An enhanced feature processing for image compression based on feature distribution learning

Abstract

Some embodiments of an image or video processing method may include a learning-based point cloud geometry method in which the method may include: obtaining a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; transforming the first feature map to a second feature map based on the first set of distribution parameters; and outputting the second feature map to a succeeding neural network layer. For some embodiments of the image or video processing method, the first feature map further includes n feature vectors respectively corresponding to n distinct pixels in an image space.

Description

AN ENHANCED FEATURE PROCESSING FOR IMAGE COMPRESSION BASED ON FEATURE DISTRIBUTION LEARNINGCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims benefit of U.S. Provisional Patent Application No. 63 / 536,340, entitled “AN ENHANCED FEATURE PROCESSING FOR IMAGE COMPRESSION BASED ON FEATURE DISTRIBUTION LEARNING” and filed September 1, 2023, which is hereby incorporated by reference.INCORPORATION BY REFERENCE

[0002] The present application incorporates by reference in their entirety the following applications: International Application No. PCT / US2023 / 027424, entitled “DEEP DISTRIBUTION-AWARE POINT FEATURE EXTRACTOR FOR AI-BASED POINT CLOUD COMPRESSION” and filed July 11, 2023 (‘“424 application”), U.S. Provisional Patent Application Serial No. 63 / 388,600, entitled “DEEP DISTRIBUTION-AWARE POINT FEATURE EXTRACTOR FOR AI-BASED POINT CLOUD COMPRESSION” and filed July 12, 2022 (‘“600 application”); International Application No. PCT / US2022 / 052861 , entitled “SCALABLE FRAMEWORK FOR POINT CLOUD COMPRESSION” and filed December 14, 2022 (“‘861 application”); and U.S. Provisional Patent Application Serial No. 63 / 536,321 , entitled “AN ENHANCED FEATURE PROCESSING FOR POINT CLOUD COMPRESSION BASED ON FEATURE DISTRIBUTION LEARNING” and filed September 1, 2023 (“‘321 application”).BACKGROUND

[0003] Video and image data are universal data formats across several business domains, such as autonomous driving, robotics, civil engineering, video streaming, teleconferencing, computer vision, and the animation / movie industry. These universal data are believed to consume a large portion of network traffic, e.g., among connected cars over 5G network, and video communications teleconferencing. Efficient representation formats are necessary for image and video understanding and communication. Raw data needs to be properly organized and processed for the purposes of, e.g., world modeling, sensing, and rendering. Compression onthese raw input signals is essential when storage and transmission of the data are required in the related scenarios.SUMMARY

[0004] Embodiments described herein include methods that are used in video encoding and decoding (collectively “coding”).

[0005] An example method in accordance with some embodiments may include: obtaining a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; transforming the first feature map to a second feature map based on the first set of distribution parameters; and encoding a bitstream based on the second feature map.

[0006] A first example method in accordance with some embodiments may include: obtaining a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; transforming the first feature map to a second feature map based on the first set of distribution parameters; and outputting the second feature map to a succeeding neural network layer.

[0007] For some embodiments of the first example method, the first feature map further includes n feature vectors respectively corresponding to n distinct pixels in an image space.

[0008] For some embodiments of the first example method, each feature vector represents a feature of the corresponding pixel in the image space, and the first feature map has dimensions n by C.

[0009] Some embodiments of the first example method may further include: generating a second set of distribution parameters using a second set of neural networks layers; transforming the first feature map to a third feature map based on the second set of distribution parameters; and updating the first feature map to the third feature map.

[0010] For some embodiments of the first example method, generating the first set of distribution parameters may include: updating a given feature map via neural network layers; and simplifying the updated feature map to obtain the first set of per channel distribution parameters.

[0011] For some embodiments of the first example method, transforming the first feature map may include: generating, for each of the C channels in the first feature map, a respective reshaped vector, wherein each reshaped vector is generated by reshaping a respective channel vector in the first feature map; obtaining a distribution parameter for each of the C channels corresponding to one of the respective reshaped vectors; expanding each of the distribution parameters into a respective feature channel; and transforming, for each of the C channels, each element in the reshaped vector by the expanded distribution parameter.

[0012] Some embodiments of the first example method may further include updating the first feature map by normalizing each feature element of the first feature map.

[0013] For some embodiments of the first example method, normalizing each feature element of the first feature map may include: generating a third set of distribution parameters using a third set of neural network layers; centering the first feature map to a fourth feature map based on the third set of distribution parameters; and updating the first feature map to the fourth feature map.

[0014] Some embodiments of the first example method may further include: generating a fourth set of distribution parameters from the fourth feature map; and normalizing the fourth feature map to a fifth feature map based on the fourth set of distribution parameters.

[0015] Some embodiments of the first example method may further include: generating a fifth set of distribution parameters using a fifth set of neural networks layers; and normalizing the fourth feature map to a sixth feature map based on the fifth set of distribution parameters.

[0016] Some embodiments of the first example method may further include downsampling the second feature map to generate a first feature vector, wherein downsampling the second feature map uses a pooling function.

[0017] For some embodiments of the first example method, the pooling function is selected from the group consisting of average pooling and max pooling.

[0018] Some embodiments of the first example method may further include: generating an expanded feature map by expanding the first feature vector towards a channel dimension; concatenating the second feature map with the expanded feature map to generate a concatenated feature map; and passing the concatenated feature map through a filtering neural network layers to generate a filtered feature map.

[0019] Some embodiments of the first example method may further include aggregating the second feature map using an additional neural network.

[0020] For some embodiments of the first example method, the additional neural network includes a convolution neural network (CNN) configured for image data.

[0021] Some embodiments of the first example method may further include aggregating the second feature map using a residual network.

[0022] For some embodiments of the first example method, the residual network is a ResNet block.

[0023] Some embodiments of the first example method may further include aggregating the second feature map using a transformer block.

[0024] For some embodiments of the first example method, the transformer block is selected from the group consisting of a point transformer and a voxel transformer.

[0025] Some embodiments of the first example method may further include: generating a seventh feature map by aggregating the second feature map using a neural network in parallel to transform the first feature map to the second feature map; and concatenating the seventh feature map to the second feature map.

[0026] Some embodiments of the first example method may further include: repeating the learning-based point cloud geometry process one or more times to generate an eighth feature map; and adding the eighth feature map to the first feature map to generate a ninth feature map.

[0027] An example apparatus in accordance with some embodiments may include: a processor; and a non- transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map; transform the first feature map to a second feature map based on the first set of distribution parameters; and encode a bitstream based on the second feature map.

[0028] A first example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map; transform thefirst feature map to a second feature map based on the first set of distribution parameters; and output the second feature map to a succeeding neural network layer.

[0029] For some embodiments of the first example apparatus, the first feature map further includes n feature vectors respectively corresponding to n distinct pixels in an image space.

[0030] For some embodiments of the first example apparatus, each feature vector represents a feature of the corresponding pixel in the image space, and the first feature map has dimensions n by C.

[0031] An example method in accordance with some embodiments may include: obtaining a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; expanding the first set of distribution parameters to dimensionally-match the first feature map; subtracting the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generating a second set of distribution parameters using a second set of neural network layers based on the first feature map; expanding the second set of distribution parameters to dimensionally-match the centered feature map; dividing the centered feature map by the expanded second set of distribution parameters to generate a normalized feature map; and encoding a bitstream based on the normalized feature map.

[0032] A second example method in accordance with some embodiments may include: obtaining a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; expanding the first set of distribution parameters to dimensionally-match the first feature map; subtracting the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generating a second set of distribution parameters using a second set of neural network layers based on the first feature map; expanding the second set of distribution parameters to dimensionally-match the centered feature map; dividing the centered feature map by the expanded second set of distribution parameters to generate a normalized feature map; and outputting the normalized feature map to a succeeding neural network layer.

[0033] For some embodiments of the second example method, the first feature map further includes n feature vectors respectively corresponding to n distinct pixels in an image space.

[0034] For some embodiments of the second example method, each feature vector represents a feature of the corresponding pixel in the image space, and the first feature map has dimensions n by C.

[0035] For some embodiments of the second example method, expanding the first set of distribution parameters to dimensionally-match the first feature map includes copying a first distribution parameter vector one or more times to dimensionally-match a corresponding dimension of the first feature map, the first distribution parameter vector includes the first set of distribution parameters, expanding the second set of distribution parameters to dimensionally-match the centered feature map includes copying a second distribution parameter vector one or more times to dimensionally-match a corresponding dimension of the centered feature map, and the second distribution parameter vector includes the second set of distribution parameters.

[0036] An example apparatus in accordance with some embodiments may include: a processor; and a non- transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map; expand the first set of distribution parameters to dimensionally-match the first feature map; subtract the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generate a second set of distribution parameters using a second set of neural network layers based on the first feature map; expand the second set of distribution parameters to dimensionally-match the centered feature map; divide the centered feature map by the expanded second set of distribution parameters to generate a normalized feature map; and encode a bitstream based on the normalized feature map.

[0037] A second example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map; expand the first set of distribution parameters to dimensionally-match the first feature map; subtract the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generate a second set of distribution parameters using a second set of neural network layers based on the first feature map; expand the second set of distribution parameters to dimensionally-match the centered feature map; divide the centeredfeature map by the expanded second set of distribution parameters to generate a normalized feature map; and output the normalized feature map to a succeeding neural network layer.

[0038] An example method in accordance with some embodiments may include: obtaining a normalized feature map, wherein the normalized feature map includes C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the normalized feature map; expanding the first set of distribution parameters to dimensionally- match the normalized feature map; scaling the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map; generating a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expanding the second set of distribution parameters to dimensionally-match the scaled feature map; adding the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and encoding a bitstream based on the transformed feature map.

[0039] A third example method in accordance with some embodiments may include: obtaining a normalized feature map, wherein the normalized feature map includes C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the normalized feature map; expanding the first set of distribution parameters to dimensionally- match the normalized feature map; scaling the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map; generating a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expanding the second set of distribution parameters to dimensionally-match the scaled feature map; adding the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and outputting the transformed feature map to a succeeding neural network layer.

[0040] For some embodiments of the third example method, the first feature map further includes n feature vectors respectively corresponding to n distinct pixels in an image space.

[0041] For some embodiments of the third example method, each feature vector represents a feature of the corresponding pixel in the image space, and the first feature map has dimensions n by C.

[0042] For some embodiments of the third example method, an input to the second set of neural network layers is the scaled feature map.

[0043] For some embodiments of the third example method, an input to the second set of neural network layers is the normalized feature map.

[0044] Some embodiments of the third example method may further include downsampling the transformed feature map to generate a first feature vector, wherein downsampling the transformed feature map uses a pooling function.

[0045] For some embodiments of the third example method, the pooling function is selected from the group consisting of average pooling and max pooling.

[0046] Some embodiments of the third example method may further include: generating an expanded feature map by expanding the first feature vector towards a channel dimension; concatenating the first feature map with the expanded feature map to generate a concatenated feature map; and passing the concatenated feature map through a filtering neural network layers to generate a filtered feature map.

[0047] Some embodiments of the third example method may further include aggregating the first feature map using an additional neural network.

[0048] For some embodiments of the third example method, the additional neural network includes a convolution neural network (CNN) configured for image data.

[0049] Some embodiments of the third example method may further include aggregating the first feature map using a residual network.

[0050] For some embodiments of the third example method, the residual network is a ResNet block.

[0051] Some embodiments of the third example method may further include: repeating the learning-based image processing method one or more times to generate an enhanced feature map; and adding the enhanced feature map to the first feature map to generate an enhanced output feature map.

[0052] Some embodiments of the third example method may further include repeating the learning-based image processing method one or more times within a point cloud encoder.

[0053] Some embodiments of the third example method may further include repeating the learning-based image processing method one or more times within a point cloud decoder.

[0054] For some embodiments of the third example method, the image processing encoder processes pixels.

[0055] For some embodiments of the third example method, the image processing encoder processes images.

[0056] An example apparatus in accordance with some embodiments may include: a processor; and a non- transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a normalized feature map, wherein the normalized feature map includes C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the normalized feature map; expand the first set of distribution parameters to dimensionally-match the normalized feature map; scale the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map; generate a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expand the second set of distribution parameters to dimensionally-match the scaled feature map; add the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and encode a bitstream based on the transformed feature map.

[0057] A third example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a normalized feature map, wherein the normalized feature map includes C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the normalized feature map; expand the first set of distribution parameters to dimensionally-match the normalized feature map; scale the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map; generate a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expand the second set of distribution parameters to dimensionally-match the scaled feature map; add the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and output the transformed feature map to a succeeding neural network layer.

[0058] In additional embodiments, encoder and decoder apparatus are provided to perform the methods described herein. An encoder or decoder apparatus may include a processor configured to perform the methodsdescribed herein. The apparatus may include a computer-readable medium (e.g. a non-transitory medium) storing instructions for performing the methods described herein. In some embodiments, a computer-readable medium (e.g. a non-transitory medium) stores a video encoded using any of the methods described herein.

[0059] One or more of the present embodiments also provide a computer readable storage medium having stored thereon instructions for performing bi-directional optical flow, encoding or decoding video data according to any of the methods described above. The present embodiments also provide a computer readable storage medium having stored thereon a bitstream generated according to the methods described above. The present embodiments also provide a method and apparatus for transmitting the bitstream generated according to the methods described above. The present embodiments also provide a computer program product including instructions for performing any of the methods described.BRIEF DESCRIPTION OF THE DRAWINGS

[0060] FIG. 1A is a system diagram illustrating an example communications system according to some embodiments.

[0061] FIG. 1 B is a system diagram illustrating an example wireless transmit / receive unit (WTRU) that may be used within the communications system illustrated in FIG. 1 A according to some embodiments.

[0062] FIG. 1 C is a system diagram illustrating an example set of interfaces for a system according to some embodiments.

[0063] FIG. 2A is a functional block diagram of block-based video encoder, such as an encoder used for Versatile Video Coding (WC), according to some embodiments.

[0064] FIG. 2B is a functional block diagram of a block-based video decoder, such as a decoder used for WC, according to some embodiments.

[0065] FIG. 3A is a schematic side view illustrating an example waveguide display that may be used with extended reality (XR) applications according to some embodiments.

[0066] FIG. 3B is a schematic side view illustrating an example alternative display type that may be used with extended reality applications according to some embodiments.

[0067] FIG. 3C is a schematic side view illustrating an example alternative display type that may be used with extended reality applications according to some embodiments.

[0068] FIG. 4 is a schematic illustration showing example feature vectors and feature map in image space according to some embodiments.

[0069] FIG. 5 is a schematic illustration showing an example distribution parameter corresponding to a reshaped vector according to some embodiments.

[0070] FIG. 6 is a functional block diagram showing example neural network layers for learning feature map distribution parameters according to some embodiments.

[0071] FIG. 7 is a schematic illustration showing an example of three convolutional neural network layers according to some embodiments.

[0072] FIG. 8 is a functional block diagram showing a first example feature map normalization process according to some embodiments.

[0073] FIG. 9 is a schematic illustration showing an example expansion to a feature map dimension according to some embodiments.

[0074] FIG. 10 is a functional block diagram showing a second example feature map normalization process according to some embodiments.

[0075] FIG. 11 is a functional block diagram showing a first example feature map transformation process according to some embodiments.

[0076] FIG. 12 is a functional block diagram showing a second example feature map transformation process according to some embodiments.

[0077] FIG. 13 is a functional block diagram showing an example downsampling of a feature map according to some embodiments.

[0078] FIG. 14 is a functional block diagram showing an example filtering of a feature map according to some embodiments.

[0079] FIG. 15 is a functional block diagram showing an example parallel architecture for an enhanced distribution learning network according to some embodiments.

[0080] FIG. 16 is a functional block diagram showing an example serial architecture for an enhanced distribution learning network according to some embodiments.

[0081] FIG. 17 is a functional block diagram showing an example deep enhanced distribution learning network according to some embodiments.

[0082] FIG. 18 is a functional block diagram showing an example deep enhanced distribution learning network in an image compression framework according to some embodiments.

[0083] FIG. 19 is a flowchart illustrating an example learning-based image compression process according to some embodiments.

[0084] FIG. 20 is a flowchart illustrating an example learning-based image compression process according to some embodiments.

[0085] FIG. 21 is a flowchart illustrating an example learning-based image compression process according to some embodiments.

[0086] The entities, connections, arrangements, and the like that are depicted in— and described in connection with— the various figures are presented by way of example and not byway of limitation. As such, any and all statements or other indications as to what a particular figure “depicts,” what a particular element or entity in a particular figure “is” or “has,” and any and all similar statements— that may in isolation and out of context be read as absolute and therefore limiting— may only properly be read as being constructively preceded by a clause such as “In at least one embodiment, ... " For brevity and clarity of presentation, this implied leading clause is not repeated ad nauseam in the detailed description.DETAILED DESCRIPTION

[0087] FIG. 1A is a diagram illustrating an example communications system 100 in which one or more disclosed embodiments may be implemented. The communications system 100 may be a multiple access system that provides content, such as voice, data, video, messaging, broadcast, etc., to multiple wireless users. The communications system 100 may enable multiple wireless users to access such content through the sharing of system resources, including wireless bandwidth. For example, the communications systems 100 may employ one or more channel access methods, such as code division multiple access (CDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal FDMA (OFDMA), single-carrier FDMA (SC-FDMA), zero-tail unique-word DFT-Spread OFDM (ZT UW DTS-s OFDM), unique word OFDM (UW-OFDM), resource block-filtered OFDM, filter bank multicarrier (FBMC), and the like.

[0088] As shown in FIG. 1A, the communications system 100 may include wireless transmit / receive units (WTRUs) 102a, 102b, 102c, 102d, a RAN 104 / 113, a CN 106, a public switched telephone network (PSTN) 108, the Internet 110, and other networks 112, though it will be appreciated that the disclosed embodiments contemplate any number of WTRUs, base stations, networks, and / or network elements. Each of the WTRUs 102a, 102b, 102c, 102d may be any type of device configured to operate and / or communicate in a wireless environment. By way of example, the WTRUs 102a, 102b, 102c, 102d, any of which may be referred to as a “station” and / or a “STA”, may be configured to transmit and / or receive wireless signals and may include a user equipment (UE), a mobile station, a fixed or mobile subscriber unit, a subscription-based unit, a pager, a cellular telephone, a personal digital assistant (PDA), a smartphone, a laptop, a netbook, a personal computer, a wireless sensor, a hotspot or Mi-Fi device, an Internet of Things (loT) device, a watch or other wearable, a head-mounted display (HMD), a vehicle, a drone, a medical device and applications (e.g. , remote surgery), an industrial device and applications (e.g., a robot and / or other wireless devices operating in an industrial and / or an automated processing chain contexts), a consumer electronics device, a device operating on commercial and / or industrial wireless networks, and the like. Any of the WTRUs 102a, 102b, 102c and 102d may be interchangeably referred to as a UE.

[0089] The communications systems 100 may also include a base station 114a and / or a base station 114b. Each of the base stations 114a, 114b may be any type of device configured to wirelessly interface with at least one of the WTRUs 102a, 102b, 102c, 102d to facilitate access to one or more communication networks, such as the CN 106, the Internet 110, and / or the other networks 112. By way of example, the base stations 114a, 114b may be a base transceiver station (BTS), a Node-B, an eNode B, a Home Node B, a Home eNode B, a gNB, a NR NodeB, a site controller, an access point (AP), a wireless router, and the like. While the base stations 114a, 114b are each depicted as a single element, it will be appreciated that the base stations 114a, 114b may include any number of interconnected base stations and / or network elements.

[0090] The base station 114a may be part of the RAN 104 / 113, which may also include other base stations and / or network elements (not shown), such as a base station controller (BSC), a radio network controller (RNC), relay nodes, etc. The base station 114a and / or the base station 114b may be configured to transmit and / or receive wireless signals on one or more carrier frequencies, which may be referred to as a cell (not shown). These frequencies may be in licensed spectrum, unlicensed spectrum, or a combination of licensed and unlicensed spectrum. A cell may provide coverage for a wireless service to a specific geographical area that may be relatively fixed or that may change over time. The cell may further be divided into cell sectors. For example,the cell associated with the base station 114a may be divided into three sectors. Thus, in one embodiment, the base station 114a may include three transceivers, i.e., one for each sector of the cell. In an embodiment, the base station 114a may employ multiple-input multiple output (MIMO) technology and may utilize multiple transceivers for each sector of the cell. For example, beamforming may be used to transmit and / or receive signals in desired spatial directions.

[0091] The base stations 114a, 114b may communicate with one or more of the WTRUs 102a, 102b, 102c, 102d over an air interface 116, which may be any suitable wireless communication link (e.g., radio frequency (RF), microwave, centimeter wave, micrometer wave, infrared (IR), ultraviolet (UV), visible light, etc.). The air interface 116 may be established using any suitable radio access technology (RAT).

[0092] More specifically, as noted above, the communications system 100 may be a multiple access system and may employ one or more channel access schemes, such as CDMA, TDMA, FDMA, OFDMA, SC-FDMA, and the like. For example, the base station 114a in the RAN 104 / 113 and the WTRUs 102a, 102b, 102c may implement a radio technology such as Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access (UTRA), which may establish the air interface 116 using wideband CDMA (WCDMA). WCDMA may include communication protocols such as High-Speed Packet Access (HSPA) and / or Evolved HSPA (HSPA+). HSPA may include High-Speed Downlink (DL) Packet Access (HSDPA) and / or High-Speed UL Packet Access (HSUPA).

[0093] In an embodiment, the base station 114a and the WTRUs 102a, 102b, 102c may implement a radio technology such as Evolved UMTS Terrestrial Radio Access (E-UTRA), which may establish the air interface 116 using Long Term Evolution (LTE) and / or LTE-Advanced (LTE-A) and / or LTE-Advanced Pro (LTE-A Pro).

[0094] In an embodiment, the base station 114a and the WTRUs 102a, 102b, 102c may implement a radio technology such as NR Radio Access , which may establish the air interface 116 using New Radio (NR).

[0095] In an embodiment, the base station 114a and the WTRUs 102a, 102b, 102c may implement multiple radio access technologies. For example, the base station 114a and the WTRUs 102a, 102b, 102c may implement LTE radio access and NR radio access together, for instance using dual connectivity (DC) principles. Thus, the air interface utilized by WTRUs 102a, 102b, 102c may be characterized by multiple types of radio access technologies and / or transmissions sent to / from multiple types of base stations (e.g., a eNB and a gNB).

[0096] In other embodiments, the base station 114a and the WTRUs 102a, 102b, 102c may implement radio technologies such as IEEE 802.11 (i.e., Wireless Fidelity (WiFi), IEEE 802.16 (i.e., Worldwide Interoperability forMicrowave Access (WiMAX)), CDMA2000, CDMA2000 1X, CDMA2000 EV-DO, Interim Standard 2000 (IS- 2000), Interim Standard 95 (IS-95), Interim Standard 856 (IS-856), Global System for Mobile communications (GSM), Enhanced Data rates for GSM Evolution (EDGE), GSM EDGE (GERAN), and the like.

[0097] The base station 114b in FIG. 1A may be a wireless router, Home Node B, Home eNode B, or access point, for example, and may utilize any suitable RAT for facilitating wireless connectivity in a localized area, such as a place of business, a home, a vehicle, a campus, an industrial facility, an air corridor (e. g . , for use by drones), a roadway, and the like. In one embodiment, the base station 114b and the WTRUs 102c, 102d may implement a radio technology such as IEEE 802.11 to establish a wireless local area network (WLAN). In an embodiment, the base station 114b and the WTRUs 102c, 102d may implement a radio technology such as IEEE 802.15 to establish a wireless personal area network (WPAN). In yet another embodiment, the base station 114b and the WTRUs 102c, 102d may utilize a cellular-based RAT (e.g., WCDMA, CDMA2000, GSM, LTE, LTE-A, LTE-A Pro, NR etc.) to establish a picocell or femtocell. As shown in FIG. 1 A, the base station 114b may have a direct connection to the Internet 110. Thus, the base station 114b may not be required to access the Internet 110 via the CN 106.

[0098] The RAN 104 / 113 may be in communication with the CN 106, which may be any type of network configured to provide voice, data, applications, and / or voice over internet protocol (VoIP) services to one or more of the WTRUs 102a, 102b, 102c, 102d. The data may have varying quality of service (QoS) requirements, such as differing throughput requirements, latency requirements, error tolerance requirements, reliability requirements, data throughput requirements, mobility requirements, and the like. The CN 106 may provide call control, billing services, mobile location-based services, pre-paid calling, Internet connectivity, video distribution, etc., and / or perform high-level security functions, such as user authentication. Although not shown in FIG. 1A, it will be appreciated that the RAN 104 / 113 and / or the CN 106 may be in direct or indirect communication with other RANs that employ the same RAT as the RAN 104 / 113 or a different RAT. For example, in addition to being connected to the RAN 104 / 113, which may be utilizing a NR radio technology, the CN 106 may also be in communication with another RAN (not shown) employing a GSM, UMTS, CDMA 2000, WiMAX, E-UTRA, or WiFi radio technology.

[0099] The CN 106 may also serve as a gateway for the WTRUs 102a, 102b, 102c, 102d to access the PSTN108, the Internet 110, and / or the other networks 112. The PSTN 108 may include circuit-switched telephone networks that provide plain old telephone service (POTS). The Internet 110 may include a global system of interconnected computer networks and devices that use common communication protocols, such as thetransmission control protocol (TCP), user datagram protocol (UDP) and / or the internet protocol (IP) in the TCP / IP internet protocol suite. The networks 112 may include wired and / or wireless communications networks owned and / or operated by other service providers. For example, the networks 112 may include another CN connected to one or more RANs, which may employ the same RAT as the RAN 104 / 113 or a different RAT.

[0100] Some or all of the WTRUs 102a, 102b, 102c, 102d in the communications system 100 may include multi-mode capabilities (e.g., the WTRUs 102a, 102b, 102c, 102d may include multiple transceivers for communicating with different wireless networks over different wireless links). For example, the WTRU 102c shown in FIG. 1A may be configured to communicate with the base station 114a, which may employ a cellularbased radio technology, and with the base station 114b, which may employ an IEEE 802 radio technology.

[0101] FIG. 1B is a system diagram illustrating an example WTRU 102. As shown in FIG. 1 B, the WTRU 102 may include a processor 118, a transceiver 120, a transmit / receive element 122, a speaker / microphone 124, a keypad 126, a display / touchpad 128, non-removable memory 130, removable memory 132, a power source 134, a global positioning system (GPS) chipset 136, and / or other peripherals 138, among others. It will be appreciated that the WTRU 102 may include any sub-combination of the foregoing elements while remaining consistent with an embodiment.

[0102] The processor 118 may be a general purpose processor, a special purpose processor, a conventional processor, a digital signal processor (DSP), a plurality of microprocessors, one or more microprocessors in association with a DSP core, a controller, a microcontroller, Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) circuits, any other type of integrated circuit (IC), a state machine, and the like. The processor 118 may perform signal coding, data processing, power control, input / output processing, and / or any other functionality that enables the WTRU 102 to operate in a wireless environment. The processor 118 may be coupled to the transceiver 120, which may be coupled to the transmit / receive element 122. While FIG. 1 B depicts the processor 118 and the transceiver 120 as separate components, it will be appreciated that the processor 118 and the transceiver 120 may be integrated together in an electronic package or chip.

[0103] The transmit / receive element 122 may be configured to transmit signals to, or receive signals from, a base station (e.g., the base station 114a) over the air interface 116. For example, in one embodiment, the transmit / receive element 122 may be an antenna configured to transmit and / or receive RF signals. In an embodiment, the transmit / receive element 122 may be an emitter / detector configured to transmit and / or receive IR, UV, or visible light signals, for example. In yet another embodiment, the transmit / receive element 122 maybe configured to transmit and / or receive both RF and light signals. It will be appreciated that the transmit / receive element 122 may be configured to transmit and / or receive any combination of wireless signals.

[0104] Although the transmit / receive element 122 is depicted in FIG. 1 B as a single element, the WTRU 102 may include any number of transmit / receive elements 122. More specifically, the WTRU 102 may employ MIMO technology. Thus, in one embodiment, the WTRU 102 may include two or more transmit / receive elements 122 (e.g., multiple antennas) for transmitting and receiving wireless signals over the air interface 116.

[0105] The transceiver 120 may be configured to modulate the signals that are to be transmitted by the transmit / receive element 122 and to demodulate the signals that are received by the transmit / receive element 122. As noted above, the WTRU 102 may have multi-mode capabilities. Thus, the transceiver 120 may include multiple transceivers for enabling the WTRU 102 to communicate via multiple RATs, such as NR and IEEE 802.11, for example.

[0106] The processor 118 of the WTRU 102 may be coupled to, and may receive user input data from, the speaker / microphone 124, the keypad 126, and / or the display / touchpad 128 (e.g., a liquid crystal display (LCD) display unit or organic light-emitting diode (OLED) display unit). The processor 118 may also output user data to the speaker / microphone 124, the keypad 126, and / or the display / touchpad 128. In addition, the processor 118 may access information from, and store data in, any type of suitable memory, such as the non-removable memory 130 and / or the removable memory 132. The non-removable memory 130 may include random-access memory (RAM), read-only memory (ROM), a hard disk, or any other type of memory storage device. The removable memory 132 may include a subscriber identity module (SIM) card, a memory stick, a secure digital (SD) memory card, and the like. In other embodiments, the processor 118 may access information from, and store data in, memory that is not physically located on the WTRU 102, such as on a server or a home computer (not shown).

[0107] The processor 118 may receive power from the power source 134, and may be configured to distribute and / or control the power to the other components in the WTRU 102. The power source 134 may be any suitable device for powering the WTRU 102. For example, the power source 134 may include one or more dry cell batteries (e.g., nickel-cadmium (NiCd), nickel-zinc (NiZn), nickel metal hydride (NiMH), lithium-ion (Li-ion), etc.), solar cells, fuel cells, and the like.

[0108] The processor 118 may also be coupled to the GPS chipset 136, which may be configured to provide location information (e.g., longitude and latitude) regarding the current location of the WTRU 102. In addition to,or in lieu of, the information from the GPS chipset 136, the WTRU 102 may receive location information over the air interface 116 from a base station (e.g., base stations 114a, 114b) and / or determine its location based on the timing of the signals being received from two or more nearby base stations. It will be appreciated that the WTRU 102 may acquire location information by way of any suitable location-determination method while remaining consistent with an embodiment.

[0109] The processor 118 may further be coupled to other peripherals 138, which may include one or more software and / or hardware modules that provide additional features, functionality and / or wired or wireless connectivity. For example, the peripherals 138 may include an accelerometer, an e-compass, a satellite transceiver, a digital camera (for photographs and / or video), a universal serial bus (USB) port, a vibration device, a television transceiver, a hands free headset, a Bluetooth® module, a frequency modulated (FM) radio unit, a digital music player, a media player, a video game player module, an Internet browser, a Virtual Reality and / or Augmented Reality (VR / AR) device, an activity tracker, and the like. The peripherals 138 may include one or more sensors, the sensors may be one or more of a gyroscope, an accelerometer, a hall effect sensor, a magnetometer, an orientation sensor, a proximity sensor, a temperature sensor, a time sensor; a geolocation sensor; an altimeter, a light sensor, a touch sensor, a magnetometer, a barometer, a gesture sensor, a biometric sensor, and / or a humidity sensor.

[0110] The WTRU 102 may include a full duplex radio for which transmission and reception of some or all of the signals (e.g., associated with particular subframes for both the UL (e.g., for transmission) and downlink (e.g., for reception) may be concurrent and / or simultaneous. The full duplex radio may include an interference management unit to reduce and or substantially eliminate self-interference via either hardware (e.g., a choke) or signal processing via a processor (e.g., a separate processor (not shown) or via processor 118). In an embodiment, the WTRU 102 may include a half-duplex radio for which transmission and reception of some or all of the signals (e.g., associated with particular subframes for either the UL (e.g., for transmission) or the downlink (e.g, for reception)).

[0111] Although the WTRU is described in FIGs. 1 A-1 B as a wireless terminal, it is contemplated that in certain representative embodiments that such a terminal may use (e.g, temporarily or permanently) wired communication interfaces with the communication network.

[0112] In representative embodiments, the other network 112 may be a WLAN.

[0113] In view of FIGs. 1A-1 B, and the corresponding description, one or more, or all, of the functions described herein may be performed by one or more emulation devices (not shown). The emulation devices may be one or more devices configured to emulate one or more, or all, of the functions described herein. For example, the emulation devices may be used to test other devices and / or to simulate network and / or WTRU functions.

[0114] The emulation devices may be designed to implement one or more tests of other devices in a lab environment and / or in an operator network environment. For example, the one or more emulation devices may perform the one or more, or all, functions while being fully or partially implemented and / or deployed as part of a wired and / or wireless communication network in order to test other devices within the communication network. The one or more emulation devices may perform the one or more, or all, functions while being temporarily implemented / deployed as part of a wired and / or wireless communication network. The emulation device may be directly coupled to another device for purposes of testing and / or may performing testing using over-the-air wireless communications.

[0115] The one or more emulation devices may perform the one or more, including all, functions while not being implemented / deployed as part of a wired and / or wireless communication network. For example, the emulation devices may be utilized in a testing scenario in a testing laboratory and / or a non-deployed (e.g., testing) wired and / or wireless communication network in order to implement testing of one or more components. The one or more emulation devices may be test equipment. Direct RF coupling and / or wireless communications via RF circuitry (e.g., which may include one or more antennas) may be used by the emulation devices to transmit and / or receive data.

[0116] FIG. 1 C is a system diagram illustrating an example set of interfaces for a system according to some embodiments. An extended reality display device, together with its control electronics, may be implemented using a system such as, e.g., the example systems of FIGS. 3B, or 3C. System 150 can be embodied as a device including the various components described below and is configured to perform one or more of the aspects described in this document. Examples of such devices, include, but are not limited to, various electronic devices such as personal computers, laptop computers, smartphones, tablet computers, digital multimedia set top boxes, digital television receivers, personal video recording systems, connected home appliances, and servers. Elements of system 150, singly or in combination, can be embodied in a single integrated circuit (IC), multiple ICs, and / or discrete components. For example, in at least one embodiment, the processing and encoder / decoder elements of system 150 are distributed across multiple ICs and / or discrete components. In various embodiments, the system 150 is communicatively coupled to one or more other systems, or other electronic devices, via, forexample, a communications bus or through dedicated input and / or output ports. In various embodiments, the system 1000 is configured to implement one or more of the aspects described in this document.

[0117] The system 150 includes at least one processor 152 configured to execute instructions loaded therein for implementing, for example, the various aspects described in this document. Processor 152 may include embedded memory, input output interface, and various other circuitries as known in the art. The system 150 includes at least one memory 154 (e.g., a volatile memory device, and / or a non-volatile memory device). System 150 may include a storage device 158, which can include non-volatile memory and / or volatile memory, including, but not limited to, Electrically Erasable Programmable Read-Only Memory (EEPROM), Read-Only Memory (ROM), Programmable Read-Only Memory (PROM), Random Access Memory (RAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), flash, magnetic disk drive, and / or optical disk drive. The storage device 158 can include an internal storage device, an attached storage device (including detachable and non-detachable storage devices), and / or a network accessible storage device, as non-limiting examples.

[0118] System 150 includes an encoder / decoder module 156 configured, for example, to process data to provide an encoded video or decoded video, and the encoder / decoder module 156 can include its own processor and memory. The encoder / decoder module 156 represents module(s) that can be included in a device to perform the encoding and / or decoding functions. As is known, a device can include one or both of the encoding and decoding modules. Additionally, encoder / decoder module 156 can be implemented as a separate element of system 150 or can be incorporated within processor 152 as a combination of hardware and software as known to those skilled in the art.

[0119] Program code to be loaded onto processor 152 or encoder / decoder 156 to perform the various aspects described in this document can be stored in storage device 158 and subsequently loaded onto memory 154 for execution by processor 152. In accordance with various embodiments, one or more of processor 152, memory 154, storage device 158, and encoder / decoder module 156 can store one or more of various items during the performance of the processes described in this document. Such stored items can include, but are not limited to, the input video, the decoded video or portions of the decoded video, the bitstream, matrices, variables, and intermediate or final results from the processing of equations, formulas, operations, and operational logic.

[0120] In some embodiments, memory inside of the processor 152 and / or the encoder / decoder module 156 is used to store instructions and to provide working memory for processing that is needed during encoding or decoding. In other embodiments, however, a memory external to the processing device (for example, theprocessing device can be either the processor 152 or the encoder / decoder module 152) is used for one or more of these functions. The external memory can be the memory 154 and / or the storage device 158, for example, a dynamic volatile memory and / or a non-volatile flash memory. In several embodiments, an external non-volatile flash memory is used to store the operating system of, for example, a television. In at least one embodiment, a fast external dynamic volatile memory such as a RAM is used as working memory for video coding and decoding operations, such as for MPEG-2 (MPEG refers to the Moving Picture Experts Group, MPEG-2 is also referred to as ISO / IEC 13818, and 13818-1 is also known as H.222, and 13818-2 is also known as H.262), HEVC (HEVC refers to High Efficiency Video Coding, also known as H.265 and MPEG-H Part 2), or WC (Versatile Video Coding, a new standard being developed by JVET, the Joint Video Experts Team).

[0121] The input to the elements of system 150 can be provided through various input devices as indicated in block 172. Such input devices include, but are not limited to, (i) a radio frequency (RF) portion that receives an RF signal transmitted, for example, over the air by a broadcaster, (ii) a Component (COMP) input terminal (or a set of COMP input terminals), (iii) a Universal Serial Bus (USB) input terminal, and / or (iv) a High Definition Multimedia Interface (HDMI) input terminal. Other examples, not shown in FIG. 1 C, include composite video.

[0122] In various embodiments, the input devices of block 172 have associated respective input processing elements as known in the art. For example, the RF portion can be associated with elements suitable for (i) selecting a desired frequency (also referred to as selecting a signal, or band-limiting a signal to a band of frequencies), (ii) downconverting the selected signal, (iii) band-limiting again to a narrower band of frequencies to select (for example) a signal frequency band which can be referred to as a channel in certain embodiments, (iv) demodulating the downconverted and band-limited signal, (v) performing error correction, and (vi) demultiplexing to select the desired stream of data packets. The RF portion of various embodiments includes one or more elements to perform these functions, for example, frequency selectors, signal selectors, bandlimiters, channel selectors, filters, downconverters, demodulators, error correctors, and demultiplexers. The RF portion can include a tuner that performs various of these functions, including, for example, downconverting the received signal to a lower frequency (for example, an intermediate frequency or a near-baseband frequency) or to baseband. In one set-top box embodiment, the RF portion and its associated input processing element receives an RF signal transmitted over a wired (for example, cable) medium, and performs frequency selection by filtering, downconverting, and filtering again to a desired frequency band. Various embodiments rearrange the order of the above-described (and other) elements, remove some of these elements, and / or add otherelements performing similar or different functions. Adding elements can include inserting elements in between existing elements, such as, for example, inserting amplifiers and an analog-to-digital converter. In various embodiments, the RF portion includes an antenna.

[0123] Additionally, the USB and / or HDMI terminals can include respective interface processors for connecting system 150 to other electronic devices across USB and / or HDMI connections. It is to be understood that various aspects of input processing, for example, Reed-Solomon error correction, can be implemented, for example, within a separate input processing IC or within processor 152 as necessary. Similarly, aspects of USB or HDMI interface processing can be implemented within separate interface ICs or within processor 152 as necessary. The demodulated, error corrected, and demultiplexed stream is provided to various processing elements, including, for example, processor 152, and encoder / decoder 156 operating in combination with the memory and storage elements to process the datastream as necessary for presentation on an output device.

[0124] Various elements of system 150 can be provided within an integrated housing, Within the integrated housing, the various elements can be interconnected and transmit data therebetween using suitable connection arrangement 174, for example, an internal bus as known in the art, including the I nter-IC (I2C) bus, wiring, and printed circuit boards.

[0125] The system 150 includes communication interface 160 that enables communication with other devices via communication channel 162. The communication interface 160 can include, but is not limited to, a transceiver configured to transmit and to receive data over communication channel 162. The communication interface 160 can include, but is not limited to, a modem or network card and the communication channel 162 can be implemented, for example, within a wired and / or a wireless medium.

[0126] Data is streamed, or otherwise provided, to the system 150, in various embodiments, using a wireless network such as a Wi-Fi network, for example IEEE 802.11 (IEEE refers to the Institute of Electrical and Electronics Engineers). The Wi-Fi signal of these embodiments is received over the communications channel 162 and the communications interface 160 which are adapted for Wi-Fi communications. The communications channel 162 of these embodiments is typically connected to an access point or router that provides access to external networks including the Internet for allowing streaming applications and other over-the-top communications. Other embodiments provide streamed data to the system 150 using a set-top box that delivers the data over the HDMI connection of the input block 172. Still other embodiments provide streamed data to the system 150 using the RF connection of the input block 172. As indicated above, various embodiments providedata in a non-streaming manner. Additionally, various embodiments use wireless networks other than Wi-Fi, for example a cellular network or a Bluetooth network.

[0127] The system 150 can provide an output signal to various output devices, including a display 176, speakers 178, and other peripheral devices 180. The display 176 of various embodiments includes one or more of, for example, a touchscreen display, an organic light-emitting diode (OLED) display, a curved display, and / or a foldable display. The display 176 can be for a television, a tablet, a laptop, a cell phone (mobile phone), or other device. The display 176 can also be integrated with other components (for example, as in a smart phone), or separate (for example, an external monitor for a laptop). The other peripheral devices 180 include, in various examples of embodiments, one or more of a stand-alone digital video disc (or digital versatile disc) (DVR, for both terms), a disk player, a stereo system, and / or a lighting system. Various embodiments use one or more peripheral devices 180 that provide a function based on the output of the system 150. For example, a disk player performs the function of playing the output of the system 150.

[0128] In various embodiments, control signals are communicated between the system 150 and the display 176, speakers 178, or other peripheral devices 180 using signaling such as AV.Link, Consumer Electronics Control (CEC), or other communications protocols that enable device-to-device control with or without user intervention. The output devices can be communicatively coupled to system 1000 via dedicated connections through respective interfaces 164, 166, and 168. Alternatively, the output devices can be connected to system 150 using the communications channel 162 via the communications interface 160. The display 176 and speakers 178 can be integrated in a single unit with the other components of system 150 in an electronic device such as, for example, a television. In various embodiments, the display interface 164 includes a display driver, such as, for example, a timing controller (T Con) chip.

[0129] The display 176 and speaker 178 can alternatively be separate from one or more of the other components, for example, if the RF portion of input 172 is part of a separate set-top box. In various embodiments in which the display 176 and speakers 178 are external components, the output signal can be provided via dedicated output connections, including, for example, HDMI ports, USB ports, or COMP outputs.

[0130] The system 150 may include one or more sensor devices 168. Examples of sensor devices that may be used include one or more GPS sensors, gyroscopic sensors, accelerometers, light sensors, cameras, depth cameras, microphones, and / or magnetometers. Such sensors may be used to determine information such as user’s position and orientation. Where the system 150 is used as the control module for an extended realitydisplay (such as control modules 124, 132), the user’s position and orientation may be used in determining how to render image data such that the user perceives the correct portion of a virtual object or virtual scene from the correct point of view. In the case of head-mounted display devices, the position and orientation of the device itself may be used to determine the position and orientation of the user for the purpose of rendering virtual content. In the case of other display devices, such as a phone, a tablet, a computer monitor, or a television, other inputs may be used to determine the position and orientation of the user for the purpose of rendering content. For example, a user may select and / or adjust a desired viewpoint and / or viewing direction with the use of a touch screen, keypad or keyboard, trackball, joystick, or other input. Where the display device has sensors such as accelerometers and / or gyroscopes, the viewpoint and orientation used for the purpose of rendering content may be selected and / or adjusted based on motion of the display device.

[0131] The embodiments can be carried out by computer software implemented by the processor 152 or by hardware, or by a combination of hardware and software. As a non-limiting example, the embodiments can be implemented by one or more integrated circuits. The memory 154 can be of any type appropriate to the technical environment and can be implemented using any appropriate data storage technology, such as optical memory devices, magnetic memory devices, semiconductor-based memory devices, fixed memory, and removable memory, as non-limiting examples. The processor 152 can be of any type appropriate to the technical environment, and can encompass one or more of microprocessors, general purpose computers, special purpose computers, and processors based on a multi-core architecture, as non-limiting examples.Block-Based Video Coding

[0132] Like HEVC, the WC is built upon the block-based hybrid video coding framework. FIG. 2A gives the block diagram of a block-based hybrid video encoding system 200. Variations of this encoder 200 are contemplated, but the encoder 200 is described below for purposes of clarity without describing all expected variations.

[0133] Before being encoded, a video sequence may go through pre-encoding processing (204), for example, applying a color transform to an input color picture (e.g., conversion from RGB 4:4:4 to YCbCr 4:2:0), or performing a remapping of the input picture components in order to get a signal distribution more resilient to compression (for instance using a histogram equalization of one of the color components). Metadata can be associated with the pre-processing and attached to the bitstream.

[0134] The input video signal 202 including a picture to be encoded is partitioned (206) and processed block by block in units of, for example, CUs. Different CUs may have different sizes. In VTM-1.0, a CU can be up to 128x128 pixels. However, different from the HEVC which partitions blocks only based on quad-trees, in the VTM- 1.0, a coding tree unit (CTU) is split into CUs to adapt to varying local characteristics based on quad / binary / ternary-tree. Additionally, the concept of multiple partition unit type in the HEVC is removed, such that the separation of CU, prediction unit (PU) and transform unit (TU) does not exist in the VVC-1.0 anymore; instead, each CU is always used as the basic unit for both prediction and transform without further partitions. In the multi-type tree structure, a CTU is firstly partitioned by a quad-tree structure. Then, each quad-tree leaf node can be further partitioned by a binary and ternary tree structure. Different splitting types may be used, such as quaternary partitioning, vertical binary partitioning, horizontal binary partitioning, vertical ternary partitioning, and horizontal ternary partitioning.

[0135] In the encoder of FIG. 2A, spatial prediction (208) and / or temporal prediction (210) may be performed. Spatial prediction (or “intra prediction”) uses pixels from the samples of already coded neighboring blocks (which are called reference samples) in the same video picture / slice to predict the current video block. Spatial prediction reduces spatial redundancy inherent in the video signal. Temporal prediction (also referred to as “inter prediction” or “motion compensated prediction”) uses reconstructed pixels from the already coded video pictures to predict the current video block. Temporal prediction reduces temporal redundancy inherent in the video signal. A temporal prediction signal for a given CU may be signaled by one or more motion vectors (MVs) which indicate the amount and the direction of motion between the current CU and its temporal reference. Also, if multiple reference pictures are supported, a reference picture index may additionally be sent, which is used to identify from which reference picture in the reference picture store (212) the temporal prediction signal comes.

[0136] The mode decision block (214) in the encoder chooses the best prediction mode, for example based on a rate-distortion optimization method. This selection may be made after spatial and / or temporal prediction is performed. The intra / inter decision may be indicated by, for example, a prediction mode flag. The prediction block is subtracted from the current video block (216) to generate a prediction residual. The prediction residual is de-correlated using transform (218) and quantized (220). (For some blocks, the encoder may bypass both transform and quantization, in which case the residual may be coded directly without the application of the transform or quantization processes.) The quantized residual coefficients are inverse quantized (222) and inverse transformed (224) to form the reconstructed residual, which is then added back to the prediction block (226) to form the reconstructed signal of the CU. Further in-loop filtering, such as deblocking / SAO (SampleAdaptive Offset) filtering, may be applied (228) on the reconstructed CD to reduce encoding artifacts before it is put in the reference picture store (212) and used to code future video blocks. To form the output video bit-stream 230, coding mode (inter or intra), prediction mode information, motion information, and quantized residual coefficients are all sent to the entropy coding unit (108) to be further compressed and packed to form the bitstream.

[0137] FIG. 2B gives a block diagram of a block-based video decoder 250. In the decoder 250, a bitstream is decoded by the decoder elements as described below. Video decoder 250 generally performs a decoding pass reciprocal to the encoding pass as described in FIG. 2A. The encoder 200 also generally performs video decoding as part of encoding video data.

[0138] In particular, the input of the decoder includes a video bitstream 252, which can be generated by video encoder 200. The video bit-stream 252 is first unpacked and entropy decoded at entropy decoding unit 254 to obtain transform coefficients, motion vectors, and other coded information. Picture partition information indicates how the picture is partitioned. The decoder may therefore divide (256) the picture according to the decoded picture partitioning information. The coding mode and prediction information are sent to either the spatial prediction unit 258 (if intra coded) or the temporal prediction unit 260 (if inter coded) to form the prediction block. The residual transform coefficients are sent to inverse quantization unit 262 and inverse transform unit 264 to reconstruct the residual block. The prediction block and the residual block are then added together at 266 to generate the reconstructed block. The reconstructed block may further go through in-loop filtering 268 before it is stored in reference picture store 270 for use in predicting future video blocks.

[0139] The decoded picture 272 may further go through post-decoding processing (274), for example, an inverse color transform (e.g. conversion from YCbCr 4:2:0 to RGB 4:4:4) or an inverse remapping performing the inverse of the remapping process performed in the pre-encoding processing (204). The post-decoding processing can use metadata derived in the pre-encoding processing and signaled in the bitstream. The decoded, processed video may be sent to a display device 276. The display device 276 may be a separate device from the decoder 250, or the decoder 250 and the display device 276 may be components of the same device.

[0140] Various methods and other aspects described in this disclosure can be used to modify modules of a video encoder 200 or decoder 250. Moreover, the systems and methods disclosed herein are not limited to WC or HEVC, and can be applied, for example, to other standards and recommendations, whether pre-existing or future-developed, and extensions of any such standards and recommendations (including WC and HEVC).Unless indicated otherwise, or technically precluded, the aspects described in this disclosure can be used individually or in combination.

[0141] FIG. 3A is a schematic side view illustrating an example waveguide display that may be used with extended reality (XR) applications according to some embodiments. An image is projected by an image generator 302. The image generator 302 may use one or more of various techniques for projecting an image. For example, the image generator 302 may be a laser beam scanning (LBS) projector, a liquid crystal display (LCD), a lightemitting diode (LED) display (including an organic LED (OLED) or micro LED (pLED) display), a digital light processor (DLP), a liquid crystal on silicon (LCoS) display, or other type of image generator or light engine.

[0142] Light representing an image 312 generated by the image generator 302 is coupled into a waveguide 304 by a diffractive in-coupler 306. The in-coupler 306 diffracts the light representing the image 312 into one or more diffractive orders. For example, light ray 308, which is one of the light rays representing a portion of the bottom of the image, is diffracted by the in-coupler 306, and one of the diffracted orders 310 (e.g. the second order) is at an angle that is capable of being propagated through the waveguide 304 by total internal reflection. The image generator 302 displays images as directed by a control module 324, which operates to render image data, video data, point cloud data, or other displayable data.

[0143] At least a portion of the light 310 that has been coupled into the waveguide 304 by the diffractive incoupler 306 is coupled out of the waveguide by a diffractive out-coupler 314. At least some of the light coupled out of the waveguide 304 replicates the incident angle of light coupled into the waveguide. For example, in the illustration, out-coupled light rays 316a, 316b, and 316c replicate the angle of the in-coupled light ray 308. Because light exiting the out-coupler replicates the directions of light that entered the in-coupler, the waveguide substantially replicates the original image 312. A user’s eye 318 can focus on the replicated image.

[0144] In the example of FIG. 3A, the out-coupler 314 out-couples only a portion of the light with each reflection allowing a single input beam (such as beam 308) to generate multiple parallel output beams (such as beams 316a, 316b, and 316c). In this way, at least some of the light originating from each portion of the image is likely to reach the user’s eye even if the eye is not perfectly aligned with the center of the out-coupler. For example, if the eye 318 were to move downward, beam 316c may enter the eye even if beams 316a and 316b do not, so the user can still perceive the bottom of the image 312 despite the shift in position. The out-coupler 314 thus operates in part as an exit pupil expander in the vertical direction. The waveguide may also include one or more additional exit pupil expanders (not shown in FIG. 3A) to expand the exit pupil in the horizontal direction.

[0145] In some embodiments, the waveguide 304 is at least partly transparent with respect to light originating outside the waveguide display. For example, at least some of the light 320 from real-world objects (such as object 322) traverses the waveguide 304, allowing the user to see the real-world objects while using the waveguide display. As light 320 from real-world objects also goes through the diffraction grating 314, there will be multiple diffraction orders and hence multiple images. To minimize the visibility of multiple images, it is desirable for the diffraction order zero (no deviation by 314) to have a great diffraction efficiency for light 320 and order zero, while higher diffraction orders are lower in energy. Thus, in addition to expanding and out-coupling the virtual image, the out-coupler 314 is preferably configured to let through the zero order of the real image. In such embodiments, images displayed by the waveguide display may appear to be superimposed on the real world.

[0146] FIG. 3B is a schematic side view illustrating an example alternative display type that may be used with extended reality applications according to some embodiments. In an XR head-mounted display device 330, a control module 332 controls a display 334, which may be an LCD, to display an image. The head-mounted display includes a partly-reflective surface 336 that reflects (and in some embodiments, both reflects and focuses) the image displayed on the LCD to make the image visible to the user. The partly-reflective surface 336 also allows the passage of at least some exterior light, permitting the user to see their surroundings.

[0147] FIG. 3C is a schematic side view illustrating an example alternative display type that may be used with extended reality applications according to some embodiments. In an XR head-mounted display device 340, a control module 342 controls a display 344, which may be an LCD, to display an image. The image is focused by one or more lenses of display optics 346 to make the image visible to the user. In the example of FIG. 3C, exterior light does not reach the user’s eyes directly. However, in some such embodiments, an exterior camera 348 may be used to capture images of the exterior environment and display such images on the display 344 together with any virtual content that may also be displayed.

[0148] The embodiments described herein are not limited to any particular type or structure of XR display device.

[0149] This application relates to learning-based image compression and processing systems, among other things. The field of image compression and processing aims to develop tools for compression, analysis, interpolation, representation, and understanding of input signals, such as image or video.

[0150] Video and image data are universal data formats across several business domains, such as autonomous driving, robotics, civil engineering, video streaming, teleconferencing, computer vision, and the animation / movie industry. These universal data are believed to consume a large portion of network traffic, e.g., among connected cars over 5G network, and video communications teleconferencing. Efficient representation formats are necessary for image and video understanding and communication. Raw data needs to be properly organized and processed for the purposes of, e.g., world modeling, sensing, and rendering. Compression on these raw input signals is essential when storage and transmission of the data are required in the related scenarios.

[0151] 2D image and video data include discrete samples on visible objects and background from a capturing scene. To fully represent the real world with point samples, in practice, 3D point cloud data requires a huge number of pixels.

[0152] For instance, 8K resolution is frequently used and the corresponding 8K format, such as UHDTV2 consists of more than 30 million pixels for only one frame. The processing of such large-scale input signals is computationally expensive, especially for consumer devices, such as smartphones, tablets, and automotive navigation systems, which have limited computational power.

[0153] The processing or inference of such an input signal tends to need an efficient storage methodology. To store and process the input signal with affordable computational cost, the input signal may be down-sampled first for some embodiments, in which the down-sampled signal summarizes the color of the input signal while having much fewer pixels. The down-sampled signal is inputted into a subsequent machine task for further processing. An efficient feature extractor may be used to improve the quality of packing data within the given resource budget.

[0154] Due to a recent evolution of computation hardware, the article, Bengio, Yoshua, et al., Representation Learning: A Review and New Perspectives, ARXI PREPRINT, arXiv: 1206.5538 (2012), mentions how an autoencoder has become fusible with the traditional image compression framework such as a linear transform coding system as described in JPEG ITU-R Rec. T.81 & ISO / IEC 10918-1: Digital Compression and Coding of Continuous Tone Still Images, ITU-R (1992), which consists of transformation analysis, entropy coding, and transformation synthesis modules. Regarding transformation analysis, a Discrete Cosine Transform (DCT) function transforms image data from the signal domain to the transformation domain. A DCT is followed by a quantizer, which makes the transformed coefficients discrete before inputting them into an entropy coder. Theset of entropy decoded coefficients is re-synthesized via an inverse DCT function to reconstruct the compressed image back into the signal domain.

[0155] Based on the above traditional image compression framework, recent challenges for learning-based image compression are efficiently replacing a DCT and inverse DCT blocks with artificial neural networks, such as convolutional neural networks (CNN or CNNs). Such neural networks approximate functions that transfer the input signals from the data domain to the latent domain or vice versa as was done with an early proposal of the autoencoder. Another challenge for learning-based compression is making the entropy coder differentiable so that an entire compression system becomes end-to-end learnable.

[0156] Unlike DCT, an approximated function from artificial neural network frequently embeds input signals as a form of feature vectors in latent space. A latent space is a mathematical space in which the features extracted from input signals are condensed to vector form. This space usually has a lower dimensionality than the extracted feature space. So, for compression purposes, such a conversion may be done for some embodiments. The elements in a latent space vector may be inferred via a trained neural network based encoder. In the downsample case, a block of input signals is condensed to a single feature vector, in which the collection of the feature vectors representing the entire downsampled input is referred to as a feature map. This compressed feature map may be passed through a set of neural network layers and an entropy bottleneck and then decompressed through a decoding architecture. An entropy bottleneck may include an entropy encoder, bitstream, and entropy decoder. For some embodiments, an entropy bottleneck is an entropy coder. Considering this pipeline, one of the main challenges of learning-based compression is generating a highly abstracted feature map that contains the finest details of the given input signal and that is decodable through a carefully designed neural network architecture.

[0157] The importance of generating efficient feature maps in learning-based compression system in many applications is described above. One possible way to meet this criterion is to design neural network layers that preserve the detailed property of the compressing point samples. However, the varying rate vs. distortion optimization targets or complex architecture designed for multiple purposes, makes preserving full details within a computed feature map difficult. Benefits may be derived from providing a treatment or rectification process directly onto the feature map for the purpose of preserving and even enhancing the feature representation for better compression.

[0158] Recent works in end-to-end learning-based image compression have demonstrated significant coding gains by analyzing probability distribution within the entropy coding process. For example, the article, Balle, Johannes, et al., Variational Image Compression with a Scale Hyperprior, INTERN! CONF, ON LEARNING REPRESENTATIONS (2018) (“Balle"), described changes to the entropy model of the autoencoder in which side information is sent in an additional bitstream. This hyperprior network performs scale capturing on the quantized output of the encoded feature map. The estimated scale is later applied in both arithmetic encoder and decoder to boost performance of the entropy model.

[0159] Based on the observation of the related works in learning-based compression, adjustment of the feature distribution is a challenging process. Since the feature map passes in-between most of the neural network layers and since the encoding and decoding times are important in a compression framework, the distribution adjustment may be kept simple, unsupervised, and easily modularizable for some embodiments. Moreover, the distribution adjustment may emphasize feature elements differently per an input signal. Therefore, a neural network-based distribution parameters learning framework may approximate a more suitable nonlinear function for the compression.

[0160] A typical learning-based image compression system often generates a feature map as a representation of the input signal. The compression of the input signal is converted to the entropy coding of the feature map. The efficiency of entropy coding tends to heavily depend on the distribution of the feature map. Such a tradeoff may be beneficial for the entropy coding when updating the feature distribution based on a predetermined and fixed distribution parameter set. The coding benefits may be limited because the manipulation of feature distribution is unable to be adapted to the input signal. Therefore, a learnable approach may adaptively update the feature distribution to improve the coding efficiency of an input signal, such as image or video data.

[0161] A compression framework uses a well-balanced trade-off between the accuracy of reconstruction and the computational cost. A distribution learning network is a flexible and extendible micro-architecture that may be applied to various end-to-end learning-based image compression systems to improve the coding gain of existing compression frameworks.

[0162] The architecture learns sets of distribution parameters from an input feature map via neural network layers. These distribution parameters may be applied to further preserve or even enhance the level of abstraction of details for the feature maps going through the image compression system. A distribution learning network (DLN) may be further enhanced by combining the DLN with other feature extractors, such as residual networksor transformer-based blocks. A distribution learning network also may be extended to a multi-stage architecture for some embodiments.

[0163] Distribution analysis and adjustment on a feature map are processes that may be used to improve coding performance of a learning-based image compression system. As an example, the hyperprior model in Balle conditioned gaussian scale parameters of the hyperprior for enhancing the factorized entropy model. The compressed hyperprior is added to the bitstream as side information, which allows the decoder to use a conditional entropy model. The entropy model became image-dependent, spatially adaptive, and more accurate.

[0164] An autoregressive model with hyperprior extended the above hyperprior entropy model by applying a gaussian mixture model (GMM) and adding an autoregressive component between the autoencoder and hyperprior networks according to article Minnen, David, et al., Joint Autoregressive and Hierarchical Priors for Learned Image Compression, ADVANCES IN NEURAL INFO MATION PROCESSING SYSTEMS (2018) (“Minnen”). This model learns a probabilistic model over quantized latent used for entropy coding. The entropy parameters network generates the mean and scale parameters for a conditional gaussian entropy model. The parameters are applied to the bottleneck of the autoencoder structure. For some embodiments, the bottleneck is a latent space vector. An autoencoder does not have entropy encoder / decoder in the bottleneck but has a condensed feature instead of bitstream. However, in practice, this autoregressive model is less desirable than hierarchical models since the autoregressive models are inherently serial, and therefore cannot be sped up using techniques such as parallelization. Due to the differentiable design and proven accuracy, the factorized and hyperprior models become base entropy models for many learning-based image compression systems. One such design, which is mentioned in article Cheng, Zhengxue, et al., Learned Image Compression with Discretized Gaussian Mixture Likelihoods and Attention Modules, PROC, OF THE IEEE / CVF CONF, ON COMP. VISION AND PATTERN RECOGNITION 7939-7948 (2020), parameterizes the distributions of latent codes for a more accurate entropy model. Another article, Hu, Zhihao, et al., Coarse-to-Fine Deep Video Coding with Hyperprior-Guided Mode Prediction, PROC, OF THE IEEE / CVF CONF, ON COMP. VISION AND PATTERN RECOGNITION 5921-5930 (2022), applied hyperprior model for motion compression of the video coding and improved performance of the motion compensation without increasing bit cost. The above feature distribution adjustments are mainly applicable to the entropy bottleneck part of the compression system. A certain feature adjustment needs to be processed throughout the entire pipeline. For this purpose, a generalized divisive normalization (GDN) operation is introduced in article Balle, Johannes, et al., Density Modeling of Images Using a Generalized Normalization Transformation, ARXIV PREPRINT arXiv: 1511.06281 (2015). This biologically inspired GDN aims at reducingredundancies between channels, which makes it suitable for compression of images. The operation is also invertible, hence a GDN can be paired with its inverse operation (IGDN) to favor the analysis-by-synthesis framework. Along with convolution and sub-sampling layers, an end-to-end image compression framework, which was discussed in article Balle, Johannes, et al., End-to-End Optimized Image Compression, ARXIV PREPRINT arXiv:1611 .01704 (2016), has been used in GDN / IGDN operations. However, the learnt distribution parameters are rather complex and lack flexibility due to the presence of an inverse module. A more light-weight yet efficient distribution adjustment module is necessary for exploiting further coding gains in a learning-based image compression system. Using dedicated neural network blocks for each set of distribution parameters may be beneficial, for some embodiments, for adaptively dealing with different types of image data.

[0165] This application introduces, in accordance with some embodiments, a feature extractor that extracts a feature from a sample point, using an example neural network micro-architecture. The example neural network micro-architecture, in accordance with some embodiments, further enhances the abstraction detail of a feature map (FM) passing through a learning-based compression framework. This example architecture, in accordance with some embodiments, takes any features as inputs and manipulates each feature element by applying multiple sets of learnt distribution parameters. For some embodiments, the feature elements are then made more discriminative for better reconstruction. This example architecture may be plugged in between neural network blocks in one or more locations and may be applied in both encoder and decoder architectures.

[0166] A distribution learning network is first introduced, in accordance with some embodiments, with further details on the manipulation of feature-elements via normalization and transformation processes. Both normalization and transformation processes may learn their own sets of distribution parameters based on their neural network (NN) layers. For some embodiments, FM transformation is a major process. For some embodiments, an FM normalization process may precede transformation to further enhance the feature map.

[0167] An extension with down-sampling and filtering followed by a multi-stage architecture is introduced in accordance with some embodiments. Further evolved distribution learning architectures are introduced in accordance with some embodiments by combining with other feature extracting NN layers such as residual network, IRN, and transformer-based methods. Article He, Kaiming, et al., Deep Residual Learning for Image Recognition, PROC, OF IEEE / CVF CONF, ON COMP. VISION AND PATTERN RECOGNITION, 770-778 (2016) (“He”) describes a residual network. Article Szegedy, Christian, et al., Inception-v4, Inception-Resnet and the Impact of Residual Connections on Learning, PROC, OF AAAI CONF, ON ARTIFICIAL INTELLIGENCE, vol. 31, no. 1 (2017) (“Szegedy”) describes IRN. Articles Dosovitskiy, Alexey, et al., An Image is Worth 16x16 Words: Transformersfor Image Recognition at Scale, ARXI PREPRINT arXiv: 2010.11929 (2020); Liu, Ze, et al., Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows, Proc, of the IEEE / CVF Intern’l Conf, on Comp. Vision 10012-10022 (2021); and Si, Chenyang, et al., Inception Transformer, ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 35 23495-23509 (2022) describe voxel transformer-based methods.Distribution Learning Network

[0168] FIG. 4 is a schematic illustration showing example feature vectors and feature map in image space according to some embodiments. In accordance with some embodiments, a neural network layer generates a feature vector 402, 406, 410 for each output point 404, 408, 412, where the feature vector lives in a C-dimensional (or C-channel) space 416. Each of the channels is a reshaped vector, and each of the elements in a reshaped vector is a feature element. The set of the feature vectors is referenced as a feature map 414. Example feature maps 414 in image space 400 and in 3D space are illustrated in FIG. 4. As shown in the example of FIG. 4, a feature map 414 may correspond to n points spread in 3D. A quantity of C feature elements 416 may be associated with each of the n points. As such, a C x n feature map (FM) 414 may correspond to this example scenario.

[0169] In some applications, one of the main purposes of a feature extraction network is to learn multiple sets of distribution parameters on a feature map and apply them to adjust the elements of the feature map for image or video compression. This operation contrasts with supervised classification tasks, which homogenize features within the same semantic classes. The present application, in accordance with some embodiments, aims to learn distinct features that are descriptive for the given signal reconstruction by inserting NN-based adjustment module(s) within a learning-based compression framework. The example feature adjustment block, in accordance with some embodiments, may be trained as part of an end-to-end compression system based on a geometric error as the loss function (e.g., chamfer distance or cross entropy). For some embodiments, feature adjustment may be implemented in two steps: element normalization and transformation, in which each step is designed with its own NN layers.Normalizing Feature Elements via Neural Network Layers

[0170] FIG. 5 is a schematic illustration showing an example distribution parameter corresponding to a reshaped vector according to some embodiments.

[0171] Normalization is known as a training technique that can benefit both convergence speed and accuracy.A feature map (FM) may be normalized for a more reliable architecture design of a neural network. For someembodiments, each feature vector may be normalized independently such that the norms (lengths) of the feature vectors are identical (all equal to a predefined value, e.g., 1).

[0172] For some embodiments, as illustrated in the breakdown 500 shown in FIG. 5, all feature elements in the FM 502 from a certain feature channel are put into a reshaped vector 508. There are a total of “C” reshaped vectors 504. Each reshaped vector is normalized independently by applying an estimated distribution of parameters 510 for each channel. For some embodiments, an estimated mean nc, which is shown in FIG. 5, is removed first from the reshaped vector and then a normalization is performed based on an estimated standard deviation a. For example, suppose the shaped vector in FIG. 5 has n elements. The parameter / cwould be the mean value of the n elements 506. FIG 5 may be seen, for some embodiments, as a conceptual diagram for explaining the reshaped vector.

[0173] FIG. 6 is a functional block diagram showing example neural network layers for learning feature map distribution parameters according to some embodiments. A set of distribution parameters 608, such as mean [i, may be learnt via a set of neural network (NN) layers 604, an example process 600 of which is shown in FIG. 6. For some embodiments, the parameter . (or others) may be either a set of parameters that each parameter normalizes its corresponding vector in the FM 602, or a single parameter that represents the entire FM. A pooling operation 606 (e.g., average pooling) may define the final form of the estimating parameter(s). Similarly, for some embodiments, the standard deviation a may be learnt either per channel ac(c = 1 ... C) or per feature map 602 by pooling the entire FM. For some embodiments, z or o may be computed directly from the initial FM elements without passing the initial FM elements through a set of NN layers. As shown in FIG. 6, the neural network (NN)-layers 604 may be located in-between the original feature map 602 and pooling 606 to enhance the FM before computing or estimating the distribution parameters 608, such as [i.

[0174] FIG. 7 is a schematic illustration showing an example of three convolutional neural network layers according to some embodiments. FIG. 7 shows an example architecture 700 for a set of neural network (NN) layers. The output of the NN layers may be pooled to estimate a set of distribution parameters. The pooling operation may be average or max operations. The range of pooling operation may be the entire FM, or perchannel basis, and etc. An original feature map, FM±(702), is inputted into an input CNN layer 706. The original feature map 702 is transformed through the input (first) CNN layer 706 to become a second feature map, FM2(708). The second feature map 708 goes through the hidden (second) CNN layer 712 to become a third feature map, FM3(714). The third feature map 714 goes through the output (third) CNN layer 718 to become a fourth feature map, FM4(720). This CNN architecture 700 learns how to refine a feature map before pooling to a setof distribution parameters. The type of distribution parameters may be defined by the network design to be shown in FIGs. 8, 10, 11 , or 12. During the training phase, each “NN Layers” learns how to better estimate distribution parameters, such as mean, standard deviation, scale, and bias, by adjusting an input feature map (FM 702 to an output feature map (FM4) 720. For some embodiments, the depth of the input feature map may be expanded to three feature maps, one for each R (red), G (green), and B (blue) color.

[0175] FIG. 8 is a functional block diagram showing a first example feature map normalization process according to some embodiments. FIG. 8 shows an example of a learning-based FM normalization process 800. An initial FM 802 is inputted into an “NN Layers 1” block 804 in which a set of mean parameters 806 are estimated. The estimated mean parameters ( / z) 806 are expanded back into a feature map and subtracted off the original feature map. Expanding back is a process 808 in which a parameter value, e.g., .c, is copied towards the c-channel axis. The centered feature map, FMC810, is passed to a block 812 to estimate a. The estimated standard deviation parameters (a) 814 are expanded 816 back into a feature map. The mean parameters 806 and standard deviation parameters 814 are examples of distribution parameters. The centered feature map, FMC810, is divided by the estimated standard deviation parameters to generate the normalized feature map, FMn818. For each of the columns in the centered feature map matrix, FMC810, each element in that particular column is divided by the element of the estimated standard deviation parameters array ( ) corresponding to that particular column for some embodiments.

[0176] Equation 1 shows a normalization equation:

[0177] As shown in FIG. 8, the ® operator translates the center of each c-th vector element to zero by an element-wise addition (which is a subtraction as shown in FIG. 8) of the FM and expanded mean parameters to generate an estimated re-centered feature map, FMC. The ® operator scales the elements by an element-wise multiplication of FMCand an expanded then inverted standard deviation parameters, to generate the normalized feature map, FMn. For some embodiments, expansion is done first, inversion is done second, and element-wise multiplication is done third. For some embodiments, an epsilon e may be added, as shown in Eq. 1, to the estimated standard deviation distribution parameters as a very small float value to avoid the possibility of zero division.

[0178] FIG. 9 is a schematic illustration showing an example expansion to a feature map dimension according to some embodiments. For some embodiments, the “Expansion to FM” blocks 808, 816 of FIG. 8 operate as theexample process 900 shown in FIG. 9. The estimated set of mean parameters is inputted into an expansion to FM dimension process 910 to match the dimension of the original FM. For some embodiments, the distribution parameters array 908 may be repeated for each of the n rows 906 of the FM matrix to increase the dimension of the distribution parameters array 908 to match the FM matrix. For some embodiments, a matrix of expanded distribution parameters 902 may have dimensions of C-dimensions 904 by n-pixels 906.

[0179] FIG. 10 is a functional block diagram showing a second example feature map normalization process according to some embodiments. For some embodiments, as illustrated in the example process 1000 in FIG. 10, a set of distribution parameters a 1014 may be learnt by an “NN Layers 2” block 1012. For some embodiments (not shown in FIG. 10), the input of the “NN Layers 2” block may be the initial FM instead of the centered FMC.

[0180] An initial FM 1002 is inputted into an “NN Layers 1” block 1004 in which a set of mean distribution parameters 1006 are estimated. The estimated mean distribution parameters ( / J.) 1006 are expanded 1008 back into a feature map and subtracted off the original feature map. The centered feature map, FMC1010, is passed to the “NN Layers 2” block 1012 to estimate o 1014. The estimated standard deviation distribution parameters (cr) 1014 are expanded 1016 back into a feature map. The centered feature map, FMC1010, is divided by the estimated standard deviation distribution parameters to generate the normalized feature map, FMn1018. For each of the columns in the centered feature map matrix, FMC, each element in that particular column is divided by the element of the estimated standard deviation distribution parameters array (a ) corresponding to that particular column for some embodiments. For some embodiments, an epsilon e may be added to the estimated standard deviation distribution parameters as a very small float value to avoid the possibility of zero division.Transforming Feature Elements via Neural Network Layers

[0181] Transforming vector elements in a feature map (FM) by its distribution parameter(s) may emphasize the property of the reconstructing points embedded in the FM. Such a channel-wise (or pixel-wise) independent vector elements transform, may efficiently adjust the expressivity of the abstracted FM by optimizing the proposed distribution learning model through an end-to-end training process. An independent parameter estimation per vector further discriminates the feature representation by breaking the rigidity of the deformation within the FM.

[0182] FIG. 11 is a functional block diagram showing a first example feature map transformation process according to some embodiments. A transformation may scale and translate the entire set of elements withparameters y and p. For some embodiments, the parameters may be predetermined. Different pairs of scale and translation values may be applied separately per vector in a feature map. A feature vector may be chosen for this separation, but, as depicted in FIG. 5, a feature vector may be reshaped into a per-channel vector. The transformation parameters ycand pc, which may be an element of the distribution parameters y and p , respectively, are defined and applied independently for each reshaped vector (channel-wise). These neural network layers, “NN Layers 3” 1104 and “NN Layers 4” 1112, may be trained by an end-to-end image compression framework. During an inference, an optimal set of distribution parameters y and p may be estimated from the given input feature map FMn1102.

[0183] Form some embodiments, such as when transforming an entire FM at once, y and p are used to perform a linear transformation of the feature map. For example, suppose a 3D object is in a virtual scene. The parameters y and p scale and translate this 3D object to another place. The 3D object’s point coordinates may be represented as a feature map (FM) after a feature extraction process. The parameters y and p also may be applied to the FM. To adjust the distribution of a feature map more precisely the parameters y and p may be learnt / estimated on a per channel basis. In this case, each parameter y and p becomes a vectorized parameter and the transformation becomes non-linear with more degree of freedom to deform the FM.

[0184] For some embodiments of an example process 1100, given a normalized FMn1102 as an input, each set of per channel distribution parameters 1106, 1114, which forms a vector in the C dimension, is learnt by a set of neural network (NN) layers 1104, 1112. These sets of NN layers 1104, 1112, which may learn a core set of distribution parameters 1106, 1114, may operate similar to the description given for FIG. 7.

[0185] Using these sets of NN layers 1104, 1112, two sets of distribution parameters y (1106) and p (1114) are learnt via their own “NN Layers 3” block 1104 and “NN Layers 4” block 1112, respectively as shown in FIG. 11. Each set of parameters y and p go through an “Expansion to FM” block 1108, 1116 to match the FM dimensions and prepare for their respective 0 and © operations. This transformation learning architecture via NN layers is based on the transformation equation 2:FMt= (FMnExpand(y)) ® Expand( ?) (2)

[0186] As shown in FIG. 11 , the 0 operator scales the elements by an element-wise multiplication of FMnwith an expanded set of distribution parameters y 1106 to generate the scaled feature map, FMS1110. The scaled feature map, FMS1110, generated by the 0 operation, and FMn1102 is used as the input to the “NN Layers 4” block 1112. The © operator translates each element in a particular column of the scaled feature map,FMS1110, by the expanded element of the set of distribution parameters / ? 1114 that corresponds to that particular column. This © operator generates the transformed feature map, FMt1118.

[0187] FIG. 12 is a functional block diagram showing a second example feature map transformation process according to some embodiments. For some embodiments, the “NN Layers 3” block 1204, the “NN Layers 4” block 1212, and the “Expansion to FM” blocks 1208, 1216 operate as described earlier.

[0188] For some embodiments, as depicted in FIG. 12, the FM input 1202 to the “NN Layers 4” 1212 is parallel. Instead of using the scaled feature map, FMS1210, as an input to the “NN Layers 4” block 1212, the normalized feature map, FMn1202, is used directly as an input to the “NN Layers 4” 1212. This variation of the input FM may improve the training of the NN layers by independently optimizing the set of distribution parameters ft.

[0189] For some embodiments of the example process 1200, a © operator adds the scaled feature map, FMS1210, to the output of the expansion to FM process 1216 to generate the transformed feature map, FMt1218.

[0190] For some embodiments, a single transformation block may be used without a normalization process. In this case, the initial FM is inputted directly into a “NN Layers 3” and / or “NN Layers 4” block instead of inputting the normalized feature map, FMn.

[0191] For some embodiments, distribution parameters may be shared by: (1) processes within the encoder; (2) processes within the decoder; and / or (3) a process in the encoder and a process in the decoder and vice versa.Extension Towards a Multi-Stage Architecture

[0192] A multi-stage architecture may further refine a feature map computed from the above distribution learning network. To deeply and smoothly communicate between stages, additional functionalities, such as downsampling or filtering, may follow the feature map transformation.Downsampling Feature Map

[0193] FIG. 13 is a functional block diagram showing an example downsampling of a feature map according to some embodiments. Pixel-based CNN layers may downsample an image by a later max or average pooling operation. The degree of downsampling is further adjustable by the size of the local group in image (usually via kernel size). In FIG 13, there are m input feature maps. By controlling this number m, the size of the local group may be adjusted. If m is larger, then the size of the local group gets bigger.

[0194] In the example process 1300 of FIG. 13, an example set of m feature maps 1302, 1304, 1306 is processed by a shared “Distribution Learning Network” block 1308, 1310, 1312. For some embodiments, a DLN block, e.g, may be a combination of the process shown in either FIG. 7 or FIG. 9 plus the process shown in either FIG. 10 or FIG. 11 . Each FM 1302, 1304, 1306 of a local group pixels is transformed via the distribution learning network 1308, 1310, 1312 into a transformed feature map 1314, 1316, 1318. The transformed feature maps 1314, 1316, 1318 are each pooled 1320, 1322, 1324 to a feature vector, FV 1326, 1328, 1330. In this manner, the features aggregate progressively across multi-stage layers. For some embodiments, the pooling operation, for each of the C columns of an n x C transformed feature map FMt, may, e.g., determine an average of the elements in a particular column or may, e.g., select the maximum value among the elements in a particular column.Filtering Feature Map

[0195] FIG. 14 is a functional block diagram showing an example filtering of a feature map according to some embodiments. For some embodiments, the size of pixels and / or feature dimensions is kept constant across different stages of a multi-stage design. This structure may improve the compatibility and allow the abovedescribed methods to be plugged between layers of any end-to-end compression framework. For some embodiments, as illustrated in the example structure 1400 in FIG. 14, a downsampled feature vector (FV) 1406 is expanded to the original number of pixels by copying the same vector for each expanded pixel. For some embodiments, downsampling may occur via a pool process 1404. The feature map (FM) 1402 before downsampling is concatenated 1410 to the expanded FV 1408. The expansion of the feature vector (FV) to match the size of the original feature map is done before concatenation 1410. For some embodiments, an additional feed forward network 1412 may be connected to generate a filtered feature map 1414 after the concatenation 1410. This additional neural network layer also matches the in-out dimension of the FM. This process is depicted in FIG. 14. For some embodiments, a feed forward network block 1412 may be a neural network in which information flows in one direction. A feed forward network 1412 may be, for example, an MLP or a CNN. For some embodiments, the purpose of a feed forward network 1412 is to refine a concatenated FM to smoothly adapt to the next block in the process.

[0196] Based on these downsampling and filtering functions, the distribution learning network may be connected multiple times with numerous options in accordance with some embodiments.

[0197] For some embodiments, the first row of FIG. 13 may be combined with FIG. 14. The first feature map of FIG. 13, FM1, may be concatenated with an expanded first feature vector of FIG. 13, FV1. See the expanded FV in FIG. 14. This expanded and concatenated feature map, FMconcat, may be inputted into the Feed Forward Network block of FIG. 14, which outputs a filtered feature map, FMf. For some embodiments, the first transformed feature map of FIG. 13, FM , may be concatenated with an expanded first feature of FIG. 13, FV1. For some embodiments, this combination of FIGs. 13 and 14 to generate a filtered feature map may be used to further filter a transformed feature map, such as FMtof FIGs. 10 and 11 .Combination with Another Feature Extractor

[0198] For some embodiments, to further enrich the abstraction level of the proposed feature extractor, different purpose neural networks may be put in place along with the distribution learning network. The residual networks of He and Szegedy and the transformer-based models of Zhao, Mao, and Zhang are examples of the other feature extractors that may be combined with the distribution learning network. In the following subsections, two such combined architectures are discussed.Parallel Architecture

[0199] FIG. 15 is a functional block diagram showing an example parallel architecture for an enhanced distribution learning network according to some embodiments. FIG. 15 illustrates an enhanced distribution learning network process 1500 in which the distribution learning network 1504 outputting a transformed FM is combined with another feature extraction network 1514 in a parallel manner. A feature map FM 1502 is fed to both networks 1504, 1514 in parallel. These neural networks extract a distribution adjusted feature map, FMd1506, from another feature map, FMa. After concatenating both feature maps, an additional feed forward network 1510, such as a pixel-based CNN or a CNN configured for image or video data, further merges the concatenated FM 1508 and outputs an enhanced feature map, FMe1512. For some embodiments, to avoid degradation in a deeper network with multiple stages, the architecture may combine with, e.g., a residual network. This residual network block (“another feature extractor” block 1514 in FIG. 15) works jointly with the distribution learning block in parallel, thereby propagating a rich representation through a deep network. In some embodiments, the “Distribution Learning Network” block 1504 in FIG. 15 may process a downsampled and filtered FM. In this way, both global and local features are preserved by the “filtering” process.Serial Architecture

[0200] FIG. 16 is a functional block diagram showing an example serial architecture for an enhanced distribution learning network according to some embodiments. Similar to the parallel architecture of FIG. 15, FIG. 16 shows another enhanced distribution learning network process 1600. In FIG. 16, the output of the distributed learning network block 1604, which may be a distribution-adjusted feature map, FMd1606, is combined in series with another feature extraction network block. In this architecture 1600, a feature map FM 1602 is inputted into the distribution learning network block 1604. The distribution learning feature map, FMd1606, becomes the input of the other network block (“another feature extractor” block 1608 in FIG. 16), which outputs another feature map, FMa. The enhanced feature map, FMe1612, is generated by a feed forward network 1610.

[0201] For some embodiments, the serial enhanced “Distribution Learning Network” in FIG. 16 may also process a downsampled and filtered FM. For both parallel and serial architectures, for some embodiments, multiple other purpose networks (more than one) may be combined with the distribution learning network.Deep Enhanced Distribution Learning Network

[0202] FIG. 17 is a functional block diagram showing an example deep enhanced distribution learning network according to some embodiments. The enhanced distribution learning network may be designed in a multi-stage architecture 1700. Such a deep neural network, which may be called a deep enhanced distribution learning network (Deep EDL-Net) 1704, 1706, 1708, is illustrated in detail in FIG. 17. In each stage, the enhanced distribution learning network (either a parallel or serial EDL network combined with residual or transformer layers) 1704, 1706, 1708 outputs an enhanced feature map, FMe., in which i represents the i-th stage. After the last stage, a ® operation stabilizes the deep network by estimating the residual of the input feature map 1702. For some embodiments, the networks try to learn network parameters through the chain of enhanced distributed learning network blocks, and this ® operation leads to an estimate of a residual of the FM. The output feature map, FM01710, is the final enriched feature map generated through the deep EDL network.Application within Learning-Based Compression Frameworks

[0203] Feature extraction in a learning-based compression system is a relatively new area compared to the classic image classifications or segmentations processes. As observed from the recent learning-based image compression frameworks of Balle and Minnen, a feature extractor needs to be designed carefully to handle the trade-off between the fine abstraction and coding time.

[0204] A deep EDL-Net architecture enriches the pixel feature with an extendable design. For some embodiments, the architecture may be followed by any neural network layer outputting a compatible feature map. The architecture is also applicable within both encoder and decoder of any learning-based compression framework.

[0205] FIG. 18 is a functional block diagram showing an example deep enhanced distribution learning network in an image compression framework according to some embodiments. In an image compression system, the pixels are regularly defined in 2D image space. The images are downsampled and upsampled usually by CNN layers. Examples of recent image compression architecture is introduced in articles Balle and Minnen.

[0206] A pixel-based image compression architecture may be combined with Deep EDL-Net blocks as illustrated in the architecture 1800 of FIG. 18. For example, in between the pixel upsampling layers 1 (1822) and 2 (1818) in the decoder, a deep EDL-Net 1820 may be inserted to enhance the feature map upsampled from the Pixel Up NN Layers 2 block 1818. FIG. 18 also shows other possible locations to apply a deep EDL-Net 1806, 1810, 1816, 1820 for some embodiments. FIG. 18 shows other locations to apply deep EDL-Net 1806, 1810, 1816, 1820 for some embodiments.

[0207] For some embodiments, an input image 1802 may be inputted into a pixel down NN layers 1 block 1804 and the output passed through a deep EDL-Net block 1806. The output of the deep EDL-Net block 1806 may be passed through a pixel down NN layers 2 block 1808 and the output passed through a second deep EDL-Net block 1810. The output of the second deep EDL-Net block 1810 may be passed through an entropy encoder 1812 to generate a bitstream. On the decoder side, the bitstream may be passed through an entropy decoder 1814. The output of the entropy decoder 1814 may be passed through a deep EDL-Net block 1816. The output of the deep EDL-Net block 1816 may be passed through a pixel up NN layers 2 block 1818, the output of which may be passed through a second deep EDL-Net block 1820. The output of the second deep EDL-Net block 1820 may be passed through a pixel up NN layers 1 block 1822 to generate an output image 1824.

[0208] FIG. 19 is a flowchart illustrating an example learning-based image compression process according to some embodiments. For some embodiments, an example process 1900 may include obtaining 1902 a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers. For some embodiments, the example process 1900 may further include generating 1904 a first set of distribution parameters using a first set of neural network layers based on the first feature map. For some embodiments, the example process 1900 may further include transforming 1906the first feature map to a second feature map based on the first set of distribution parameters. For some embodiments, the example process 1900 may further include outputting 1908 the second feature map to a succeeding neural network layer.

[0209] FIG. 20 is a flowchart illustrating an example learning-based image compression process according to some embodiments. For some embodiments, an example process 2000 may include obtaining 2002 a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers. For some embodiments, the example process 2000 may further include generating 2004 a first set of distribution parameters using a first set of neural network layers based on the first feature map. For some embodiments, the example process 2000 may further include expanding 2006 the first set of distribution parameters to dimensionally-match the first feature map. For some embodiments, the example process 2000 may further include subtracting 2008 the expanded first set of distribution parameters from the first feature map to generate a centered feature map. For some embodiments, the example process 2000 may further include generating 2010 a second set of distribution parameters using a second set of neural network layers based on the first feature map. For some embodiments, the example process 2000 may further include expanding 2012 the second set of distribution parameters to dimensionally-match the centered feature map. For some embodiments, the example process 2000 may further include dividing 2014 the centered feature map by the expanded second set of distribution parameters to generate a normalized feature map. For some embodiments, the example process 2000 may further include outputting 2016 the normalized feature map to a succeeding neural network layer.

[0210] FIG. 21 is a flowchart illustrating an example learning-based image compression process according to some embodiments. For some embodiments, an example process 2100 may include obtaining 2102 a normalized feature map, wherein the normalized feature map includes C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers. For some embodiments, the example process 2100 may further include generating 2104 a first set of distribution parameters using a first set of neural network layers based on the normalized feature map. For some embodiments, the example process 2100 may further include expanding 2106 the first set of distribution parameters to dimensionally-match the normalized feature map. For some embodiments, the example process 2100 may further include scaling 2108 the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map. For some embodiments, the example process 2100 may further include generating 2110 a second set of distribution parameters using asecond set of neural network layers based on the normalized feature map. For some embodiments, the example process 2100 may further include expanding 2112 the second set of distribution parameters to dimensionally- match the scaled feature map. For some embodiments, the example process 2100 may further include adding 2114 the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map. For some embodiments, the example 2100 process may further include outputting 2116 the transformed feature map to a succeeding neural network layer.

[0211] A first example method in accordance with some embodiments may include: obtaining a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; transforming the first feature map to a second feature map based on the first set of distribution parameters; and outputting the second feature map to a succeeding neural network layer.

[0212] For some embodiments of the first example method, the first feature map further includes n feature vectors respectively corresponding to n distinct pixels in an image space.

[0213] For some embodiments of the first example method, each feature vector represents a feature of the corresponding pixel in the image space, and the first feature map has dimensions n by C.

[0214] Some embodiments of the first example method may further include: generating a second set of distribution parameters using a second set of neural networks layers; transforming the first feature map to a third feature map based on the second set of distribution parameters; and updating the first feature map to the third feature map.

[0215] For some embodiments of the first example method, generating the first set of distribution parameters may include: updating a given feature map via neural network layers; and simplifying the updated feature map to obtain the first set of per channel distribution parameters.

[0216] For some embodiments of the first example method, transforming the first feature map may include: generating, for each of the C channels in the first feature map, a respective reshaped vector, wherein each reshaped vector is generated by reshaping a respective channel vector in the first feature map; obtaining a distribution parameter for each of the C channels corresponding to one of the respective reshaped vectors; expanding each of the distribution parameters into a respective feature channel; and transforming, for each of the C channels, each element in the reshaped vector by the expanded distribution parameter.

[0217] Some embodiments of the first example method may further include updating the first feature map by normalizing each feature element of the first feature map.

[0218] For some embodiments of the first example method, normalizing each feature element of the first feature map may include: generating a third set of distribution parameters using a third set of neural network layers; centering the first feature map to a fourth feature map based on the third set of distribution parameters; and updating the first feature map to the fourth feature map.

[0219] Some embodiments of the first example method may further include: generating a fourth set of distribution parameters from the fourth feature map; and normalizing the fourth feature map to a fifth feature map based on the fourth set of distribution parameters.

[0220] Some embodiments of the first example method may further include: generating a fifth set of distribution parameters using a fifth set of neural networks layers; and normalizing the fourth feature map to a sixth feature map based on the fifth set of distribution parameters.

[0221] Some embodiments of the first example method may further include downsampling the second feature map to generate a first feature vector, wherein downsampling the second feature map uses a pooling function.

[0222] For some embodiments of the first example method, the pooling function is selected from the group consisting of average pooling and max pooling.

[0223] Some embodiments of the first example method may further include: generating an expanded feature map by expanding the first feature vector towards a channel dimension; concatenating the second feature map with the expanded feature map to generate a concatenated feature map; and passing the concatenated feature map through a filtering neural network layers to generate a filtered feature map.

[0224] Some embodiments of the first example method may further include aggregating the second feature map using an additional neural network.

[0225] For some embodiments of the first example method, the additional neural network includes a convolution neural network (CNN) configured for image data.

[0226] Some embodiments of the first example method may further include aggregating the second feature map using a residual network.

[0227] For some embodiments of the first example method, the residual network is a ResNet block.

[0228] Some embodiments of the first example method may further include aggregating the second feature map using a transformer block.

[0229] For some embodiments of the first example method, the transformer block is selected from the group consisting of a point transformer and a voxel transformer.

[0230] Some embodiments of the first example method may further include: generating a seventh feature map by aggregating the second feature map using a neural network in parallel to transform the first feature map to the second feature map; and concatenating the seventh feature map to the second feature map.

[0231] Some embodiments of the first example method may further include: repeating the learning-based point cloud geometry process one or more times to generate an eighth feature map; and adding the eighth feature map to the first feature map to generate a ninth feature map.

[0232] A first example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map; transform the first feature map to a second feature map based on the first set of distribution parameters; and output the second feature map to a succeeding neural network layer.

[0233] For some embodiments of the first example apparatus, the first feature map further includes n feature vectors respectively corresponding to n distinct pixels in an image space.

[0234] For some embodiments of the first example apparatus, each feature vector represents a feature of the corresponding pixel in the image space, and the first feature map has dimensions n by C.

[0235] A second example method in accordance with some embodiments may include: obtaining a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; expanding the first set of distribution parameters to dimensionally-match the first feature map; subtracting the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generating a second set of distribution parameters using a second set of neural network layers based on the first feature map; expanding the second set of distribution parameters to dimensionally-match the centered feature map; dividing the centered feature map by the expandedsecond set of distribution parameters to generate a normalized feature map; and outputting the normalized feature map to a succeeding neural network layer.

[0236] For some embodiments of the second example method, the first feature map further includes n feature vectors respectively corresponding to n distinct pixels in an image space.

[0237] For some embodiments of the second example method, each feature vector represents a feature of the corresponding pixel in the image space, and the first feature map has dimensions n by C.

[0238] For some embodiments of the second example method, expanding the first set of distribution parameters to dimensionally-match the first feature map includes copying a first distribution parameter vector one or more times to dimensionally-match a corresponding dimension of the first feature map, the first distribution parameter vector includes the first set of distribution parameters, expanding the second set of distribution parameters to dimensionally-match the centered feature map includes copying a second distribution parameter vector one or more times to dimensionally-match a corresponding dimension of the centered feature map, and the second distribution parameter vector includes the second set of distribution parameters.

[0239] A second example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map includes C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map; expand the first set of distribution parameters to dimensionally-match the first feature map; subtract the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generate a second set of distribution parameters using a second set of neural network layers based on the first feature map; expand the second set of distribution parameters to dimensionally-match the centered feature map; divide the centered feature map by the expanded second set of distribution parameters to generate a normalized feature map; and output the normalized feature map to a succeeding neural network layer.

[0240] A third example method in accordance with some embodiments may include: obtaining a normalized feature map, wherein the normalized feature map includes C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the normalized feature map; expanding the first set of distribution parameters to dimensionally-match the normalized feature map; scaling the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map; generating a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expanding the second set of distribution parameters to dimensionally-match the scaled feature map; adding the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and outputting the transformed feature map to a succeeding neural network layer.

[0241] For some embodiments of the third example method, the first feature map further includes n feature vectors respectively corresponding to n distinct pixels in an image space.

[0242] For some embodiments of the third example method, each feature vector represents a feature of the corresponding pixel in the image space, and the first feature map has dimensions n by C.

[0243] For some embodiments of the third example method, an input to the second set of neural network layers is the scaled feature map.

[0244] For some embodiments of the third example method, an input to the second set of neural network layers is the normalized feature map.

[0245] Some embodiments of the third example method may further include downsampling the transformed feature map to generate a first feature vector, wherein downsampling the transformed feature map uses a pooling function.

[0246] For some embodiments of the third example method, the pooling function is selected from the group consisting of average pooling and max pooling.

[0247] Some embodiments of the third example method may further include: generating an expanded feature map by expanding the first feature vector towards a channel dimension; concatenating the first feature map with the expanded feature map to generate a concatenated feature map; and passing the concatenated feature map through a filtering neural network layers to generate a filtered feature map.

[0248] Some embodiments of the third example method may further include aggregating the first feature map using an additional neural network.

[0249] For some embodiments of the third example method, the additional neural network includes a convolution neural network (CNN) configured for image data.

[0250] Some embodiments of the third example method may further include aggregating the first feature map using a residual network.

[0251] For some embodiments of the third example method, the residual network is a ResNet block.

[0252] Some embodiments of the third example method may further include: repeating the learning-based image processing method one or more times to generate an enhanced feature map; and adding the enhanced feature map to the first feature map to generate an enhanced output feature map.

[0253] Some embodiments of the third example method may further include repeating the learning-based image processing method one or more times within a point cloud encoder.

[0254] Some embodiments of the third example method may further include repeating the learning-based image processing method one or more times within a point cloud decoder.

[0255] For some embodiments of the third example method, the image processing encoder processes pixels.

[0256] For some embodiments of the third example method, the image processing encoder processes images.

[0257] A third example apparatus in accordance with some embodiments may include: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a normalized feature map, wherein the normalized feature map includes C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the normalized feature map; expand the first set of distribution parameters to dimensionally-match the normalized feature map; scale the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map; generate a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expand the second set of distribution parameters to dimensionally-match the scaled feature map; add the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and output the transformed feature map to a succeeding neural network layer.

[0258] An example apparatus in accordance with some embodiments may include at least one processor configured to perform any one of the methods listed above.

[0259] An example apparatus in accordance with some embodiments may include a computer-readable medium storing instructions for causing one or more processors to perform any one of the methods listed above.

[0260] An example apparatus in accordance with some embodiments may include at least one processor and at least one non-transitory computer-readable medium storing instructions for causing the at least one processor to perform any one of the methods listed above.

[0261] An example apparatus in accordance with some embodiments may include a computer-readable medium storing a scene description file generated according to any one of the methods listed above.

[0262] An example apparatus in accordance with some embodiments may include a signal including a scene description file generated according to any one of the methods listed above.

[0263] This disclosure describes a variety of aspects, including tools, features, embodiments, models, approaches, etc. Many of these aspects are described with specificity and, at least to show the individual characteristics, are often described in a manner that may sound limiting. However, this is for purposes of clarity in description, and does not limit the disclosure or scope of those aspects. Indeed, all of the different aspects can be combined and interchanged to provide further aspects. Moreover, the aspects can be combined and interchanged with aspects described in earlier filings as well.

[0264] The aspects described and contemplated in this disclosure can be implemented in many different forms. While some embodiments are illustrated specifically, other embodiments are contemplated, and the discussion of particular embodiments does not limit the breadth of the implementations. At least one of the aspects generally relates to video encoding and decoding, and at least one other aspect generally relates to transmitting a bitstream generated or encoded. These and other aspects can be implemented as a method, an apparatus, a computer readable storage medium having stored thereon instructions for encoding or decoding video data according to any of the methods described, and / or a computer readable storage medium having stored thereon a bitstream generated according to any of the methods described.

[0265] In the present disclosure, the terms “reconstructed” and “decoded” may be used interchangeably, the terms “pixel” and “sample” may be used interchangeably, the terms “image,” “picture” and “frame” may be used interchangeably. Usually, but not necessarily, the term “reconstructed” is used at the encoder side while “decoded” is used at the decoder side.

[0266] The terms HDR (high dynamic range) and SDR (standard dynamic range) often convey specific values of dynamic range to those of ordinary skill in the art. However, additional embodiments are also intended in whicha reference to HDR is understood to mean “higher dynamic range” and a reference to SDR is understood to mean “lower dynamic range.” Such additional embodiments are not constrained by any specific values of dynamic range that might often be associated with the terms “high dynamic range” and “standard dynamic range.”

[0267] Various methods are described herein, and each of the methods includes one or more steps or actions for achieving the described method. Unless a specific order of steps or actions is required for proper operation of the method, the order and / or use of specific steps and / or actions may be modified or combined. Additionally, terms such as “first”, “second”, etc. may be used in various embodiments to modify an element, component, step, operation, etc., such as, for example, a “first decoding” and a “second decoding”. Use of such terms does not imply an ordering to the modified operations unless specifically required. So, in this example, the first decoding need not be performed before the second decoding, and may occur, for example, before, during, or in an overlapping time period with the second decoding.

[0268] Various numeric values may be used in the present disclosure, for example. The specific values are for example purposes and the aspects described are not limited to these specific values.

[0269] Embodiments described herein may be carried out by computer software implemented by a processor or other hardware, or by a combination of hardware and software. As a non-limiting example, the embodiments can be implemented by one or more integrated circuits. The processor can be of any type appropriate to the technical environment and can encompass one or more of microprocessors, general purpose computers, special purpose computers, and processors based on a multi-core architecture, as non-limiting examples.

[0270] Various implementations involve decoding. “Decoding”, as used in this disclosure, can encompass all or part of the processes performed, for example, on a received encoded sequence in order to produce a final output suitable for display. In various embodiments, such processes include one or more of the processes typically performed by a decoder, for example, entropy decoding, inverse quantization, inverse transformation, and differential decoding. In various embodiments, such processes also, or alternatively, include processes performed by a decoder of various implementations described in this disclosure, for example, extracting a picture from a tiled (packed) picture, determining an upsampling filter to use and then upsampling a picture, and flipping a picture back to its intended orientation.

[0271] As further examples, in one embodiment “decoding” refers only to entropy decoding, in another embodiment “decoding” refers only to differential decoding, and in another embodiment “decoding” refers to a combination of entropy decoding and differential decoding. Whether the phrase “decoding process” is intendedto refer specifically to a subset of operations or generally to the broader decoding process will be clear based on the context of the specific descriptions.

[0272] Various implementations involve encoding. In an analogous way to the above discussion about “decoding”, “encoding” as used in this disclosure can encompass all or part of the processes performed, for example, on an input video sequence in order to produce an encoded bitstream. In various embodiments, such processes include one or more of the processes typically performed by an encoder, for example, partitioning, differential encoding, transformation, quantization, and entropy encoding. In various embodiments, such processes also, or alternatively, include processes performed by an encoder of various implementations described in this disclosure.

[0273] As further examples, in one embodiment “encoding” refers only to entropy encoding, in another embodiment “encoding” refers only to differential encoding, and in another embodiment “encoding” refers to a combination of differential encoding and entropy encoding. Whether the phrase “encoding process” is intended to refer specifically to a subset of operations or generally to the broader encoding process will be clear based on the context of the specific descriptions.

[0274] When a figure is presented as a flow diagram, it should be understood that it also provides a block diagram of a corresponding apparatus. Similarly, when a figure is presented as a block diagram, it should be understood that it also provides a flow diagram of a corresponding method / process.

[0275] Various embodiments refer to rate distortion optimization. In particular, during the encoding process, the balance or trade-off between the rate and distortion is usually considered, often given the constraints of computational complexity. The rate distortion optimization is usually formulated as minimizing a rate distortion function, which is a weighted sum of the rate and of the distortion. There are different approaches to solve the rate distortion optimization problem. For example, the approaches may be based on an extensive testing of all encoding options, including all considered modes or coding parameters values, with a complete evaluation of their coding cost and related distortion of the reconstructed signal after coding and decoding. Faster approaches may also be used, to save encoding complexity, in particular with computation of an approximated distortion based on the prediction or the prediction residual signal, not the reconstructed one. A mix of these two approaches can also be used, such as by using an approximated distortion for only some of the possible encoding options, and a complete distortion for other encoding options. Other approaches only evaluate a subset of the possible encoding options. More generally, many approaches employ any of a variety of techniques toperform the optimization, but the optimization is not necessarily a complete evaluation of both the coding cost and related distortion.

[0276] The implementations and aspects described herein can be implemented in, for example, a method or a process, an apparatus, a software program, a data stream, or a signal. Even if only discussed in the context of a single form of implementation (for example, discussed only as a method), the implementation of features discussed can also be implemented in other forms (for example, an apparatus or program). An apparatus can be implemented in, for example, appropriate hardware, software, and firmware. The methods can be implemented in, for example, a processor, which refers to processing devices in general, including, for example, a computer, a microprocessor, an integrated circuit, or a programmable logic device. Processors also include communication devices, such as, for example, computers, cell phones, portable / personal digital assistants (“PDAs”), and other devices that facilitate communication of information between end-users.

[0277] Reference to “one embodiment” or “an embodiment” or “one implementation” or “an implementation”, as well as other variations thereof, means that a particular feature, structure, characteristic, and so forth described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of the phrase “in one embodiment” or “in an embodiment” or “in one implementation” or “in an implementation”, as well any other variations, appearing in various places throughout this disclosure are not necessarily all referring to the same embodiment.

[0278] Additionally, this disclosure may refer to “determining” various pieces of information. Determining the information can include one or more of, for example, estimating the information, calculating the information, predicting the information, or retrieving the information from memory.

[0279] Further, this disclosure may refer to “accessing” various pieces of information. Accessing the information can include one or more of, for example, receiving the information, retrieving the information (for example, from memory), storing the information, moving the information, copying the information, calculating the information, determining the information, predicting the information, or estimating the information.

[0280] Additionally, this disclosure may refer to “receiving” various pieces of information. Receiving is, as with “accessing”, intended to be a broad term. Receiving the information can include one or more of, for example, accessing the information, or retrieving the information (for example, from memory). Further, “receiving” is typically involved, in one way or another, during operations such as, for example, storing the information, processing the information, transmitting the information, moving the information, copying the information, erasingthe information, calculating the information, determining the information, predicting the information, or estimating the information.

[0281] It is to be appreciated that the use of any of the following 7”, “and / or”, and “at least one of”, for example, in the cases of “A / B”, “A and / or B” and “at least one of A and B”, is intended to encompass the selection of the first listed option (A) only, or the selection of the second listed option (B) only, or the selection of both options (A and B). As a further example, in the cases of “A, B, and / or C” and “at least one of A, B, and C”, such phrasing is intended to encompass the selection of the first listed option (A) only, or the selection of the second listed option (B) only, or the selection of the third listed option (C) only, or the selection of the first and the second listed options (A and B) only, or the selection of the first and third listed options (A and C) only, or the selection of the second and third listed options (B and C) only, or the selection of all three options (A and B and C). This may be extended for as many items as are listed.

[0282] Also, as used herein, the word “signal” refers to, among other things, indicating something to a corresponding decoder. For example, in certain embodiments the encoder signals a particular one of a plurality of parameters for region-based filter parameter selection for de-artifact filtering. In this way, in an embodiment the same parameter is used at both the encoder side and the decoder side. Thus, for example, an encoder can transmit (explicit signaling) a particular parameter to the decoder so that the decoder can use the same particular parameter. Conversely, if the decoder already has the particular parameter as well as others, then signaling can be used without transmitting (implicit signaling) to simply allow the decoder to know and select the particular parameter. By avoiding transmission of any actual functions, a bit savings is realized in various embodiments. It is to be appreciated that signaling can be accomplished in a variety of ways. For example, one or more syntax elements, flags, and so forth are used to signal information to a corresponding decoder in various embodiments. While the preceding relates to the verb form of the word “signal”, the word “signal” can also be used herein as a noun.

[0283] Implementations can produce a variety of signals formatted to carry information that can be, for example, stored or transmitted. The information can include, for example, instructions for performing a method, or data produced by one of the described implementations. For example, a signal can be formatted to carry the bitstream of a described embodiment. Such a signal can be formatted, for example, as an electromagnetic wave (for example, using a radio frequency portion of spectrum) or as a baseband signal. The formatting can include, for example, encoding a data stream and modulating a carrier with the encoded data stream. The information that the signal carries can be, for example, analog or digital information. The signal can be transmitted over avariety of different wired or wireless links, as is known. The signal can be stored on a processor-readable medium.

[0284] We describe a number of embodiments. Features of these embodiments can be provided alone or in any combination, across various claim categories and types. Further, embodiments can include one or more of the following features, devices, or aspects, alone or in any combination, across various claim categories and types:• Adapting residues at an encoder according to any of the embodiments discussed.• A bitstream or signal that includes one or more of the described syntax elements, or variations thereof.• A bitstream or signal that includes syntax conveying information generated according to any of the embodiments described.• Inserting in the signaling syntax elements that enable the decoder to adapt residues in a manner corresponding to that used by an encoder.• Creating and / or transmitting and / or receiving and / or decoding a bitstream or signal that includes one or more of the described syntax elements, or variations thereof.• Creating and / or transmitting and / or receiving and / or decoding according to any of the embodiments described.• A method, process, apparatus, medium storing instructions, medium storing data, or signal according to any of the embodiments described.• A TV, set-top box, cell phone, tablet, or other electronic device that performs adaptation of filter parameters according to any of the embodiments described.• A TV, set-top box, cell phone, tablet, or other electronic device that performs adaptation of filter parameters according to any of the embodiments described, and that displays (e.g. using a monitor, screen, or other type of display) a resulting image.• A TV, set-top box, cell phone, tablet, or other electronic device that selects (e.g. using a tuner) a channel to receive a signal including an encoded image, and performs adaptation of filter parameters according to any of the embodiments described.• A TV, set-top box, cell phone, tablet, or other electronic device that receives (e.g. using an antenna) a signal over the air that includes an encoded image, and performs adaptation of filter parameters according to any of the embodiments described.

[0285] Note that various hardware elements of one or more of the described embodiments are referred to as “modules” that carry out (i.e., perform, execute, and the like) various functions that are described herein in connection with the respective modules. As used herein, a module includes hardware (e.g., one or more processors, one or more microprocessors, one or more microcontrollers, one or more microchips, one or more application-specific integrated circuits (ASICs), one or more field programmable gate arrays (FPGAs), one or more memory devices) deemed suitable by those of skill in the relevant art for a given implementation. Each described module may also include instructions executable for carrying out the one or more functions described as being carried out by the respective module, and it is noted that those instructions could take the form of or include hardware (i.e., hardwired) instructions, firmware instructions, software instructions, and / or the like, and may be stored in any suitable non-transitory computer-readable medium or media, such as commonly referred to as RAM, ROM, etc.

[0286] Although features and elements are described above in particular combinations, one of ordinary skill in the art will appreciate that each feature or element can be used alone or in any combination with the other features and elements. In addition, the methods described herein may be implemented in a computer program, software, or firmware incorporated in a computer-readable medium for execution by a computer or processor. Examples of computer-readable storage media include, but are not limited to, a read only memory (ROM), a random access memory (RAM), a register, cache memory, semiconductor memory devices, magnetic media such as internal hard disks and removable disks, magneto-optical media, and optical media such as CD-ROM disks, and digital versatile disks (DVDs). A processor in association with software may be used to implement a radio frequency transceiver for use in a WTRU, UE, terminal, base station, RNC, or any host computer.

Claims

CLAIMS1. A learning-based image processing method, the method comprising: obtaining a first feature map, wherein the first feature map comprises C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; transforming the first feature map to a second feature map based on the first set of distribution parameters; and encoding a bitstream based on the second feature map.

2. A learning-based image processing method, the method comprising: obtaining a first feature map, wherein the first feature map comprises C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; transforming the first feature map to a second feature map based on the first set of distribution parameters; and outputting the second feature map to a succeeding neural network layer.

3. The method of claim 2, wherein the first feature map further comprises n feature vectors respectively corresponding to n distinct pixels in an image space.

4. The method of claim 3, wherein each feature vector represents a feature of the corresponding pixel in the image space, and wherein the first feature map has dimensions n by C.

5. The method of claim 2, further comprising: generating a second set of distribution parameters using a second set of neural networks layers; transforming the first feature map to a third feature map based on the second set of distribution parameters; andupdating the first feature map to the third feature map.

6. The method of claim 2, wherein generating the first set of distribution parameters comprises: updating a given feature map via neural network layers; and simplifying the updated feature map to obtain the first set of per channel distribution parameters.

7. The method of claim 2, wherein transforming the first feature map comprises: generating, for each of the C channels in the first feature map, a respective reshaped vector, wherein each reshaped vector is generated by reshaping a respective channel vector in the first feature map; obtaining a distribution parameter for each of the C channels corresponding to one of the respective reshaped vectors; expanding each of the distribution parameters into a respective feature channel; and transforming, for each of the C channels, each element in the reshaped vector by the expanded distribution parameter.

8. The method of claim 2, further comprising updating the first feature map by normalizing each feature element of the first feature map.

9. The method of claim 7, wherein normalizing each feature element of the first feature map comprises: generating a third set of distribution parameters using a third set of neural network layers; centering the first feature map to a fourth feature map based on the third set of distribution parameters; and updating the first feature map to the fourth feature map.

10. The method of claim 9, further comprising: generating a fourth set of distribution parameters from the fourth feature map; and normalizing the fourth feature map to a fifth feature map based on the fourth set of distribution parameters.11 . The method of claim 9, further comprising: generating a fifth set of distribution parameters using a fifth set of neural networks layers; andnormalizing the fourth feature map to a sixth feature map based on the fifth set of distribution parameters.

12. The method of claim 2, further comprising: downsampling the second feature map to generate a first feature vector, wherein downsampling the second feature map uses a pooling function.

13. The method of claim 12, wherein the pooling function is selected from the group consisting of average pooling and max pooling.

14. The method of claim 12, further comprising: generating an expanded feature map by expanding the first feature vector towards a channel dimension; concatenating the second feature map with the expanded feature map to generate a concatenated feature map; and passing the concatenated feature map through a filtering neural network layers to generate a filtered feature map.

15. The method of claim 2, further comprising aggregating the second feature map using an additional neural network.

16. The method of claim 15, wherein the additional neural network comprises a convolution neural network (CNN) configured for image data.

17. The method of claim 2, further comprising aggregating the second feature map using a residual network.

18. The method of claim 17, wherein the residual network is a ResNet block.

19. The method of claim 2, further comprising aggregating the second feature map using a transformer block.

20. The method of claim 19, wherein the transformer block is selected from the group consisting of a point transformer and a voxel transformer.21 . The method of claim 2, further comprising:generating a seventh feature map by aggregating the second feature map using a neural network in parallel to transform the first feature map to the second feature map; and concatenating the seventh feature map to the second feature map.

22. The method of claim 2, further comprising: repeating the learning-based point cloud geometry process one or more times to generate an eighth feature map; and adding the eighth feature map to the first feature map to generate a ninth feature map.

23. An apparatus comprising: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map comprises C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map; transform the first feature map to a second feature map based on the first set of distribution parameters; and encode a bitstream based on the second feature map.

24. An apparatus comprising: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map comprises C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map;transform the first feature map to a second feature map based on the first set of distribution parameters; and output the second feature map to a succeeding neural network layer.

25. The apparatus of claim 24, wherein the first feature map further comprises n feature vectors respectively corresponding to n distinct pixels in an image space.

26. The apparatus of claim 24, wherein each feature vector represents a feature of the corresponding pixel in the image space, and wherein the first feature map has dimensions n by C.

27. A learning-based image processing method, the method comprising: obtaining a first feature map, wherein the first feature map comprises C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; expanding the first set of distribution parameters to dimensionally-match the first feature map; subtracting the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generating a second set of distribution parameters using a second set of neural network layers based on the first feature map; expanding the second set of distribution parameters to dimensionally-match the centered feature map; dividing the centered feature map by the expanded second set of distribution parameters to generate a normalized feature map; and encoding a bitstream based on the normalized feature map.

28. A learning-based image processing method, the method comprising: obtaining a first feature map, wherein the first feature map comprises C channels, and wherein the first feature map is generated by one or more preceding neural network layers;generating a first set of distribution parameters using a first set of neural network layers based on the first feature map; expanding the first set of distribution parameters to dimensionally-match the first feature map; subtracting the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generating a second set of distribution parameters using a second set of neural network layers based on the first feature map; expanding the second set of distribution parameters to dimensionally-match the centered feature map; dividing the centered feature map by the expanded second set of distribution parameters to generate a normalized feature map; and outputting the normalized feature map to a succeeding neural network layer.

29. The method of claim 28, wherein the first feature map further comprises n feature vectors respectively corresponding to n distinct pixels in an image space.

30. The method of claim 28, wherein each feature vector represents a feature of the corresponding pixel in the image space, and wherein the first feature map has dimensions n by C.

31. The method of claim 28, wherein expanding the first set of distribution parameters to dimensionally-match the first feature map comprises copying a first distribution parameter vector one or more times to dimensionally-match a corresponding dimension of the first feature map, wherein the first distribution parameter vector comprises the first set of distribution parameters, wherein expanding the second set of distribution parameters to dimensionally-match the centered feature map comprises copying a second distribution parameter vector one or more times to dimensionally- match a corresponding dimension of the centered feature map, and wherein the second distribution parameter vector comprises the second set of distribution parameters.

32. An apparatus comprising: a processor; anda non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map comprises C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map; expand the first set of distribution parameters to dimensionally-match the first feature map; subtract the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generate a second set of distribution parameters using a second set of neural network layers based on the first feature map; expand the second set of distribution parameters to dimensionally-match the centered feature map; divide the centered feature map by the expanded second set of distribution parameters to generate a normalized feature map; and encode a bitstream based on the normalized feature map.

33. An apparatus comprising: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a first feature map, wherein the first feature map comprises C channels, and wherein the first feature map is generated by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the first feature map; expand the first set of distribution parameters to dimensionally-match the first feature map; subtract the expanded first set of distribution parameters from the first feature map to generate a centered feature map; generate a second set of distribution parameters using a second set of neural network layers based on the first feature map;expand the second set of distribution parameters to dimensionally-match the centered feature map; divide the centered feature map by the expanded second set of distribution parameters to generate a normalized feature map; and output the normalized feature map to a succeeding neural network layer.

34. A learning-based image processing method, the method comprising: obtaining a normalized feature map, wherein the normalized feature map comprises C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the normalized feature map; expanding the first set of distribution parameters to dimensionally-match the normalized feature map; scaling the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map; generating a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expanding the second set of distribution parameters to dimensionally-match the scaled feature map; adding the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and encoding a bitstream based on the transformed feature map.

35. A learning-based image processing method, the method comprising: obtaining a normalized feature map, wherein the normalized feature map comprises C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generating a first set of distribution parameters using a first set of neural network layers based on the normalized feature map;expanding the first set of distribution parameters to dimensionally-match the normalized feature map; scaling the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map; generating a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expanding the second set of distribution parameters to dimensionally-match the scaled feature map; adding the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and outputting the transformed feature map to a succeeding neural network layer.

36. The method of claim 35, wherein the first feature map further comprises n feature vectors respectively corresponding to n distinct pixels in an image space.

37. The method of claim 35, wherein each feature vector represents a feature of the corresponding pixel in the image space, and wherein the first feature map has dimensions n by C.

38. The method of claim 35, wherein an input to the second set of neural network layers is the scaled feature map.

39. The method of claim 35, wherein an input to the second set of neural network layers is the normalized feature map.

40. The method of any one of claims 35-39, further comprising: downsampling the transformed feature map to generate a first feature vector, wherein downsampling the transformed feature map uses a pooling function.41 . The method of claim 40, wherein the pooling function is selected from the group consisting of average pooling and max pooling.

42. The method of any one of claims 40-41, further comprising: generating an expanded feature map by expanding the first feature vector towards a channel dimension;concatenating the first feature map with the expanded feature map to generate a concatenated feature map; and passing the concatenated feature map through a filtering neural network layers to generate a filtered feature map.

43. The method of any one of claims 35-42, further comprising aggregating the first feature map using an additional neural network.

44. The method of claim 43, wherein the additional neural network comprises a convolution neural network (CNN) configured for image data.

45. The method of any one of claims 35-42, further comprising aggregating the first feature map using a residual network.

46. The method of claim 45, wherein the residual network is a ResNet block.

47. The method of any one of claims 35-46, further comprising: repeating the learning-based image processing method one or more times to generate an enhanced feature map; and adding the enhanced feature map to the first feature map to generate an enhanced output feature map.

48. The method of any one of claims 35-47, further comprising repeating the learning-based image processing method one or more times within a point cloud encoder.

49. The method of any one of claims 35-47, further comprising repeating the learning-based image processing method one or more times within a point cloud decoder.

50. The method of any one of claims 35-48, wherein the image processing encoder processes pixels.51 . The method of any one of claims 35-48, wherein the image processing encoder processes images.

52. An apparatus comprising: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to:obtain a normalized feature map, wherein the normalized feature map comprises C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the normalized feature map; expand the first set of distribution parameters to dimensionally-match the normalized feature map; scale the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map; generate a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expand the second set of distribution parameters to dimensionally-match the scaled feature map; add the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and encode a bitstream based on the transformed feature map.

53. An apparatus comprising: a processor; and a non-transitory computer-readable medium storing instructions operative, when executed by the processor, to cause the apparatus to: obtain a normalized feature map, wherein the normalized feature map comprises C channels corresponding to an original feature map, and wherein the original feature map and the normalized feature map are generated separately by one or more preceding neural network layers; generate a first set of distribution parameters using a first set of neural network layers based on the normalized feature map; expand the first set of distribution parameters to dimensionally-match the normalized feature map; scale the normalized feature map by the expanded first set of distribution parameters to generate a scaled feature map;generate a second set of distribution parameters using a second set of neural network layers based on the normalized feature map; expand the second set of distribution parameters to dimensionally-match the scaled feature map; add the scaled feature map to the expanded second set of distribution parameters to generate a transformed feature map; and output the transformed feature map to a succeeding neural network layer.

54. An apparatus comprising at least one processor configured to perform the method of any one of claims 1-22, 27-31, and 34-51.

55. An apparatus comprising a computer-readable medium storing instructions for causing one or more processors to perform the method of any one of claims 1-22, 27-31 , and 34-51.

56. An apparatus comprising at least one processor and at least one non-transitory computer-readable medium storing instructions for causing the at least one processor to perform the method of any one of claims 1-22, 27-31 , and 34-51.

57. A computer-readable medium storing a feature map generated according to any one of claims 1-22, 27-31 , and 34-51.

58. A signal including a feature map generated according to any one of claims 1-22, 27-31 , and 34-51.

Images