may surround the source volume at least partially such that projection of the geometry primitives happens from the centre of the projection surface outwards to the surface. For example, a cylindrical surface has a projection centre axis and a spherical surface has a projection centre point. A cubical or rectangular surface may have projection centre planes or a projection centre axis or point and the projection of the geometry primitives may take place either orthogonally to the sides of the surface or from the projection centre axis or point outwards to the surface. The projection surfaces, e.g. cylindrical and rectangular, may be open from the top and the bottom such that when the surface is cut and rolled out on a two-dimensional plane, it forms a rectangular shape. Such rectangular shape with pixel data can be encoded and decoded with a video codec.

[0073] Alternatively or in addition, the projection surface such as a planar surface or a sphere may be inside group of geometry primitives, e.g. inside a point cloud that defines a surface. In the case of an inside projection surface, the projection may take place from outside in towards the centre and may result in sub-sampling of the texture data of the source.

[0074] In a point cloud based scene model or object model, points may be represented with any floating point coordinates. A quantized point cloud may be used to reduce the amount of data, whereby the coordinate values of the point cloud are represented e.g. with lO-bit, l2-bit or 16-bit integers. Integers may be used because hardware accelerators may be able to operate on integers more efficiently. The points in the point cloud may have associated colour, reflectance, opacity etc. texture values. The points in the point cloud may also have a size, or a size may be the same for all points. The size of the points may be understood as indicating how large an object the point appears to be in the model in the projection. The point cloud is projected by ray casting from the projection surface to find out the pixel values of the projection surface. In such a manner, the topmost point remains visible in the projection, while points closer to the centre of the projection surface may be occluded. In other words, in general, the original point cloud, meshes, voxels, or any other model is projected outwards to a simple geometrical shape, this simple geometrical shape being the projection surface.

[0075] Different projection surfaces may have different characteristics in terms of projection and reconstruction. In the sense of computational complexity, a projection to a cubical surface may be the most efficient, and a cylindrical projection surface may provide accurate results efficiently. Also cones, polyhedron-based parallelepipeds (hexagonal or octagonal, for example) and spheres or a simple plane may be used as projection surfaces.

[0076] The phrase along the bitstream (e.g. indicating along the bitstream) may be defined to refer to out-of-band transmission, signalling, or storage in a manner that the out-of-band data is associated with the bitstream. The phrase decoding along the bitstream or alike may refer to decoding the referred out-of-band data (which may be obtained from out-of-band transmission, signalling, or storage) that is associated with the bitstream. For example, an indication along the bitstream may refer to metadata in a container file that encapsulates the bitstream.

[0077] As illustrated in Fig. 5a, a first texture picture may be encoded into a bitstream, and the first texture picture may comprise a first projection of texture data of a first source volume SV1 of a scene model SCE onto a first projection surface Sl. The scene model SCE may comprise a number of further source volumes SV2, SV3, SV4.

[0078] In the projection, data on the position of the originating geometry primitive may also be determined, and based on this determination, a geometry picture may be formed. This may happen for example so that depth data is determined for each or some of the texture pixels of the texture picture. Depth data is formed such that the distance from the originating geometry primitive such as a point to the projection surface is determined for the pixels. Such depth data may be represented as a depth picture, and similarly to the texture picture, such geometry picture (in this example, depth picture) may be encoded and decoded with a video codec. This first geometry picture may be seen to represent a mapping of the first projection surface to the first source volume, and the decoder may use this information to determine the location of geometry primitives in the model to be reconstructed. In order to determine the position of the first source volume and/or the first projection surface and/or the first projection in the scene model, there may be first geometry information encoded into or along the bitstream.

[0079] A picture may be defined to be either a frame or a field. A frame may be defined to comprise a matrix of luma samples and possibly the corresponding chroma samples. A field may be defined to be a set of alternate sample rows of a frame. Fields may be used as encoder input for example when the source signal is interlaced. Chroma sample arrays may be absent (and hence monochrome sampling may be in use) or may be subsampled when compared to luma sample arrays. Some chroma formats may be summarized as follows: - In monochrome sampling there is only one sample array, which may be nominally considered the luma array.

- In 4:2:0 sampling, each of the two chroma arrays has half the height and half the width of the luma array.

- In 4:2:2 sampling, each of the two chroma arrays has the same height and half the width of the luma array.

- In 4:4:4 sampling when no separate colour planes are in use, each of the two chroma

arrays has the same height and width as the luma array.

[0080] It is possible to code sample arrays as separate colour planes into the bitstream and respectively decode separately coded colour planes from the bitstream. When separate colour planes are in use, each one of them is separately processed (by the encoder and/or the decoder) as a picture with monochrome sampling.

[0081 ] An attribute picture may be defined as a picture that comprises additional information related to an associated texture picture. An attribute picture may for example comprise surface normal, opacity, or reflectance information for a texture picture. A geometry picture may be regarded as one type of an attribute picture, although a geometry picture may be treated as its own picture type, separate from an attribute picture.

[0082] Texture picture(s) and the respective geometry picture(s), if any, and the respective attribute picture(s) may have the same or different chroma format.

[0083] Terms texture image and texture picture may be used interchangeably. Terms geometry image and geometry picture may be used interchangeably. A specific type of a geometry image is a depth image. Embodiments described in relation to a geometry image equally apply to a depth image, and embodiments described in relation to a depth image equally apply to a geometry image. Terms attribute image and attribute picture may be used interchangeably. A geometry picture and/or an attribute picture may be treated as an auxiliary picture in video/image encoding and/or decoding.

[0084] Depending on the context, a pixel may be defined to a be a sample of one of the sample arrays of the picture or may be defined to comprise the collocated samples of all the sample arrays of the picture.

[0085] Multiple source volumes (objects) may be encoded as texture pictures, geometry pictures and projection geometry information into the bitstream in a similar manner. That is, as in Fig. 5a, the scene model SCE may comprise multiple objects OBJ1, OBJ2, OBJ3, OBJ4, and these may be treated as source volumes SV1, SV2, SV3, SV4 and each object may be coded as a texture picture, geometry picture and projection geometry information. [0086] In the above, the first texture picture of the first source volume SV1 and further texture pictures of the other source volumes SV2, SV3, SV4 may represent the same time instance. That is, there may be a plurality of texture and geometry pictures and projection geometry information for one time instance, and the other time instances may be coded in a similar manner. Since the various source volumes are in this way producing sequences of texture pictures and sequences of geometry pictures, as well as sequences of projection geometry information, the inter-picture redundancy in the picture sequences can be used to encode the texture and geometry data for the source volumes efficiently, compared to the presently known ways of encoding volume data.

[0087] An object OBJ3 (source volume SV3) may be projected onto a projection surface S3 and encoded into the bitstream as a texture picture, geometry picture and projection geometry information as described above. Furthermore, such source volume may be indicated to be static by encoding information into said bitstream on said fourth projection geometry being static. A static source volume or object may be understood to be an object whose position with respect to the scene model remains the same over two or more or all time instances of the video sequence. For such static source volume, the geometry data (geometry pictures) may also stay the same, that is, the object's shape remains the same over two or more time instances. For such static source volume, some or all of the texture data (texture pictures) may stay the same over two or more time instances. By encoding information into the bitstream of the static nature of the source volume the encoding efficiency may further be improved, as the same information may not need to be coded multiple times. In this manner, the decoder will also be able to use the same reconstruction or partially same reconstruction of the source volume (object) over multiple time instances.

[0088] In an analogous manner, the different source volumes may be coded into the bitstream with different frame rates. For example, a slow-moving or relatively unchanging object (source volume) may be encoded with a first frame rate, and a fast-moving and/or changing object (source volume) may be coded with a second frame rate. The first frame rate may be slower than the second frame rate, for example one half or one quarter of the second frame rate, or even slower. For example, if the second frame rate is 30 frames per second, the second frame rate may be 15 frames per second, or 1 frame per second. The first and second object (source volumes) may be "sampled" in synchrony such that some frames of the faster frame rate coincide with frames of the slower frame rate.

[0089] There may be one or more coordinate systems in the scene model. The scene model may have a coordinate system and one or more of the objects (source volumes) in the scene model may have their local coordinate systems. The shape, size, location and orientation of one or more projection surfaces may be encoded into or along the bitstream with respect to the scene model coordinates. Alternatively or in addition, the encoding may be done with respect to coordinates of the scene model or said first source volume. The choice of coordinate systems may improve the coding efficiency.

[0090] Information on temporal changes in location, orientation and size of one or more said projection surfaces may be encoded into or along the bitstream. For example, if one or more of the objects (source volumes) being encoded is moving or rotating with respect to the scene model, the projection surface moves or rotates with the object to preserve the projection as similar as possible.

[0091] If the projection volumes are changing, for example splitting or bending into two parts, the projection surfaces may be sub-divided respectively. Therefore, information on sub- division of one or more of the source volumes and respective changes in one or more of the projection surfaces may be encoded into or along the bitstream.

[0092] The resulting bitstream may then be output to be stored or transmitted for later decoding and reconstruction of the scene model.

[0093] Decoding of the information from the bitstream may happen in analogous manner.

A first texture picture may be decoded from a bitstream to obtain first decoded texture data, where the first texture picture comprises a first projection of texture data of a first source volume of the scene model to be reconstructed onto a first projection surface. The scene model may comprise a number of further source volumes. Then, a first geometry picture may be decoded from the bitstream to obtain first decoded scene model geometry data. The first geometry picture may represent a mapping of the first projection surface to the first source volume. First projection geometry information of the first projection may be decoded from the bitstream, the first projection geometry information comprising information of position of the first projection surface in the scene model. Using this information, a reconstructed scene model may be formed by projecting the first decoded texture data to a first destination volume using the first decoded scene model geometry data and said first projection geometry information to determine where the decoded texture information is to be placed in the scene model.

[0094] A 3D scene model may be classified into two parts: first all dynamic parts, and second all static parts. The dynamic part of the 3D scene model may further be sub-divided into separate parts, each representing objects (or parts of) an object in the scene model, that is, source volumes. The static parts of the scene model may include e.g. static room geometry (walls, ceiling, fixed furniture) and may be compressed either by known volumetric data compression solutions, or, similar to the dynamic part, sub-divided into individual objects for projection-based compression as described earlier, to be encoded into the bitstream.

[0095] In an example, some objects may be a chair (static), a television screen (static geometry, dynamic texture), a moving person (dynamic). For each object, a suitable projection geometry (surface) may be found, e.g. cube projection to represent the chair, another cube for the screen, a cylinder for the person's torso, a sphere for a detailed representation of the person's head, and so on. The 3D data of each object may then be projected onto the respective projection surface and 2D planes are derived by "unfolding" the projections from three dimensions to two dimensions (plane). The unfolded planes will have several channels, typically three for the colour representation of the texture, e.g. RGB, YUV, and one additional plane for the geometry (depth) of each projected point for later

reconstruction.

[0096] Frame packing may be defined to comprise arranging more than one input picture, which may be referred to as (input) constituent frames, into an output picture. In general, frame packing is not limited to any particular type of constituent frames or the constituent frames need not have a particular relation with each other. In many cases, frame packing is used for arranging constituent frames of a stereoscopic video clip into a single picture sequence. The arranging may include placing the input pictures in spatially non-overlapping areas within the output picture. For example, in a side-by-side arrangement, two input pictures are placed within an output picture horizontally adjacently to each other. The arranging may also include partitioning of one or more input pictures into two or more constituent frame partitions and placing the constituent frame partitions in spatially non-overlapping areas within the output picture. The output picture or a sequence of frame-packed output pictures may be encoded into a bitstream e.g. by a video encoder. The bitstream may be decoded e.g. by a video decoder. The decoder or a post-processing operation after decoding may extract the decoded constituent frames from the decoded picture(s) e.g. for displaying.

[0097] A standard 2D video encoder may then receive the planes as inputs, either as individual layers per object, or as a frame-packed representation of all objects. The texture picture may thus comprise a plurality of projections of texture data from further source volumes and the geometry picture may represent a plurality of mappings of projection surfaces to the source volume.

[0098] For each object, additional information may be signalled to allow for reconstruction at the decoder side: - in the case of a frame-packed representation: separation boundaries may be signalled to recreate the individual planes for each object,

- in the case of projection-based compression of static content: classification of each object as static/dynamic may be signalled,

- relevant data to create real-world geometry data from the decoded (quantised) geometry channel(s), e.g. quantisation method, depth ranges, bit depth, etc. may be signalled,

- initial state of each object: geometry shape, location, orientation, size may be signalled,

- temporal changes for each object, either as changes to the initial state on a per-picture level, or as a function of time may be signalled, and

- nature of any additional auxiliary data may be signalled.

[0099] The decoder may receive the static 3D scene model data together with the video bitstreams representing the dynamic parts of the scene model. Based on the signalled information on the projection geometries, each object may be reconstructed in 3D space and the decoded scene model is created by fusing all reconstructed parts (objects or source volumes) together.

[0100] Standard video encoding hardware may be utilized for real-time

compression/decompression of the projection surfaces that have been unfolded onto planes.

[0101] Single proj ection surfaces might suffice for the proj ection of very simple obj ects.

Complex objects or larger scenes may require several (different) projections. The relative geometry of the object/scene may remain constant over a volumetric video sequence, but the location and orientation of the projection surfaces in space can change (and can be possibly predicted in the encoding, wherein the difference from the prediction is encoded).

[0102] Depth may be coded "outside-in" (indicating the distance from the projection surface to the 3D point), or "inside-out" (indicating the distance from the 3D point to the projection surface). In inside-out coding, depth of each projected point may be positive (with positive distance PD1) or negative (with negative distance). Fig. 5b shows an example of projecting an object OBJ1 using a cube map projection format, wherein there are six projection surfaces PS 1 , ... ,PS6 of the proj ection cube PC 1. In this example, the proj ection surfaces are one on the left side PS1, one in front PS2, one on the right side PS3, one in the back PS4, one in the bottom PS5, and one in the top PS6 of the cube PC1 in the setup of Figure 5b. For clarity, only four of the projection surfaces will be shown and used in the rest of the specification. For example, in Figure 8a the projection surfaces on the left PS1, on the right PS3, in the front PS2 and at in the back PS4 are shown. It is, however, clear to a skilled person to utilize similar principles on all six projection surfaces when the cube map projection format is used.

[0103] Fig. 6 shows a projection of a source volume to a projection surface, and inpainting of a sparse projection. A three-dimensional (3D) scene model, represented as objects OBJ1 comprising geometry primitives such as mesh elements, points, and/or voxel, may be projected onto one, or more, projection surfaces, as described earlier. As shown in Fig 6, these projection surface geometries may be "unfolded" onto 2D planes (two planes per projected source volume: one for texture TP1, one for depth GP1), which may then be encoded using standard 2D video compression technologies. Relevant projection geometry information may be transmitted alongside the encoded video files to the decoder. The decoder may then decode the video and performs the inverse projection to regenerate the 3D scene model object ROBJ1 in any desired representation format, which may be different from the starting format e.g. reconstructing a point cloud from original mesh model data.

[0104] In addition to the texture picture and geometry picture shown in Fig. 6, one or more auxiliary pictures related to one or more said texture pictures and the pixels thereof may be encoded into or along with the bitstream. The auxiliary pictures may e.g. represent texture surface properties related to one or more of the source volumes. Such texture surface properties may be e.g. surface normal information (e.g. with respect to the projection direction), reflectance and opacity (e.g. an alpha channel value). An encoder may encode, in or along with the bitstream, indication(s) of the type(s) of texture surface properties represented by the auxiliary pictures, and a decoder may decode, from or along the bitstream, indication(s) of the type(s) of texture surface properties represented by the auxiliary pictures.

[0105] Mechanisms to represent an auxiliary picture may include but are not limited to the following:

A colour component sample array, such as a chroma sample array, of the geometry picture.

An additional sample array in addition to the conventional three colour component sample arrays of the texture picture or the geometry picture.

A constituent frame of a frame-packed picture that may also comprise texture picture(s) and/or geometry picture(s).

An auxiliary picture included in specific data units in the bitstream. For example, the Advanced Video Coding (H.264/AVC) standard specifies a network abstraction layer (NAL) unit for a coded slice of an auxiliary coded picture without partitioning. An auxiliary picture layer within a layered bitstream. For example, the High Efficiency Video Coding (HE VC) standard comprises the feature of including auxiliary picture layers in the bitstream. An auxiliary picture layer comprises auxiliary pictures.

An auxiliary picture bitstream separate from the bitstream(s) for the texture picture(s) and geometry picture(s). The auxiliary picture bitstream may be indicated, for example in a container file, to be associated with the bitstream(s) for the texture pictures(s) and geometry picture(s).

[0106] The mechanism(s) to be used for auxiliary pictures may be pre-defined e.g. in a coding standard, or the mechanism(s) may be selected e.g. by an encoder and indicated in or along the bitstream. The decoder may decode the mechanism(s) used for auxiliary pictures from or along the bitstream.

[0107] The projection surface of a source volume may encompass the source volume, and there may be a model of an object in that source volume. Encompassing may be understood so that the object (model) is inside the surface such that when looking from the centre axis or centre point of the surface, the object's points are closer to the centre than the points of the projection surface are. The model may be made of geometry primitives, as described. The geometry primitives of the model may be projected onto the projection surface to obtain projected pixels of the texture picture. This projection may happen from inside-out.

Alternatively or in addition, the projection may happen from outside-in.

Projecting 3D data onto 2D planes is independent from the 3D scene model representation format. There exist several approaches for projecting 3D data onto 2D planes, with the respective signalling. For example, there exist several mappings from spherical coordinates to planar coordinates, known from map projections of the globe, and the type and parameters of such projection may be signalled. For cylindrical projections, the aspect ratio of height and width may be signalled.

[0108] It may happen that when the projection of the object is performed on the projection surfaces PS1— PS6, some parts of the object OBJ1 or another object may occlude some other parts of the object OBJ1 which otherwise were visible from the projection surface in question. Hence, some parts of the object OBJ1 would not be projected to any of the surfaces of the projection format.

[0109] Figure 7 illustrates an example of this kind of situation. In this example the person’s left hand occludes a part of the body of the person so that when viewed (projected) from the left hand’s side the occluded part of the body would not be projected. In the same way, the planar object on the person’s right hand occludes some parts of the person’s stomach when viewed from the front of the person.

[0110] According to an approach, which has been proposed for occlusion handling for projection-based volumetric video coding, the 3D volume surface is analysed with respect to the target projection surface before performing the 3D-to-2D projection. Therein, an entity that maps 3D texture data on to projection planes can choose the six sides of an oriented or an axis aligned bounding box of a 3D point cloud as the initial set of projection planes. The mapping of 3D surface parts on to the projection planes only maps the closest coherent surface onto the projections planes. For example, if there are two surfaces of the 3D object where one surface occludes the other surface in the direction of the 2D planes normal, then only the occluding surface is mapped on to the projection plane. The occluded surface requires the generation of another projection plane for mapping. The pose of the projections planes for the occluded points in the point cloud can be chosen such that it maximizes the rate-distortion performance for encoding the texture, depth and other auxiliary planes.

[01 11] For the MPEG standard, there has been developed a test model for point cloud compression. MPEG W17248 discloses another projection-based approach for a test model for standardisation of dynamic point cloud compression. In intra coding of this test model, projected texture and geometry data is accompanied with an additional binary“occupancy map”, which may also referred to as“inpainting mask”, to establish which projected points are valid or not. Therein, texture and depth images are blurred/inpainted to increase coding efficiency. The occupancy map is used to signal if a 2D pixel should be reconstructed to 3D space at the decoder side. Fig. 8 illustrates the intra coding approach presented in MPEG W17248. For a comprehensive description of the test model, a reference is made to MPEG W17248.

[01 12] However, transmitting the additional occupancy map is costly in terms of the bit rate budget. W17248 uses a four- frame IPPP coding structure (intra frame followed by three inter frames), and the additional occupancy map is only transmitted in every fourth frame (I- frame). Thus, 3D resampling and additional motion information, possibly including padding and/or interpolation, is required to align the I-frame occupancy map to the three P-frames without transmitted occupancy map.

[0113] In the following, an enhanced method for point cloud frame intra coding will be described in more detail, in accordance with an embodiment.

[01 14] The method, which is disclosed in Figure 9, comprises inputting (900) a point cloud frame in an encoder; projecting (902) a 3D object represented by the point cloud frame onto a first arbitrary shaped 2D patch for generation of a geometry image, a texture image and, optionally, an attribute image; subdividing (904) the first 2D patch into image blocks of a predetermined size along a predetermined block grid; maintaining (906), in the first 2D patch, only fully filled image blocks where each pixel has a projected value; removing (908) the 3D object points having a corresponding 2D projection point in the fully filled image blocks of the first 2D patch from the 3D object; encoding (910) the first 2D patch of the geometry image, texture image or attribute image having only fully filled image blocks of the predetermined size into a bitstream of encoding video frames; and subjecting (912) remaining 3D object points iteratively for one or more new projections into subsequent 2D patches.

[01 15] Thus, compared to the test model of MPEG W17248, the use of arbitrary shaped 2D patches is significantly reduced or even completely removed through the introduction of block-aligned patches. A 3D object, represented as meshes, points, and/or voxel, is projected onto 2D planes, resulting in arbitrary shaped 2D patches for geometry, texture and

depth/attribute images. Before an individual 2D patch is packed into the 2D geometry, texture or depth/attribute image, it is subdivided to follow the block structure of any predetermined image or video codec, i.e. the individual 2D patches are subdivided into NxN blocks (e.g.,

8x8 blocks). The patch is then trimmed to contain only fully filled blocks, resulting in a block-aligned patch, which is suitable for image or video compression such that it is decodable by said predetermined image or video codec.

[01 16] Figure 10 illustrates an example of such trimming process, where an original patch and a block-aligned patch are shown in parallel. As shown in the block-aligned patch on the right, only fully filled blocks are included therein. Thus, any 3D-to-2D projection points represented in the trimmed block are removed from the original 3D shape and the trimmed patch is packed into the respective 2D images. Any 3D-to-2D projection points in the original patch, which were removed from the patch due to the trimming, remain in the original 3D shape and are available for the next patch/projection.

[0117] Accordingly, the usage of block-aligned patches may avoid or lessen the need of padding/interpolation for improved coding efficiency. Consequently, there might be no, or at least a reduced, need for transmitting occupancy information. Without an independent occupancy map, any 3D resampling and additional motion information is preferably not needed. The occupancy map may be coded at a spatial granularity of the block grid used in creating the block-aligned patches, and hence reduced in size when compared to a per-pixel occupancy map. Per-block occupancy map may be coded for example using context-adaptive variable length coding, such as CAB AC, e.g. with a l-bit bin and context state derivation from the adjacent blocks above and from the left. If the occupancy map is reduced in size, it may be possible to transmit it as per-frame, thus eliminating the need for 3D resampling and motion information.

[0118] According to an embodiment, the process is repeated until one of the following criteria is fulfilled: the complete 3D shape is projected, there are no more suitable patch projections to fill at least one full NxN block, or a predefined stopping condition is met. Thus, the remaining (i.e. not removed) 3D object points are repeatedly subjected for a new projection until one of the above criteria indicates that any further projection would not improve the efficiency of obtaining fully filled blocks.

[01 19] Figure 11 shows a flow chart of the block-alignment process for the 2D patches and various embodiments related thereto. As a starting point, a 3D object (1100), represented as meshes, points, and/or voxel, is projected onto a plurality of 2D planes, i.e. one plane for texture, one plane for geometry, and additional plane(s) for additional attributes. The result of the projection is arbitrary shaped 2D patches.

[0120] According to an embodiment, a projected patch is analysed and processed before packing it into the respective 2D images. A 2D patch is created (1102) from the projecting 3D data, wherein the projected 3D points are removed from the 3D object. The created patch is checked (1104) whether it fills at least one NxN block fully; if affirmative, the patch is divided (1106) into NxN blocks. If not affirmative, the projected points are returned to the 3D objected (1100). From a patch divided into blocks, it checked (1108), for each block, if it is fully tilled. Each partially tilled block is removed from the patch and the projected points contained by it are returned to the 3D object (1100). The block-aligned patch containing only fully filled blocks is packed (1110) into a block-aligned 2D image (1112) for encoding. This process is repeated until a stopping condition (1114) is met.

[0121] According to an embodiment, the stopping condition is one the following:

3D object has been fully projected, thus no further steps are required;

Percentage of remaining 3D points is below a certain threshold compared to original 3D points, e.g. 10% remaining 3D points mean that 90% of the 3D object have been successfully projected;

No further block-aligned patches are possible, i.e. not enough remaining points to form at least one fully filled NxN block;

A threshold number of patches is reached, i.e. limit the maximum number of projected patches; No available space in the projected 2D image, i.e. the resolution of the packed 2D images is limited;

Processing time limitation is met, i.e. the execution time of the process is limited.

[0122] It is noted that the above list is not comprehensive, but other stopping conditions are also feasible.

[0123] As becomes evident from the above, depending on the stopping condition, there may remain a number of 3D points not projected to any patches. Thus, further processing (1116) of possible remaining 3D data might be desired. According to an embodiment, the remaining non-projected 3D data is processed according to one of the following:

the remaining 3D points are dropped from the encoded representation;

the remaining 3D points are packed without block alignment in additional individual images with accompanying occupancy maps;

the remaining 3D points are packed without block alignment in the same pictures as the block-aligned patches.

[0124] The first option requires no additional signalling or decoder adaptions.

[0125] If the remaining 3D points are packed without block alignment in additional individual images with accompanying occupancy maps, such additional images would have a smaller resolution than the images containing the block-aligned patches. Herein, additional signaling of such occupancy information is required, as described further below.

[0126] If the remaining 3D points are packed without block alignment in the same pictures as the block-aligned patches, reduced signaling for occupancy map compression is introduced, as described further below.

[0127] Accordingly, the usage of the block-alignment of patches reduces such artefacts at patch boundaries that are caused by transform coefficient quantization. Thus, less or even no occupancy information at all is required.

[0128] According to an embodiment, when the remaining 3D points are packed without block alignment in additional individual images with accompanying occupancy maps, reduced signalling of occupancy information is used. Therein, a 3D object is decomposed into a set of block-aligned and non-block-aligned, arbitrary-shaped patches. The block-aligned patches are coded without any additional occupancy or motion information. The arbitrary-shaped patches may be coded following the approach outlined in MPEG W17248 regarding 3D reshaping and additional motion and occupancy information. [0129] Figure 12 shows a block chart of an encoder arranged to carry out reduced occupancy signalling. The encoder of Figure 12 has been adapted from the encoder of Figure

8.

[0130] The block-alignment and packing process results in a set of block-aligned patches and non-block-aligned, arbitrary-shaped patches. Thus, two images (block-aligned and additional image) per modality (texture, geometry, further possible attributes) are used, plus one occupancy map for the arbitrary-shaped patches image. The 2D-to-3D reconstruction can be performed sequential or in parallel for the block-aligned and the arbitrary-shaped patch data. In case of the arbitrary-shaped patch data, the occupancy map may be used to determine if a point will be reconstructed into 3D space or not.

[0131] Consequently, as a large part of the 3D object can typically be represented with block-aligned patches, the remaining arbitrary-shaped patches can be presented with a much smaller occupancy map. As a result, 3D reshaping and respective motion data is reduced as well. The size of the occupancy map may be reduced up to the point where a per-frame transmission becomes feasible. Thus, 3D reshaping and respective motion data would not be needed at all. This would also enable to use more complex random access prediction hierarchies including bi-prediction.

[0132] According to an embodiment, when the remaining 3D points are packed without block alignment in the same pictures as the block-aligned patches, block-aligned signaling for occupancy map compression is used. Therein, a 3D object is decomposed into a set of 2D patches and the patches are packed into a set of 2D images (texture, geometry, additional attributes) for video compression. Such images may contain block-aligned patches, arbitrary- shaped (non-block-aligned) patches or a combination of both. Arbitrary- shaped patches may be derived as remainders of the block-alignment process described above, or directly from the patch decomposition process, i.e. no block-alignment process is performed. If a combination of block-aligned and arbitrary-shaped patches is used, the patches may be packed freely into the 2D images, or the packing may follow certain rules, e.g. block-aligned patches are packed in one region and arbitrary-shaped patches in another region. The set of 2D images is accompanied by one single occupancy map of the same 2D resolution. The occupancy map is a binary representation comprising information about which 2D pixels shall be reconstructed into 3D space at the decoder side.

[0133] With reference to Figure 8, the occupancy map is signalled in or along the bitstream. Herein, various ways of arranging the occupancy information may be used. For example, the following order for arranging the occupancy information may be used: 1. Optionally signaling patch packing inside the 2D image. I.e. if block-aligned patches and arbitrary-shaped patches are packed in their own defined areas of the 2D image, these areas (containing only block-aligned or only arbitrary-shaped patches) can be signaled.

2. No further signaling for block-aligned only areas. For the remaining region, or if no

packing information is signaled, further signaling is performed:

3. If the 2D occupancy map, or area of the occupancy map, contains at least one arbitrary- shaped patch, the occupancy map is divided in NxN blocks. If block-alignment was used in the preceding patch generation step, the same blocks size may, but not necessarily need, be used. The blocks of the occupancy map division may be smaller, but preferably not larger than the blocks used for the block-alignment.

4. The state of each occupancy map block is signaled. There are three possible states:

I. The block is fully empty

II. The block is fully filled

III. The block is partially filled

5. The code words for signaling these three states can be derived using entropy coding, i.e. least bits for the state with the highest probability. Context adaptation may be used in the entropy coding. For example, the code words may be coded with CAB AC. In some embodiments, rather than treating the state of each occupancy map block as an individual code word, states of a set of occupancy map blocks is binarized, for example using run- length coding, and the binarized code word is further entropy coded. For example, a code word may be indicative of a state value and a run of consecutive occupancy map blocks in a particular scan order sharing the same indicated state.

6. No further signaling is required for states I and II. In case of state III, additional signaling is required to represent the per-pixel occupancy values. This can be achieved either by: a) Signaling pixel occupancy inside the block as a vector, e.g. 64 bits for an 8x8 block b) Signaling pixel occupancy inside the block as a quadtree. An example is shown in Fig.

13, where an 8x8 block is signaled as occupancy tree. For each block, three code words are possible: block is fully empty, block is fully filled, or block is partially filled. The code words for the first two cases would result in stopping the sub-division for this block. The code word for the third condition will result in sub-dividing the block into four sub blocks, until pixel level is reached. Again, entropy coding may be used to derive the code words.

[0134] For example, the following code words could be used for the example of Fig. 13:

Block empty: 0, Block full: 10, Block partially filled: 11. On the first level (8x8 block), the block is partially filled. On the second level (4x4 blocks), the first block is empty and the fourth block is full, whereas the second and third blocks are partially filled. On the third level in the second and third blocks, both contain two empty blocks, one full block and one partially filled block. The partially filled blocks are defined at the pixel level.

[0135] Thus, the quadtree signaling of the example of Fig. 13 would result in a 31 bit long bit stream: 11 (I^st level) 0 11 11 10 (2^nd level) 0 0 11 10 10 0 11 0 (3^rd level) 10 0 0 0 0 0 10 0 (pixel level).

[0136] According to an embodiment, run length coding may be used to reduce the required number of bits further.

[0137] According to an embodiment, the occupancy map may be reconstructed at the decoder side as follows:

Any empty occupancy map with the same resolution as the 2D texture and geometry images is initialized.

- Any possible patch packing information is decoded.

Regions marked as containing arbitrary-shaped patches (or if no patch packing information is signaled) are decoded reversing the respective encoder process (block sub-division, vector or quadtree coding).

[0138] Consequently, as a large part of the 3D object can typically be represented with block-aligned patches, the occupancy signalling of the remaining arbitrary-shaped patches requires much less bits. No additional video streams need to be transmitted either.

[0139] Furthermore, this approach can improve coding efficiency even without the pre processing block-alignment step. Consequently, the bit rate requirements for signalling the occupancy map may be reduced up to the point where a per-frame transmission becomes feasible. Thus, 3D reshaping and respective motion data would not be needed at all. This would also open the approach to more complex random access prediction hierarchies including bi-prediction.

[0140] In the above, some embodiments have been described with reference to encoding. It needs to be understood that said encoding may comprise one or more of the following:

encoding source image data into a bitstream, encapsulating the encoded bitstream in a container file and/or in packet(s) or stream(s) of a communication protocol, and announcing or describing the bitstream in a content description, such as the Media Presentation

Description (MPD) of ISO/IEC 23009-1 (known as MPEG-DASH) or the IETF Session Description Protocol (SDP). Similarly, some embodiments have been described with reference to decoding. It needs to be understood that said decoding may comprise one or more of the following: decoding image data from a bitstream, decapsulating the bitstream from a container file and/or from packet(s) or stream(s) of a communication protocol, and parsing a content description of the bitstream,

[0141] In the above, some embodiments have been described with reference to blocks of a particular size (NxN) along a block grid. It needs to be understood that embodiments may be realized for non-square blocks (MxN) if the underlying coding system supports such. It also needs to be understood that embodiments may be realized with multiple block partitioning levels. For example, a block of NxN may be considered filled in different embodiments, when it the underlying coding system support block partitioning of LxL blocks (e.g. CTUs of HE VC) such that the NxN block is a coding unit.

[0142] In the above, some embodiments have been described with reference to encoding or decoding texture images, geometry images, and (optionally) attribute images. It needs to be understood that these images are not necessarily separate images in encoding and/or decoding. For example the following options exist to represent a geometry image and/or one or more attribute images in relation to the associated texture image:

An additional sample array in addition to the conventional three colour component sample arrays of the texture picture.

A constituent frame of a frame-packed picture that may also comprise texture picture(s).

[0143] In the above, some embodiments have been described with reference to encoding or decoding texture pictures, geometry pictures, projection geometry information, and

(optionally) attribute pictures into or from a single bitstream. It needs to be understood that embodiments can be similarly realized when encoding or decoding texture pictures, geometry pictures, projection geometry information, and (optionally) attribute pictures into or from several bitstreams that are associated with each other, e.g. by metadata in a container file or media presentation description for streaming.

[0144] In general, the various embodiments of the invention may be implemented in hardware or special purpose circuits or any combination thereof. While various aspects of the invention may be illustrated and described as block diagrams or using some other pictorial representation, it is well understood that these blocks, apparatus, systems, techniques or methods described herein may be implemented in, as non-limiting examples, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof. [0145] Embodiments of the inventions may be practiced in various components such as integrated circuit modules. The design of integrated circuits is by and large a highly automated process. Complex and powerful software tools are available for converting a logic level design into a semiconductor circuit design ready to be etched and formed on a semiconductor substrate.

[0146] Programs, such as those provided by Synopsys, Inc. of Mountain View, California and Cadence Design, of San Jose, California automatically route conductors and locate components on a semiconductor chip using well established rules of design as well as libraries of pre stored design modules. Once the design for a semiconductor circuit has been completed, the resultant design, in a standardized electronic format (e.g., Opus, GDSII, or the like) may be transmitted to a semiconductor fabrication facility or "fab" for fabrication.

[0147] The foregoing description has provided by way of exemplary and non-limiting examples a full and informative description of the exemplary embodiment of this invention. However, various modifications and adaptations may become apparent to those skilled in the relevant arts in view of the foregoing description, when read in conjunction with the accompanying drawings and the appended claims. However, all such and similar

modifications of the teachings of this invention will still fall within the scope of this invention.

Claims

1. A method comprising:

inputting a point cloud frame in an encoder;

projecting a 3D object represented by the point cloud frame onto a first arbitrary shaped 2D patch for generation of at least a geometry image and a texture image;

subdividing the first 2D patch into image blocks of a predetermined size along a predetermined block grid;

maintaining, in the first 2D patch, only fully filled image blocks where each pixel has a projected value;

removing the 3D object points having a corresponding 2D projection point in the fully filled image blocks of the first 2D patch from the 3D object;

encoding the first 2D patch of the geometry image or texture image having only fully filled image blocks of the predetermined size into a bitstream of encoding video frames; and

subjecting remaining 3D object points iteratively for one or more new projections into subsequent 2D patches.

2. The method according to claim 1, wherein the remaining 3D object points are subjected for a new projection until one of the following criteria is fulfilled: complete shape of the 3D object is projected, there are no more suitable patch projections, or a predefined stopping condition is met.

3. The method according to claim 1 or 2, further comprising

analysing and processing said arbitrary shaped 2D patches prior to packing into the respective 2D images.

4. The method according to claim 3, wherein the analysing and processing for each of said arbitrary shaped 2D patches further comprises

checking whether the patch fills at least one predetermined image block fully; dividing the patch into image blocks of the predetermined size;

checking, for each image block, if the image block is fully filled;

removing any partially filled image block from the patch and returning the 2D projection points contained by said image block are to the 3D object; and packing the block-aligned 2D patch containing only fully filled blocks into a block- aligned 2D image for encoding.

5. The method according to any preceding claim, wherein the stopping condition is one the following:

- the 3D object has been fully projected;

- a percentage of remaining 3D points is below a certain threshold compared to original 3D points;

- no further block-aligned patches are possible;

- a threshold number of patches has been projected;

- no space available in the projected 2D images;

- time limit for the processing is met.

6. The method according to any preceding claim, wherein remaining non-projected 3D data is processed according to one of the following:

- the remaining 3D points are dropped from the encoded representation;

- the remaining 3D points are packed without block alignment in additional individual images with accompanying occupancy maps;

- the remaining 3D points are packed without block alignment in the same pictures as the block-aligned patches.

7. The method according to claim 6, wherein upon packing the remaining 3D points without block alignment in additional individual images with accompanying occupancy maps, the method further comprises

decomposing the 3D object into a set of block-aligned patches and arbitrary-shaped patches; and

using a reduced signalling of occupancy information only for the arbitrary- shaped patches.

8. The method according to claim 6, wherein upon packing the remaining 3D points without block alignment in the same pictures as the block-aligned patches, the method further comprises using block-aligned signaling for occupancy map compression, wherein the occupancy map comprises information about the 2D projection points to be reconstructed into 3D space at the decoder side.

9. The method according to claim 8, further comprising

signaling per-pixel occupancy information within a block as a quadtree.

10. An apparatus comprising at least one processor and at least one memory, said at least one memory stored with code thereon, which when executed by said at least one processor, causes the apparatus to perform the method according to at least one of claims 1 to 9.

11. A computer readable storage medium stored with code thereon for use by an apparatus, which when executed by a processor, causes the apparatus to perform the method according to at least one of claims 1 to 9.

12. An apparatus comprising:

means for inputting a point cloud frame in an encoder;

means for projecting a 3D object represented by the point cloud frame onto a first arbitrary shaped 2D patch for generation of at least a geometry image and a texture image;

means for subdividing the first 2D patch into image blocks of a predetermined size along a predetermined block grid;

means for maintaining, in the first 2D patch, only fully filled image blocks where each pixel has a projected value;

means for removing the 3D object points having a corresponding 2D projection point in the fully filled image blocks of the first 2D patch from the 3D object;

means for encoding the first 2D patch of the geometry image or texture image having only fully filled image blocks of the predetermined size into a bitstream of encoding video frames; and

means for subjecting remaining 3D object points iteratively for one or more new projections into subsequent 2D patches.