Feature Compression for Machine-Oriented Video Coding
The method of reordering and encoding feature channels in video coding for machines addresses inefficiencies in machine vision tasks by optimizing compression and interoperability through characteristic-based rearrangement and encoding.
Patent Information
- Application Number
- JP2023528039
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2022-09-08
- Filing Date
- 2022-09-09
- Publication Date
- 2024-11-06
- Estimated Expiration
- 2042-09-09
AI Technical Summary
Existing video coding technologies are inefficient for machine vision tasks due to the large data volume and varying characteristics of feature channels in video data, leading to suboptimal compression and interoperability between devices.
A method for feature compression that involves obtaining a feature map, reordering feature channels based on characteristics such as mean, variance, or correlation coefficients, and encoding the reordered channels using video coding techniques like HEVC or VVC, with signaling to maintain order.
Improves compression efficiency and interoperability by rearranging feature channels for better alignment and encoding, reducing data volume and power consumption in video coding for machines.
Smart Images

Figure 0007578380000003 
Figure 0007578380000004 
Figure 0007578380000005
Abstract
Description
[Technical field]
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority from U.S. Provisional Application No. 63 / 246,217, filed with the U.S. Patent and Trademark Office on September 20, 2021, the disclosure of which is incorporated herein by reference in its entirety.
[0002] Technical Field SUMMARY OF THE DISCLOSURE Embodiments of the present disclosure are directed to video coding, and more specifically, to feature compression consistent with machine-directed video coding (VCM). [Background technology]
[0003] Videos and images may be consumed by human users for a variety of purposes, e.g., entertainment, education, etc. Therefore, video and image coding often exploits the properties of the human visual system for better compression efficiency while maintaining good subjective quality.
[0004] In recent years, with the rise of machine learning applications coupled with sensor abundance, many intelligent platforms are utilizing video for machine vision tasks such as object detection, segmentation, or tracking. As a result, encoding videos and images for consumption by machine tasks has become an interesting and challenging problem. This has led to the introduction of Video Coding for Machines (VCM) research. To achieve this goal, the international standards group MPEG created an ad-hoc group "Video coding for machines (VCM)" to standardize related techniques for better interoperability between different devices. Summary of the Invention
[0005] In accordance with one or more embodiments, a method for performing feature compression is executed by at least one processor and includes the steps of obtaining an input video; obtaining a feature map corresponding to the input video, the feature map including a plurality of feature channels; reordering the plurality of feature channels based on at least one characteristic of the plurality of feature channels; compressing the reordered plurality of feature channels; and generating an encoded bitstream based on the compressed and reordered plurality of feature channels.
[0006] In accordance with one or more embodiments, a device for performing feature compression includes at least one memory configured to store program code; and at least one processor configured to read the program code and to operate as directed by the program code, the program code including: a first acquisition code configured to cause the at least one processor to acquire an input video; a second acquisition code configured to cause the at least one processor to acquire a feature map corresponding to the input video, the feature map including a plurality of feature channels; a reordering code configured to cause the at least one processor to reorder the plurality of feature channels based on at least one characteristic of the plurality of feature channels; a compression code configured to cause the at least one processor to compress the reordered plurality of feature channels; and a generation code configured to cause the at least one processor to generate an encoded bitstream based on the compressed and reordered plurality of feature channels.
[0007] In accordance with one or more embodiments, a non-transitory computer-readable medium stores instructions that, when executed by one or more processors of a device that performs feature compression, cause the one or more processors to: obtain an input video; obtain a feature map corresponding to the input video, the feature map including a plurality of feature channels; reorder the plurality of feature channels based on at least one characteristic of the plurality of feature channels; compress the reordered plurality of feature channels; and generate an encoded bitstream based on the compressed and reordered plurality of feature channels. [Brief description of the drawings]
[0008] Further features, nature and various advantages of the disclosed subject matter will become more apparent from the following detailed description and the accompanying drawings. [Figure 1] FIG. 1 is a diagram of an environment in which the methods, apparatus, and systems described herein may be implemented, according to an embodiment. [Diagram 2] 2 is a block diagram of example components of one or more devices of FIG. 1. [Diagram 3] 1 is a block diagram of an exemplary architecture for performing video coding, according to an embodiment. [Figure 4] FIG. 2 is a block diagram of an exemplary feature map, according to an embodiment. [Diagram 5] 1 is a flowchart of an exemplary process for performing feature compression, according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0009] Figure 1 is a diagram of an environment 100 in which the methods, apparatus, and systems described herein may be implemented, according to an embodiment. As shown in Figure 1, the environment 100 may include a user device 110, a platform 120, and a network 130. The devices of the environment 100 may be interconnected via wired connections, wireless connections, or a combination of wired and wireless connections.
[0010] User device 110 includes one or more devices that can receive, generate, store, process, and / or provide information associated with platform 120. For example, user device 110 may include a computing device (e.g., a desktop computer, a laptop computer, a tablet computer, a handheld computer, a smart speaker, a server, etc.), a mobile phone (e.g., a smartphone, a wireless phone, etc.), a wearable device (e.g., smart glasses or a smart watch), or a similar device. In some implementations, user device 110 can receive information from and / or transmit information to platform 120.
[0011] Platform 120 includes one or more devices, as described elsewhere herein. In some implementations, platform 120 may include a cloud server or a group of cloud servers. In some implementations, platform 120 may be designed to be modular, such that software components may be swapped in or out depending on particular needs. As such, platform 120 may be easily and / or quickly reconfigured for different uses.
[0012] In some implementations, as shown, platform 120 may be hosted in a cloud computing environment 122. Notably, although the implementations described herein describe platform 120 as being hosted in a cloud computing environment 122, in some implementations platform 120 may not be cloud-based (i.e., may be implemented outside of a cloud computing environment) or may be partially cloud-based.
[0013] Cloud computing environment 122 includes an environment that hosts platform 120. Cloud computing environment 122 can provide services, such as computing, software, data access, storage, etc., that do not require end-user (e.g., user device 110) knowledge of the physical location and configuration of the systems and / or devices that host platform 120. As shown, cloud computing environment 122 can include a group of computing resources 124 (collectively referred to as “computing resources 124” and individually referred to as “computing resource 124”).
[0014] The computational resources 124 include one or more personal computers, workstation computers, server devices, or other types of computing and / or communication devices. In some implementations, the computational resources 124 may host the platform 120. Cloud resources may include computational instances running on the computational resources 124, storage devices provided on the computational resources 124, data transfer devices provided on the computational resources 124, etc. In some implementations, the computational resources 124 may communicate with other computational resources 124 via wired connections, wireless connections, or a combination of wired and wireless connections.
[0015] As further shown in FIG. 1, the computational resources 124 include a group of cloud resources, such as one or more applications (“APP”) 124-1, one or more virtual machines (“VM”) 124-2, virtualized storage (“VS”) 124-3, and one or more hypervisors (“HYP”) 124-4.
[0016] Application 124-1 includes one or more software applications that may be provided to or accessed by user device 110 and / or platform 120. Application 124-1 may eliminate the need to install and run a software application on user device 110. For example, application 124-1 may include software associated with platform 120 and / or any other software that may be provided via cloud computing environment 122. In some implementations, one application 124-1 may send / receive information to / from one or more other applications 124-1 via virtual machine 124-2.
[0017] Virtual machine 124-2 includes a software implementation of a machine (e.g., a computer) that executes programs like a physical machine. Virtual machine 124-2 may be either a system virtual machine or a process virtual machine, depending on the usage and degree of correspondence of virtual machine 124-2 to any real machine. A system virtual machine can provide a complete system platform that supports the execution of a complete operating system ("OS"). A process virtual machine can execute a single program and support a single process. In some implementations, virtual machine 124-2 can run on behalf of a user (e.g., user device 110) and manage the infrastructure of cloud computing environment 122, such as data management, synchronization, long-term data transfer, etc.
[0018] Virtualized storage 124-3 includes one or more storage systems and / or one or more devices that use virtualization techniques within the storage systems or devices of the computing resources 124. In some implementations, within the context of storage systems, types of virtualization may include block virtualization and file virtualization. Block virtualization may refer to the abstraction (or separation) of logical storage from physical storage, such that a storage system may be accessed without regard to physical storage or heterogeneous structures. This separation may allow a storage system administrator flexibility in how they manage storage for end users. File virtualization may eliminate the dependency between data accessed at the file level and where the file is physically stored. This may enable optimization of storage usage, server consolidation, and / or performing non-disruptive file migration.
[0019] The hypervisor 124-4 may provide hardware virtualization technology that allows multiple operating systems (e.g., "guest operating systems") to run simultaneously on a host computer, such as the computational resource 124. The hypervisor 124-4 may present a virtual operating system platform to the guest operating systems and may manage the execution of the guest operating systems. Multiple instances of different operating systems may share virtualized hardware resources.
[0020] Network 130 may include one or more wired and / or wireless networks, such as, for example, a cellular network (e.g., a fifth generation (5G) network, a long-term evolution (LTE) network, a third generation (3G) network, a code division multiple access (CDMA) network, etc.), a public land mobile network (PLMN), a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), a telephone network (e.g., a public switched telephone network (PSTN)), a private network, an ad-hoc network, an intranet, the Internet, a fiber optic based network, etc., and / or a combination of these or other types of networks.
[0021] The number and arrangement of devices and networks shown in Figure 1 are provided as an example. In practice, there may be additional, fewer, different, or otherwise arranged devices and / or networks than those shown in Figure 1. Furthermore, two or more of the devices shown in Figure 1 may be implemented within a single device, or a single device shown in Figure 1 may be implemented as multiple distributed devices. Additionally or alternatively, a set of devices (e.g., one or more devices) of environment 100 may perform one or more functions that are described as being performed by another set of devices in environment 100.
[0022] Figure 2 is a block diagram of example components of one or more devices of Figure 1. Device 200 may correspond to user device 110 and / or platform 120. As shown in Figure 2, device 200 may include a bus 210, a processor 220, a memory 230, a storage component 240, an input component 250, an output component 260, and a communication interface 270.
[0023] The bus 210 includes components that enable communication between the components of the device 200. The processor 220 is implemented in hardware, firmware, or a combination of hardware and software. The processor 220 is a central processing unit (CPU), a graphics processing unit (GPU), an accelerated processing unit (APU), a microprocessor, a microcontroller, a digital signal processor (DSP), a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), or another type of processing component. In some implementations, the processor 220 includes one or more processors that can be programmed to perform functions. The memory 230 includes random access memory (RAM), read only memory (ROM), and / or another type of dynamic or static storage device (e.g., flash memory, magnetic memory, and / or optical memory) that stores information and / or instructions for use by the processor 220.
[0024] The storage component 240 stores information and / or software related to the operation and use of the device 200. For example, the storage component 240 may include a hard disk (e.g., a magnetic disk, an optical disk, a magneto-optical disk, and / or a solid-state disk), a compact disk (CD), a digital versatile disk (DVD), a floppy disk, a cartridge, a magnetic tape, and / or another type of non-transitory computer-readable medium along with a corresponding drive.
[0025] Input components 250 include components that enable device 200 to receive information, such as via user input (e.g., a touchscreen display, a keyboard, a keypad, a mouse, buttons, switches, and / or a microphone). Additionally or alternatively, input components 250 may include sensors that sense information (e.g., a global positioning system (GPS) component, an accelerometer, a gyroscope, and / or an actuator). Output components 260 include components that provide output information from device 200 (e.g., a display, a speaker, and / or one or more light emitting diodes (LEDs)).
[0026] Communications interface 270 includes transceiver-like components (e.g., a transceiver, and / or a separate receiver and transmitter) that enable device 200 to communicate with other devices, such as via a wired connection, a wireless connection, or a combination of wired and wireless connections. Communications interface 270 may enable device 200 to receive information from and / or provide information to another device. For example, communications interface 270 may include an Ethernet interface, an optical interface, a coaxial interface, an infrared interface, a radio frequency (RF) interface, a universal serial bus (USB) interface, a Wi-Fi interface, a cellular network interface, etc.
[0027] Device 200 can perform one or more operations described herein. Device 200 may perform these operations in response to processor 220 executing software instructions stored by a non-transitory computer-readable medium, such as memory 230 and / or storage component 240. A computer-readable medium is defined herein as a non-transitory memory device. A memory device includes memory space within a single physical storage device or memory space spread across multiple physical storage devices.
[0028] The software instructions can be read into the memory 230 and / or storage component 240 from another computer-readable medium via the communications interface 270 or from another device. When executed, the software instructions stored in the memory 230 and / or storage component 240 can cause the processor 220 to perform one or more operations described herein. Additionally or alternatively, hardwired circuitry may be used in place of or in combination with software instructions to perform one or more operations described herein. Thus, the implementations described herein are not limited to any specific combination of hardware circuitry and software.
[0029] The number and arrangement of components shown in Figure 2 are provided as an example. In practice, device 200 may include additional, fewer, different, or differently arranged components than those shown in Figure 2. Additionally or alternatively, a set of components (e.g., one or more components) of device 200 may perform one or more functions that are described as being performed by another set of components of device 200.
[0030] 3 is a block diagram of an example architecture 300 for performing video coding, according to an embodiment. In an embodiment, architecture 300 may be a machine-directed video coding (VCM) architecture or an architecture that is otherwise compatible with or configured to perform VCM coding. For example, architecture 300 may be compatible with "Use cases and requirements for Video Coding for Machines" (ISO / IEC JTC 1 / SC 29 / WG 2 N18), "Draft of Evaluation Framework for Video Coding for Machines" (ISO / IEC JTC 1 / SC 29 / WG 2 N19), and "Call for Evidence for Video Coding for Machines" (ISO / IEC JTC 1 / SC 29 / WG 2 N20), the disclosures of which are incorporated herein by reference in their entireties.
[0031] In an embodiment, one or more of the elements shown in FIG. 3 may correspond to or be implemented by one or more of the elements discussed above with respect to FIGS. 1-2, such as one or more of user device 110, platform 120, device 200, or any of the elements contained therein.
[0032] As seen in FIG. 3, the architecture 300 may include a VCM encoder 310 and a VCM decoder 320. In an embodiment, the VCM encoder may receive a sensor input 301, which may include, for example, one or more input images or an input video. The sensor input 301 may be provided to a feature extraction module 311, which may extract features from the sensor input, which may be transformed using a feature transformation module 312 and encoded using a feature encoding module 313. In an embodiment, the term "encoding" may include, correspond to, or be used interchangeably with the term "compression." The architecture 300 may include an interface 302, which may enable the feature extraction module 311 to interface with a neural network (NN) that may assist in performing the feature extraction.
[0033] The sensor input 301 may be provided to a video encoding module 314, which may generate an encoded video. In an embodiment, after the features are extracted, transformed, and encoded, the encoded features may be provided to the video encoding module 314, which may use the encoded features to assist in generating the encoded video. In an embodiment, the video encoding module 314 may output the encoded video as an encoded video bitstream, and the feature encoding module 313 may output the encoded features as an encoded feature bitstream. In an embodiment, the VCM encoder 310 may provide both the encoded video bitstream and the encoded feature bitstream to a bitstream multiplexer 315, which may generate the encoded bitstream by combining the encoded video bitstream and the encoded feature bitstream.
[0034] In an embodiment, the encoded bitstream may be received by a bitstream demultiplexer (demux), which may separate the encoded bitstream into an encoded video bitstream and an encoded feature bitstream, which may be provided to a VCM decoder 320. The encoded feature bitstream may be provided to a feature decoding module 322, which may generate decoded features, and the encoded video bitstream may be provided to a video decoding module, which may generate a decoded video. In an embodiment, the decoded features may be further provided to a video decoding module 323, which may use the decoded features to assist in generating the decoded video.
[0035] In an embodiment, the output of the video decoding module 323 and the feature decoding module 322 may be primarily used for machine consumption, e.g., machine vision module 332. In an embodiment, the output may also be used for human consumption, shown in FIG. 3 as human vision 331. From the client end, e.g., from the VCM decoder 320 side, a VCM system, e.g., architecture 300, may perform video decoding to first obtain a sample domain video. One or more machine tasks for understanding the video content may then be performed, e.g., by machine vision module 332. In an embodiment, architecture 300 may include an interface 303, which may enable the machine vision module 332 to interface with a NN that may assist in performing one or more machine tasks.
[0036] As shown in FIG. 3, in addition to the video encoding and decoding path including the video encoding module 314 and the video decoding module 323, another path included in the architecture 300 is the feature extraction, feature encoding, and feature decoding path, which includes the feature extraction module 311, the feature transformation module 312, the feature encoding module 313, and the feature decoding module 322. Examples related to these paths will be discussed in more detail below.
[0037] In an embodiment, in the VCM encoder 310, features extracted from a source image or a source video, such as the sensor input 301, can be directly utilized by a machine vision network in the VCM decoder 320 or in the machine vision module 332. In this case, there may be no need for feature extraction in the VCM decoder 320. Therefore, in the VCM encoder 310, power savings can be achieved, especially when the features can be used for multiple machine vision tasks.
[0038] A feature map, or a feature map signal, extracted from an input image or an input video can be represented as a three-dimensional array, which may be denoted as F n,h,w as shown, where n, h, w are non-negative integers, 0 ≦ n < N, 0 ≦ h < H, 0 ≦ w < W, N represents the number of feature channels, H represents the height of each feature channel, and W represents the width of each feature channel. FIG. 4 is a block diagram of an exemplary feature map 400 according to an embodiment. As shown in FIG. 4, the feature map 400 can include N channels 400-1, 400-2, 400-3,..., 400-n corresponding to n = 0, n = 1, n = 2,..., n = N-1, respectively.
[0039] In an embodiment, F n,h,wIf x, y, and z are expressed as floating-point numbers, they may be converted to integers through average subtraction, scaling, and quantization processes. Similar to source images or videos, feature maps may be compressed in the VCM encoder 310, for example using the feature encoding module 313, before being sent to the VCM decoder 320 due to their large data volume. Therefore, feature compression is also an important topic in VCM.
[0040] However, the feature map F n,h,w In a feature map, different feature channels may exhibit different characteristics. Adjacent feature channels are not necessarily similar or highly correlated. Therefore, it may be desirable to reorder feature channels in a feature map such that similar or correlated feature channels are placed adjacent to each other for better compression efficiency. Thus, embodiments of the present disclosure may relate to methods of reordering feature channels for feature compression and corresponding signaling methods.
[0041] In an embodiment, the feature channels in the feature map can be reordered based on the characteristics of each feature channel. For example, the mean or variance of each feature channel in the feature map may be calculated, and the feature channels may be reordered based on the calculated feature channel mean or feature channel variance, for example, in order of increasing or decreasing feature channel mean or feature channel variance, or based on any other statistically based criteria. The mean may be considered as the first moment of the feature channel, and the variance may be considered as the second moment of the feature channel. In an embodiment, higher order moments, for example, the third moment, may be further used to reorder the feature channels.
[0042] In an embodiment, the feature channels in the feature map may be reordered based on a metric that measures the similarity between the feature channels. For example, a correlation coefficient may be calculated between a pair of feature channels, such as the i-th feature channel and the j-th feature channel. In an embodiment, the correlation coefficient may correspond to the ratio between the covariance of the pair of feature channels and the product of the standard deviations of the pair of feature channels.
[0043] In an embodiment, the correlation coefficient may be calculated as shown in Equation 1 below.
number
[0044] In Equation 1, ρ i,j may represent the correlation coefficient between the i-th feature channel and the j-th feature channel, and μ i can represent the average of the i-th feature channel, and μ j can represent the mean of the jth feature channel, and σ i can represent the standard deviation of the i-th feature channel, and σ j may represent the standard deviation of the jth feature channel.
[0045] After computing the pair-wise feature channel correlation coefficients, one or more anchor feature channels, e.g., K anchor feature channels, can be selected, and the rest of the feature channels can be clustered based on their correlation coefficients with the K anchor feature channels. For example, the K anchor feature channels can be denoted as a(k), k=0,...,K-1, where a(k) represents the feature channel index of the kth anchor feature channel. The mth feature channel can be classified as part of a(k)'s cluster if it has the largest correlation coefficient with a(k) compared to its correlation coefficient with a(1),...,a(k-1),a(k+1),...,a(K-1) when m is not in the set {a(0),a(1),...,a(K-1)}. Within each cluster, the feature channels can be sorted based on feature channel characteristics, such as feature channel mean or variance. The clusters may be sorted based on the cluster mean or variance, or based on the correlation coefficient between a(k), k=1,...,K-1 and a(0), where the cluster of a(0) is placed first, without loss of generality.
[0046] In an embodiment, various methods can be used to select the K anchor feature channels. For example, when the N feature channels are sorted based on feature channel characteristics such as mean or variance, every N / Kth feature channel may be selected as the K feature channels. In an embodiment, apart from the above clustering methods, other clustering methods may be used, for example, a hierarchical clustering method based on correlation distance.
[0047] In an embodiment, for each feature channel, the correlation coefficients of the feature channel with all of the other feature channels can be summed to obtain a total correlation coefficient. All of the feature channels can then be reordered based on the total correlation coefficients, for example in decreasing or increasing order.
[0048] In embodiments, other feature channel similarity metrics such as the Levene Test or KL divergence may be used to reorder the feature channels.
[0049] In the VCM encoder 310, the feature channels can be reordered as discussed above, and the reordered feature channels can be compressed using a video coding technique, such as high efficiency video coding (HEVC), versatile video coding (VVC), or any other video coding technique desired. In addition to the coded bitstream, information about the order of the feature channels in the feature map can be signaled to enable the VCM decoder 320 to restore the original order of the feature channels after all of the feature channels have been decoded.
[0050] In an embodiment, information about the order of the reordered feature channels can be explicitly signaled using a permutation sequence of numbers {0, 1, ..., N-1} corresponding to the N feature channels, denoted as {P(0), P(1), ..., P(N-1)}, where P(k) for k=0, 1, ..., N-1 indicates its original order in the feature map before reordering.
[0051] In an embodiment, information about only the feature channels that change their position relative to the original position may be signaled. For example, the kth feature channel in the reordered feature map may be the P(k)th feature channel in its original order. In this case, information indicating the pair (k, P(k)) may be signaled in the bitstream. When the portion of feature channels that change position is not large, signaling information about the value pairs of the feature channels that change position may be more efficient in terms of bit savings.
[0052] In an embodiment, for the reordered feature channels, the difference value between the original position and the new position may be signaled. For example, information indicating {kP(k)} for k=0,1,...,N-1 may be signaled in the bitstream. If most of the feature channels do not change position, the corresponding kP(k)=0, which may use fewer bits to signal the reordering relationship in the bitstream.
[0053] In an embodiment, the order of the reordered feature channels may be signaled implicitly. For example, if feature channel averages are used to sort the source feature channels in the feature map, the average values of each feature channel may be signaled in the original order in the bitstream. The VCM decoder 320 may then sort the average values to determine how the feature channels were reordered. In an embodiment, if variance or another metric is used to sort the feature channels, a similar method may be applied.
[0054] As discussed above, the feature channels in the feature map are sorted and fed to an encoder, such as the feature encoding module 313, which can then determine a reference feature channel for a given feature channel based on its algorithm. In an embodiment, the feature channels can be treated as pictures when coded by a two-dimensional video codec such as HEVC, VVC, etc. For example, the feature encoding module 313 can first divide the feature channels into one or more group of pictures (GOP) structures and construct one or more reference picture lists in the GOP, such as a hierarchical reference picture list.
[0055] In an embodiment, the feature encoding module 313 may determine one or more reference feature channels for a given feature channel based on the feature channel similarity metric described above. For example, before encoding, all feature channels may be in the original order or may be reordered based on feature channel characteristics such as mean or variance. During the encoding process, the feature encoding module 313 may still divide the feature channels into GOPs or treat all feature channels as one GOP. Within a GOP, the feature channel with a larger total feature channel coefficient may be encoded first. When selecting a reference feature channel for a feature channel to be coded, only the feature channel with a relatively large similarity metric may be chosen. The selected codec may signal the result of the reordering of the reference picture list using its own standard-compliant method.
[0056] In an embodiment, the encoder may treat the K clusters mentioned above as K GOPs and encode them independently, because the feature channels in different clusters may be more dissimilar and there may be less benefit in referencing feature channels in other clusters.
[0057] 5 is a flow chart of an example process 500 for performing feature compression. In some implementations, one or more of the processing blocks in FIG. 5 may be performed by any of the elements discussed above, such as the VCM encoder 310, or elements contained therein, such as the feature extraction module 311, the feature transformation module 312, or the feature encoding module 313.
[0058] As shown in FIG. 5, process 500 may include obtaining an input video (block 502).
[0059] 5, the process 500 may include obtaining a feature map corresponding to the input video (block 504). In an embodiment, the feature map may include multiple feature channels.
[0060] As further shown in FIG. 5, the process 500 may include reordering the multiple feature channels based on at least one characteristic of the multiple feature channels (block 506).
[0061] 5, the process 500 may include compressing the reordered feature channels (block 508). In an embodiment, compressing the reordered feature channels may include or correspond to encoding the multiple feature channels.
[0062] As further shown in FIG. 5, the process 500 may include generating an encoded bitstream based on the reordered and compressed multiple feature channels (block 510).
[0063] In an embodiment, the at least one characteristic may include at least one of an average of the plurality of feature channels, a variance of the plurality of feature channels, and a similarity of the plurality of feature channels. In an embodiment, the similarity may be determined based on a correlation coefficient between two feature channels of the plurality of feature channels, and the correlation coefficient may correspond to a ratio between a covariance of the two feature channels and a product of standard deviations of the two feature channels.
[0064] In an embodiment, the two feature channels may include the i-th feature channel and the j-th feature channel of the multiple feature channels, and the correlation coefficient may be calculated according to Equation 1 above.
[0065] In an embodiment, the multiple feature channels may be sorted into at least one picture group, and at least one reference picture used to generate the encoded bitstream may be selected based on a correlation coefficient.
[0066] In an embodiment, the reordering may include selecting at least one anchor feature channel from among the plurality of feature channels, and clustering remaining feature channels from the plurality of feature channels into at least one cluster based on the at least one anchor feature channel.
[0067] In an embodiment, the reordering may further include ordering the remaining feature channels in the at least one cluster based on at least one of the mean of the remaining feature channels, the variance of the remaining feature channels, and the similarity of the remaining feature channels.
[0068] In an embodiment, the reordering may include calculating a total correlation coefficient for each feature channel of the plurality of feature channels, and ordering the plurality of feature channels based on the total correlation coefficient.
[0069] In an embodiment, the process 500 may further include signaling the order of the reordered feature channels in the encoded bitstream.
[0070] In an embodiment, the order may be signaled by indicating the feature channel that has changed position within the reordered plurality of feature channels.
[0071] In an embodiment, the order may be signaled implicitly.
[0072] 5 illustrates example blocks of process 500, in some implementations process 500 may include additional, fewer, different, or differently arranged blocks relative to those illustrated in FIG 5. Additionally or alternatively, two or more of the blocks of process 500 may be performed in parallel.
[0073] Additionally, the proposed methods may be implemented by a processing circuit (e.g., one or more processors or one or more integrated circuits). In one example, the one or more processors execute a program stored on a non-transitory computer-readable medium to perform one or more of the proposed methods.
[0074] The techniques described above may be implemented as computer software using computer readable instructions and physically stored on one or more computer readable media.
[0075] The embodiments of the present disclosure may be used separately or combined in any order. Furthermore, each of the embodiments (and methods thereof) may be implemented by processing circuitry (e.g., one or more processors or one or more integrated circuits). In one example, the one or more processors execute a program stored on a non-transitory computer-readable medium.
[0076] The foregoing disclosure provides illustration and description, but is not intended to be exhaustive or to limit the implementation to the precise form disclosed. Modifications and variations are possible in light of the above disclosure or may be acquired from practice of the implementations.
[0077] As used herein, the term component is intended to be broadly interpreted as hardware, firmware, or a combination of hardware and software.
[0078] Although combinations of features are recited in the claims and / or disclosed in the specification, these combinations are not intended to limit the disclosure of possible implementations. Indeed, many of these features can be combined in ways not specifically recited in the claims and / or disclosed in the specification. Although each dependent claim listed below may depend directly on only one claim, the disclosure of possible implementations includes each dependent claim in combination with every other claim in the claim set.
[0079] No element, act, or instruction used herein should be construed as critical or required unless expressly described as such. Also, as used herein, the articles "a" and "an" are intended to include one or more items and may be used interchangeably with "one or more." Additionally, as used herein, the term "set" is intended to include one or more items (e.g., related items, unrelated items, a combination of related and unrelated items, etc.) and may be used interchangeably with "one or more." Where only one item is intended, the term "a" or similar language is used. Also, as used herein, the terms "has," "have," "having," and the like are intended to be open-ended terms. Additionally, the phrase "based on" is intended to mean "based at least in part on," unless expressly indicated otherwise.
Claims
1. 1. A method of performing feature compression, the method being performed by at least one processor, comprising: obtaining an input video; obtaining a feature map corresponding to the input video, the feature map including a plurality of feature channels; reordering the plurality of feature channels based on at least one characteristic of the plurality of feature channels; compressing the reordered feature channels; generating an encoded bitstream based on the reordered and compressed feature channels; Including, the at least one characteristic includes a similarity of the plurality of feature channels; the similarity is determined based on a correlation coefficient between two feature channels of the plurality of feature channels; A method according to claim 1, wherein the correlation coefficient corresponds to a ratio between the covariance of the two feature channels and the product of the standard deviations of the two feature channels.
2. the two feature channels include an i-th feature channel and a j-th feature channel among the plurality of feature channels; The correlation coefficient is calculated using the following formula: [0010] where ρ represents the correlation coefficient, F represents a three-dimensional array representing the feature map, W represents the width of the feature map, w represents a width index, H represents the height of the feature map, h represents a height index, and μ i represents the average of the i feature channel, and μ j represents the average of the jth feature channel, and σ i represents the standard deviation of the i feature channel, and σ j The method of claim 1 , wherein j represents the standard deviation of the j feature channel.
3. 2. The method of claim 1, wherein the plurality of feature channels are sorted into at least one picture group, and at least one reference picture used to generate the encoded bitstream is selected based on the correlation coefficient.
4. 2. The method of claim 1 , wherein the reordering comprises selecting at least one anchor feature channel from among the plurality of feature channels, and clustering remaining feature channels from the plurality of feature channels into at least one cluster based on the at least one anchor feature channel.
5. 5. The method of claim 4, wherein the reordering further comprises ordering the remaining feature channels in the at least one cluster based on at least one of a mean of the remaining feature channels, a variance of the remaining feature channels, and a similarity of the remaining feature channels.
6. The method of claim 1 , wherein the reordering comprises: calculating a total correlation coefficient for each feature channel of the plurality of feature channels; and ordering the plurality of feature channels based on the total correlation coefficient.
7. The method of claim 1 , further comprising the step of signaling an order of the reordered feature channels in the encoded bitstream.
8. The method of claim 7 , wherein the order is signaled by indicating feature channels that have changed position within the reordered plurality of feature channels.
9. the order is implicitly signaled; and implicitly signaling the order includes signaling an average value of each feature channel in an original order in the encoded bitstream if a channel average is used to sort the multiple feature channels in the feature map. The method of claim 7.
10. 1. A device for performing feature compression, comprising: at least one memory configured to store program code; at least one processor configured to read said program code and to execute said program code to perform the method of any one of claims 1 to 9; The device that contains
11. A computer program product causing one or more processors to carry out a method according to any one of claims 1 to 9.
Citation Information
Patent Citations
Signaling feature map data
JP2023542876A
Image processing method, data processing method, image processing device, and program
WO2021095245A1
Signaling of feature map data
WO2022086376A1
