Method for segmenting a plurality of data, and corresponding coding method, decoding method, devices, systems and computer program

Abstract

The invention relates to a method for segmenting a plurality of input data, the method comprising: - determining (61) weight values to be applied to the plurality of input data before it is processed by at least one processing device configured to produce a processing result according to a criterion for optimising a quality of the processing result, as a function of the said criterion and a criterion for optimising a quantity of input data to be processed; - determining (62) segmentation information for the plurality of input data, assigned a first value or a second value, depending on the weights; and - obtaining (63) a subset of data to be processed by applying the segmentation information to the plurality of input data, which segmentation information comprises the data of the plurality of input data associated with an item of segmentation information equal to the first value.

Description

Description Method for segmenting a plurality of data, coding method, decoding method, corresponding devices, systems and computer programs Scope of the invention

[0001] The present invention relates generally to the field of processing a plurality of data, such as 3D images, multi-view type of a scene acquired by a plurality of cameras or physiological data of a patient from a plurality of sensors.

[0002] The invention relates in particular to the segmentation of this plurality of data before their coding and then their transmission via a communication network to a processing device. Previous art

[0003] In the field of virtual reality and immersive video, free navigation allows viewers, using a VR headset, to view a scene from any perspective, whether that perspective is captured by a camera or not. Such a view not captured by the camera is also called a virtual view or intermediate view because it lies between camera-captured views and must be synthesized from those captured views to render the scene to the viewer.

[0004] In an immersive video context, where the viewer has the sensation of being immersed in the scene, the scene SC is typically captured by a set of cameras, as illustrated in Figure 1. These cameras can be 2D (Ci, C2...C cameras). N(with N a non-zero integer from Figure 1), meaning that each of them captures a view from a specific viewpoint, or of the 360 ​​type, meaning that they capture the entire scene in 360 degrees around the camera (camera C 360 (from figure 1), therefore from several different points of view. The cameras can be arranged in an arc, a rectangle, or any other configuration that allows for good coverage of the scene.

[0005] In relation to Figure 2, at a given moment, we obtain a set of images representing the scene from different views. Since these are videos, temporal sampling of the captured images is performed (for example, at 30 frames per second) in order to produce an original multiview video (MVV), as illustrated in Figure 3.

[0006] The MIV standard (MPEG Immersive Video) enables the transmission of video adapted for immersive navigation. The encoder selects portions (patches) from each view to transmit in order to maximize the quality of view synthesis from these patches, while minimizing the amount of data transmitted. Patches are extracted from the views and assembled into one or more atlases, which are therefore images comprising an assembly of patches from different views. The patches are generally arranged in an atlas to fill it with in the most comprehensive way possible. Each atlas is accompanied by an occupancy map, which is an image where each pixel can take on a first or second value, distinct from the first (for example, corresponding to the color "white" or "black"), to indicate whether the atlas pixel belongs to a patch or not. This allows for the segmentation of the atlas patches from this occupancy map. Furthermore, other information is transmitted for each patch, including its coordinates within the atlas, as well as the view to which it is associated.

[0007] From the transmitted data, the MIV decoder can retrieve the patches and place them in the view to which they belong. This view is then called "partial," since it does not contain all the pixel values ​​of the original view as acquired by one of the cameras. However, if the encoder has effectively selected the portions of the views to transmit, they are sufficient to generate or synthesize any viewpoint of the scene. In this respect, view synthesis from the decoded views is not specified by the MIV standard. It relies on occupancy maps to determine whether a pixel in a given view contains relevant information or not.

[0008] We also know of a texture-based view synthesis method using neural networks from the paper by Wang et al., entitled "IBRNet: Learning Multi-View Image-Based Rendering," published by arXiv:2102.13090v2 in April 2021. Specifically, the paper describes a neural network, called IBRN, which takes as input camera parameters, texture images captured from different viewpoints, and the coordinates of the viewpoint of the scene to be synthesized. It then produces as output the synthesized view, corresponding to the view that a camera would have captured from the viewpoint corresponding to the provided coordinates. One advantage of this method is that it yields very high-quality results.

[0009] One drawback of this method, and more generally of current image synthesis methods, is that they require all the images acquired by the cameras. This represents a large amount of data to transmit and then decode, which poses a complexity problem at the decoder level, particularly when it is embedded in a mobile device such as a smartphone or an augmented reality headset.

[0010] There is therefore a need for a solution to reduce the amount of data, and in particular the number of pixels to be transmitted, while preserving as much as possible the quality of the view synthesis.

[0011] The invention improves the situation. Description of the invention

[0012] The invention addresses this need by proposing a method for segmenting a plurality of data acquired by sensors, referred to as input data, said method comprising: - determining the weight values ​​to be applied to the plurality of input data before processing by at least one processing device configured to produce a processing result according to an optimization criterion of the quality of the processing result of the input data, said weight values ​​being determined according to said criterion and an optimization criterion of the quantity of input data to be processed, - the determination of segmentation information for said plurality of input data, each segmentation piece of said data being assigned a first value or a second value distinct from the first, depending on said weights, and - obtaining a subset of data to be processed by applying the segmentation information determined to said plurality of input data, the subset of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the first value.

[0013] The invention proposes a completely new and inventive approach to segmenting a plurality of input data before processing by a given processing device. This approach consists of configuring weight values ​​to be applied to the input data so that both the quantity of data resulting from the segmentation and presented as input to the processing device, as well as the quality of processing, are optimized. The weight values ​​are determined from the plurality of input data themselves and are therefore specifically chosen for them.

[0014] For example, the optimization criterion for the quantity of input data to be processed includes minimizing the cumulative value of weight values ​​or the number of weight values ​​below a given threshold. In one variation, it includes minimizing the cumulative value of the input data retained in the subset of data to be processed or a number of these input data. For example, the optimization criterion for processing quality includes minimizing the mean squared error between a result obtained from the plurality of input data and a result obtained from the subset of data to be processed, or maximizing the signal-to-noise ratio (PSNR).

[0015] The weight values ​​of the resulting configuration are then used to determine one or more segmentation pieces of information for the plurality of input data. These segmentation pieces of information indicate: - what input data should be provided to the processing device, because they are useful for it to produce a processing result that meets the optimization criterion for the quality of the input data processing result and - which input data should not be presented to the processing device because they are not useful, in the sense that they do not contribute to improving the quality of the treatment result obtained.

[0016] The segmentation information thus determined finally allows us to obtain the subset of data to be processed. In this way, only this subset of data to The processed data can be coded and transmitted to a receiver incorporating a similar processing device, in terms of structure and configuration.

[0017] Thus, the invention does not simply propose to perform efficient segmentation of the plurality of input data, but it also takes into account, when determining the weight values ​​that perform the segmentation, the impact of the weight values ​​of the segmentation module, which performs the segmentation, on the output of the processing device. It therefore does not consider a simple segmentation module independent of the processing device, but takes into account the combination of the two, and more precisely, their successive actions on the input data.

[0018] The invention applies to any type of data acquired by any type of sensor. For example, a plurality of sensors can be arranged around a scene, object, or subject. For instance, this could involve a plurality of cameras, each with distinct viewpoints of the scene, configured to acquire a sequence of images or views of that scene. In this case, the processing device can be a device for synthesizing additional views from the original views acquired by the plurality of cameras. The segmentation of the original views according to the invention allows only the data useful for synthesizing an additional view to be retained, while redundant data is eliminated.

[0019] In another example, the data plurality consists of time-series sequences of a patient's physiological measurements captured by a variety of sensors (electrocardiogram (ECG), electroencephalogram (EEG), CT scan, magnetic resonance imaging (MRI), X-ray image, blood composition indicator, etc.). In this other example, the treatment device could be a diagnostic aid device.

[0020] According to another aspect of the invention, the determination includes learning said weight values ​​from said plurality of input data, said learning being carried out by backpropagation of a gradient of a loss function combining the two criteria.

[0021] Advantageously, determining the weight values ​​for segmenting the input data employs a learning technique for an artificial intelligence module using the input data itself. More precisely, the artificial intelligence module, referred to hereafter as the segmentation module, learns the best possible internal configuration of its weights, thereby optimizing both the amount of data from the input subset presented to the processing system and the quality of the data processing.

[0022] The invention departs from conventional machine learning practices by implementing specific learning tailored to a plurality of input data. Therefore, it does not require the prior acquisition of a large, labeled training database, but only the plurality of common data points to be processed.

[0023] According to yet another aspect of the invention, said processing device comprises weights, called processing weights, and values ​​of said processing weights have been previously determined according to the criterion of optimizing the quality of the result of processing the input data, and the process further includes the determination of modified values ​​of said processing weights.

[0024] Advantageously, the processing system also implements an artificial intelligence technique, based on applying weights to the input data. For example, this processing system was previously trained, in a conventional manner, using a labeled database, to process the plurality of input data in such a way as to provide an output result that conforms to the processing quality optimization criterion.

[0025] Advantageously, the learning according to the invention is a combined learning of the segmentation module and the processing device. It includes an update of the processing device's configuration so that it can optimally process the subset of data resulting from the segmentation of the plurality of input data.

[0026] For example, the segmentation module and the processing device are each organized into layers comprising weights, and the succession of their respective layers forms an artificial intelligence module, for example a neural network, trained to segment the input data and then process it optimally.

[0027] According to yet another aspect of the invention, said plurality of input data comprises a plurality of views acquired by a plurality of cameras, each view comprises pixels, said weights are comprised in a plurality of layers, each layer is associated with each view and comprises each weight per pixel, and the segmentation information comprises a plurality of segmentation maps, each map being associated with each view.

[0028] The invention is particularly applicable to input data of the type multi-view video sequence. Advantageously, a layered segmentation module is considered, with one layer per view and one weight per pixel, the weights of each layer allowing the segmentation information of the pixels of the associated view to be derived.

[0029] For example, for images captured by cameras, segmentation information can take the form of segmentation maps, of the same dimensions as the input images, and whose useful pixels are valued at a first value (for example, corresponding to "white") and the non-useful pixels at a second value (for example, corresponding to "black").

[0030] Advantageously, the processing device is configured to synthesize an additional view associated with a given viewpoint and from at least one original view that it receives as input.

[0031] It is understood that in this case, the learning process according to the invention can consist of presenting, as input to the segmentation module combined with the processing device, the plurality of acquired views except for one, which will be the one to be synthesized. In this way, the loss function can be calculated by comparing the synthesized view to the original view. Advantageously, this procedure will be repeated for each of the acquired views.

[0032] According to another aspect of the invention, said plurality of input data comprises a plurality of measurement data sequences acquired by a plurality of sensors, said weights are included in a plurality of layers, each layer is associated with each measurement data sequence and comprises each weight per measurement data, and the segmentation information comprises a plurality of segmentation sequences, each segmentation sequence being associated with each measurement data sequence.

[0033] According to this other example, the data set consists of time sequences of a patient's physiological measurements captured by a variety of sensors. In this other example, the treatment system may include one or more diagnostic aids.

[0034] Correspondingly, the invention also relates to a device for segmenting a plurality of data acquired by sensors, referred to as input data, said device being configured to implement: - the determination of weight values ​​to be applied to the plurality of input data before processing by at least one processing device configured to produce a processing result according to an optimization criterion for the quality of the processing result of the input data, said weight values ​​being determined according to said criterion and an optimization criterion for the quantity of input data to be processed, - the determination of segmentation information for said plurality of input data, each segmentation piece of said data being assigned a first value or a second value distinct from the first, depending on said weights, and - obtaining a subset of data to be processed by applying the segmentation information determined to said plurality of input data, the subset of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the first value.

[0035] The aforementioned device implements the segmentation process according to the invention in its various embodiments.

[0036] Advantageously, said segmentation device is integrated into a server equipment configured to receive the plurality of input data and further comprising the aforementioned input data processing device.

[0037] The invention also relates to a method for encoding a plurality of data acquired by sensors, referred to as input data, comprising: - obtaining segmentation information for the plurality of input data and a subset of data to be processed by applying said segmentation information to the plurality of input data, said segmentation including the determination of weight values ​​to be applied to the plurality of input data before processing by at least one processing device previously configured to produce a processing result according to an optimization criterion of a quality of processing, said weight values ​​being determined according to said criterion and another criterion for optimizing a quantity of input data to be processed, said segmentation information of said data being valued at a first or second value distinct from the first, according to said weight values, the subset (USS) of data to be processed comprising the data from the plurality of input data associated with segmentation information equal to the first value, and - the coding of segmentation information and the subset of data to be processed.

[0038] With the invention, the coded data contains all the information necessary to reconstruct a plurality of segmented input data at the level of a receiver comprising a processing device similar to that implemented by the transmitter, in terms of structure and configuration.

[0039] According to another aspect of the invention, obtaining it further comprises: - obtaining modified weight values, said weights being intended to be applied to the input data by the processing device, the values ​​of said weights previously determined according to said criterion of optimization of the quality of the result of processing the input data and for the processing of the plurality of input data, having been modified according to said criteria, for the processing of the subset of data to be processed, - the coding of said modified weight values.

[0040] Advantageously, the invention proposes to transmit the modified weight values ​​of the processing device on the transmitter side within the coded data, in order to update the configuration of the processing device on the receiver side. This ensures that the processing device produces an optimal result according to the optimization criterion.

[0041] Correspondingly, the invention also relates to a device for encoding a plurality of data acquired by sensors, called input data, configured to implement: - obtaining segmentation information for the plurality of input data and a subset of data to be processed by applying said segmentation information to the plurality of input data, said segmentation including the determination of weight values ​​to be applied to the plurality of input data before processing by at least one processing device (previously configured to produce a processing result according to a criterion for optimizing the quality of the processing), said weight values ​​being determined according to said criterion and another criterion for optimizing the quantity of input data to be processed, said segmentation information for said data being assigned a first or second value distinct from the first, according to said weight values,the subset of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the first value, and, - the coding of segmentation information and the subset of data to be processed.

[0042] The aforementioned device implements the coding process according to the invention in its various embodiments.

[0043] Advantageously, the said coding device is integrated into the aforementioned server equipment.

[0044] The invention also relates to a method for decoding coded data, comprising: - the decoding of coded data, said coded data comprising segmentation information for a plurality of data acquired by sensors, said input data, a subset of data to be processed by a processing device configured to apply weights to the plurality of decoded input data and to produce a processing result according to a processing quality optimization criterion, said segmentation information for said input data being valued at a first value or a second value distinct from the first, said subset of data to be processed having been obtained by applying said segmentation information to the plurality of input data, the subset of data to be processed comprising the data from the plurality of input data associated with said segmentation information equal to the first value, said coded data further comprising modified values ​​of said weights,said modified values ​​having been determined for the processing of the plurality of segmented input data, according to the criterion of optimizing the quality of the processing and a criterion of optimizing the quantity of data in the subset of data to be processed, - the construction of a plurality of segmented input data decoded from the subset of decoded data to be processed and the decoded segmentation information, and - the provision of the plurality of segmented input data decoded and the modified values ​​of said weights to the processing device.

[0045] Advantageously, the invention proposes to encode the modified weight values ​​of the processing device on the transmitter side, for transmission to a receiver and updating of the processing device's configuration on the receiver side. This ensures that the processing device produces an optimal result according to the optimization criterion.

[0046] Correspondingly, the invention also relates to a decoding device comprising: - the decoding of coded data, said coded data comprising segmentation information for a plurality of data acquired by sensors, called input data, a subset of data to be processed by a processing device configured to apply weights to the plurality of decoded input data and to produce a processing result according to an optimization criterion of processing quality, said segmentation information for said input data being valued at a first value or a second distinct value of the first, said subset of data to be processed having been obtained by applying said segmentation information to the plurality of input data, the subset of data to be processed comprising the data from the plurality of input data associated with said segmentation information equal to the first value, said coded data further comprising modified values ​​of said weights, said modified values ​​having been determined for the processing of the plurality of segmented input data, according to the optimization criterion of a quality of processing and a criterion of optimization of a quantity of data of the subset of data to be processed, - the construction of a plurality of segmented input data decoded from the subset of decoded data to be processed and the decoded segmentation information, and - the provision of the plurality of segmented input data decoded and the modified values ​​of said weights to the processing device.

[0047] The aforementioned device implements the decoding process according to the invention in its various embodiments.

[0048] Advantageously, said decoding device is integrated into terminal equipment configured to receive the coded data and further comprising the aforementioned processing device.

[0049] The invention also relates to a signal carrying coded data. Said coded data comprises segmentation information for a plurality of data acquired by sensors, called input data, a subset of data to be processed obtained by applying said segmentation information to said plurality of input data, said segmentation information for said input data being valued at a first value or a second value distinct from the first, the subset of data to be processed comprising the data from the plurality of input data associated with segmentation information equal to the first value, said subset of data to be processed being intended to be decoded, then used to reconstruct a plurality of segmented input data decoded from the decoded segmentation information, for the purpose of processing said plurality of segmented input data by a processing device.configured to produce a processing result based on a processing quality optimization criterion, by applying weights to the plurality of said input data. Said encoded data further include modified values ​​of said weights, said modified values ​​having been determined for the processing of the plurality of segmented input data, based on the processing quality optimization criterion and a data quantity optimization criterion of the subset of data to be processed, and being intended to be used by said processing device to update said weights before processing the plurality of reconstructed decoded segmented input data.

[0050] The invention also relates to a system comprising the segmentation device, the encoding device, the decoding device and the processing device. as mentioned above. This system, the segmentation, coding and decoding devices according to the invention, present at least the same advantages as those conferred by the aforementioned methods.

[0051] The invention also relates to computer program products comprising program code instructions for the respective implementation of the aforementioned segmentation, encoding and decoding processes, when executed by a processor.

[0052] Such a program can use any programming language, and be in the form of source code, object code, or code intermediate between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0053] The invention also relates to at least one computer-readable recording medium on which computer programs are recorded, including program code instructions for executing the steps of the processes according to the invention as described above.

[0054] Such a recording medium can be any entity or device capable of storing the program. For example, the medium can include a storage means, such as a ROM, for example a CD-ROM or a microelectronic circuit ROM, or a magnetic recording means, for example a mobile medium (memory card) or a hard drive or an SSD.

[0055] On the other hand, such a recording medium can be a transmissible medium such as an electrical or optical signal, which can be transmitted via an electrical or optical cable, by radio, or by other means, so that the computer programs it contains can be executed remotely. The programs according to the invention can, in particular, be downloaded onto a network, for example, the Internet.

[0056] Alternatively, the recording medium or media may be one or more integrated circuits in which each program is incorporated, the circuit or circuits being adapted to execute or to be used in the execution of the aforementioned processes.

[0057] In an example implementation, this technique is implemented using software and / or hardware components. In this context, the term "module" may refer to a software component, a hardware component, or a set of hardware and software components.

[0058] A software component corresponds to one or more computer programs, one or more subroutines of a program, or more generally, any element of a program or software capable of implementing a function or set of functions, as described below for the module in question. Such a software component is executed by a data processor of a physical entity (terminal, server, gateway, set-top box, router, etc.) and is capable of accessing the hardware resources of that physical entity (memory, storage media, communication buses, input / output cards, user interfaces, etc.). Hereafter, "resources" refers to all sets of hardware and / or software elements supporting a function or service, whether they are single or combined.

[0059] Similarly, a hardware component corresponds to any element of a hardware assembly capable of implementing a function or set of functions, as described below for the module in question. This could be a programmable hardware component or one with an integrated processor for software execution, for example, an integrated circuit, a smart card, a memory card, an electronic board for running firmware, etc.

[0060] Each component of the system described above naturally implements its own software modules.

[0061] The different embodiments mentioned above can be combined with each other for the implementation of this technique. Brief description of the drawings

[0062] Other features and advantages of the present invention will become apparent from the description below, with reference to the accompanying drawings, which illustrate an example of an embodiment without being limiting in any way. In the figures:

[0063] [Fig 1]: presents an example of the arrangement of a plurality of cameras forming a multi-view video acquisition system of a scene, according to the prior art;

[0064] [Fig 2]: schematically illustrates a plurality of images of the scene, captured by the plurality of cameras at a given moment, according to the earlier art;

[0065] [Fig 3]: schematically illustrates a sequence of the plurality of images, captured by the plurality of cameras at several successive moments and forming the original multi-view video, according to the prior art;

[0066] [Fig 4]: schematically illustrates a first example of the architecture of a system according to an embodiment of the invention comprising a device for segmenting a plurality of input data, a device for encoding segmentation information and a subset of data to be processed from the segmentation, a device for decoding coded data representative of the subset of data to be processed and the associated segmentation information and a device for processing the decoded data, when the plurality of data is a multi-view video of the scene acquired by a plurality of cameras;

[0067] [Fig 5]: schematically illustrates a second example of the architecture of a system according to an embodiment of the invention comprising a device for segmenting a plurality of input data, a device for encoding segmentation information and a subset of data to be processed from the segmentation, a device for decoding coded data representative of the subset of data to be processed and the associated segmentation information and a device for processing the decoded data, where the plurality of data includes physiological measurements of a patient acquired by a plurality of sensors;

[0068] [Fig 6]: describes in the form of a flowchart the steps of a process for segmenting a plurality of input data, according to an example of an embodiment of the invention;

[0069] [Fig 7]: describes in the form of a flowchart the steps of a process for encoding data resulting from the segmentation of the plurality of coded data according to an embodiment of the invention;

[0070] [Fig 8]: describes in the form of a flowchart the steps of a process for decoding data resulting from the segmentation of the plurality of data according to an embodiment of the invention;

[0071] [Fig 9]: details a first example of the implementation of the aforementioned segmentation and coding processes when the plurality of data includes a multi-view video;

[0072] [Fig 10]: details a first example of implementation of the aforementioned decoding process when the plurality of input data includes a multi-view video;

[0073] [Fig 11]: details a second example of implementation of the aforementioned segmentation and coding processes when the plurality of input data includes a plurality of sequences of physiological measurements of a patient;

[0074] [Fig 12]: details a second example of implementation of the aforementioned decoding process when the plurality of input data includes a plurality of sequences of physiological measurements of a patient;

[0075] [Fig 13]: describes an example of the material structure of a device for segmenting a plurality of data according to the invention;

[0076] [Fig 14]: describes an example of the hardware structure of a device for encoding a plurality of data according to the invention; and

[0077] [Fig 15]: describes an example of the material structure of a device for decoding data resulting from segmentation and coding according to the invention. Description of the invention

[0078] The principle of the invention is based on the segmentation of a plurality of input data, acquired by sensors arranged around a scene, object, or subject, for the purpose of processing them by a given processing device. Here, segmentation refers to the process of identifying, among the plurality of input data, those to be retained for a given subsequent processing and grouping them into a subset of data to be processed.

[0079] In other words, it involves pruning data from one or more of these sensors when it is not useful or relevant for the processing intended by the processing device, for example, because it is redundant with data acquired by one or more other sensors. According to the invention, the decision to prune or retain the input data is based on two criteria: a criterion for optimizing the quantity of segmented data, that is, data retained in the subset of data to be processed, and a criterion for optimizing the quality of the processing performed on the subset of data to be processed.

[0080] According to one embodiment of the invention, the segmentation is implemented by an artificial intelligence module placed upstream of the processing device and capable of configuring itself by machine learning. The structure of the automatic segmentation module includes a plurality of weights intended to be applied to the plurality of input data, and learning a configuration of the automatic segmentation module consists of the module learning the best possible internal configuration of its weights, allowing both to minimize the amount of data from the subset of input data to be presented as input to the processing device and to maximize the quality of the data processing.

[0081] According to the invention, this learning process is, on the one hand, specific to the plurality of input data and, on the other hand, performed by considering the result of the combined actions of the segmentation module and the processing device located downstream of the segmentation module. This processing device thus receives as input the plurality of data weighted by the configuration weights of the automatic segmentation module, according to at least two criteria. The first performance criterion, relating to the segmentation module, is to minimize the amount of data to be provided to the processing device, which amounts to pruning the plurality of input data as much as possible. The second performance criterion is linked to the processing device and requires maximizing a measure of the quality of the processing performed by the processing device.

[0082] The configuration values ​​of the automatic segmentation module are then used to determine segmentation information for the plurality of data, and a subset of data to be processed is finally obtained by applying the segmentation information to the plurality of input data.

[0083] The invention thus makes it possible to reduce the amount of data to be coded and transmitted to a remote receiving equipment incorporating a processing device similar to those implemented by a transmitting equipment, while preserving the quality of the processing of this data.

[0084] The invention applies to any type of input data and is independent of the artificial intelligence technique implemented by the automatic segmentation module, provided that it can be configured using a plurality of configuration weights associated with the plurality of input data.

[0085] The invention finds a particular application in a free navigation system within a multi-view video, for example embedded in terminal equipment, such as a mobile phone or virtual reality headset. In this case, the processing involves synthesizing an additional view, desired by the user of the terminal equipment, from the segmented data.

[0086] Of course, the invention is not limited to this use case, but can be applied to any other plurality of data acquired by sensors, such as sensors for physiological measurements of a patient.

[0087] In the following, we will focus on describing examples of implementation of the invention for these two particular use cases.

[0088] The term "scene" in the following broad sense refers to any object, subject or plurality of objects or subjects in their environment.

[0089] In relation to Figure 4, we will now present an example of the architecture of a system S according to the invention, comprising a server equipment ES configured to receive input data acquired by a plurality of sensors Cl, C2, ...CN. In this case, it consists of a plurality of cameras and the input data received is a plurality of video sequences taken from different viewpoints of a scene SC, also called multi-view video (MVV).

[0090] In this example, the ES server equipment comprises an E / R transceiver module, a segmentation device 100 according to the invention, a PROCI processing device, SYNT1 for the plurality of input data, and an encoding device 200 for the segmented data according to the invention. Such ES server equipment is configured to transmit the encoded data, for example, as an STR data stream or an F file. D, using its E / R module. It also includes an M ES memory in which it stores, for example, coded data.

[0091] According to the invention, the segmentation device 100 is configured to determine weight values ​​to be applied to the plurality of input data before processing by at least one processing device (PROCI, SYNT1) configured to produce a processing result according to a criterion for maximizing a quality measure of the processing result of the input data, said weight values ​​being determined according to said criterion and another criterion for minimizing a quantity of input data to be processed, to determine segmentation information for said plurality of input data, said segmentation information for said data being valued at a first value or a second value distinct from the first, according to said weights, and to obtain a subset of data to be processed by applying the segmentation information determined to said plurality of input data,the subset of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the first value.

[0092] In this example, the system S includes an artificial intelligence module or segmentation module SEG1 comprising said weights and located upstream of the PROCI processing device, SYNT1. Advantageously, the determination includes learning said weight values, from said plurality of input data, by backpropagation of a gradient of a loss function combining the two criteria.

[0093] Alternatively, the SEG1 module is integrated into device 100.

[0094] Device 100 thus implements the process of segmenting a plurality of data representative of a scene according to the invention which will be detailed below in relation to figure 6.

[0095] According to the invention, the encoding device 200 is configured to obtain segmentation information from the plurality of data and the subset of data to be processed produced by the segmentation device 100, and to encode the subset of data to be processed and the segmentation information obtained. In Figure 4, the Device 100 is independent of device 200, but according to one variant (not shown), it is integrated into device 200.

[0096] Device 200 thus implements the process of encoding data representative of a scene according to the invention which will be detailed below in relation to figure 7.

[0097] In this example, the PROCI processing device, SYNT1, is configured to synthesize a user-selected view from the decoded input views when that view does not correspond to any of the views in the multiview video that were transmitted in the STR, F encoded data stream or file. D It is understood that according to the invention, the PROCI, SYNT1 processing device implemented in the ES server equipment is used only for the purpose of training the SEG1 automatic segmentation module.

[0098] In this example, the system S also includes a terminal equipment UE, for example a smartphone or an augmented reality headset, for example of the head-mounted device or HMD type, worn by a user UT, remote from the server equipment ES and for example connected to it via an RC communications network.

[0099] According to the invention, the UE terminal equipment comprises a transmit / receive E / R module capable of receiving the STR, F encoded data stream or file D via the RC communication network, a 300 device for decoding received coded data, a PROC2, SYNT2 processing device for segmented data, similar to that of the ES equipment in terms of structure and configuration, and an M UE- memory

[0100] In this example, device 300 is configured to decode the received coded data and make it available to the processing device PROC2, SYNT2. According to the invention, the coded data includes the segmentation information of the plurality of input data (here the multiview video VMV) and the subset of data to be processed, resulting from the application of the determined segmentation information to said plurality of input data, produced by device 100 of the ES equipment, as previously described.

[0101] Device 300 thus implements the decoding process according to the invention which will be detailed below in relation to figure 8.

[0102] In the example in Figure 4, the PROC2, SYNT2 processing device of the terminal equipment (UE) is therefore configured to synthesize the view chosen by a user from the other decoded views of the multiview video, when this view does not correspond to any of the views of the multiview video transmitted in the STR, F encoded data stream or file. D It is similar to the PROCI, SYNT1 device of the ES server equipment in that it has the same structure and configuration. In a particular embodiment, the PROCI, PROC2 processing devices also include an artificial intelligence module comprising a weight structure, the values ​​of which are configured using a machine learning technique. According to this embodiment, the prior configuration of the two devices PROCI, PROC2 processing (ES and UE side) is automatically learned from the same input database, for example labeled and under the constraint of a maximization criterion of a quality measure of the processing result.

[0103] The M UE memory is used for example to store the additional view synthesized by the PROC2 processing device, SYNT2.

[0104] In relation to Figure 5, we now present another example of the architecture of a system S' according to the invention, comprising a server device ES' configured to receive input data acquired by a plurality of sensors SI, S2, ...SN, of various types, for example ECG, EEG, CT scanner, MRI, blood composition analysis, X-ray, etc. In this case, it involves a plurality of physiological sensors placed on or near a patient SB, and the input data received consists of a plurality of sequences of physiological measurements acquired at different successive times by the plurality of sensors. According to this example, the server device ES' has a structure similar to that of the server device ES previously described in relation to Figure 4.It comprises a device 100' for segmenting the plurality of physiological measurements according to the invention, a segmentation module SEG1' for the input data, configured by device 100' to segment the input data according to the invention, a processing device PROCI' DIAG1 for the data resulting from the segmentation by module SEG1', a device 200' for encoding the segmented data according to the invention, an E / R transmit / receive module, and an ES- memory. Devices 100' and 200' are similar to the aforementioned devices 100 and 200. Device 100' thus implements the input data segmentation method (here, physiological measurements of patient SB) according to the invention, which will be detailed below with reference to Figure 6. Device 200' thus implements the input data encoding method according to the invention, which will be detailed below with reference to Figure 7.

[0105] In this example, the PROCI', DIAG1 processing device is configured to process the physiological measurements provided by the sensors to perform diagnostic support. It is understood that, according to the invention, on the server equipment side (ES'), this PROCI', DIAG1 processing device is used here only to configure the SEG1' segmentation module of device 100' according to the invention.

[0106] The system S' also includes a terminal device UE', located remotely from the server device ES' and connected to it via the RC communication network. The UE' has a structure similar to that of the terminal device UE previously described in relation to Figure 4. It comprises at least one decoding device 300' similar to the aforementioned decoding device 300 in relation to Figure 4, a processing device PROC2', DIAG2 pre-configured to process the plurality of data from the sensors once decoded by the decoding device 300', an E / R transmit / receive module, and a memory M UE-. The device 300' is similar to the aforementioned device 300 and implements the method for decoding data representative of a scene according to the invention, which will be detailed below in relation to figure 8. Advantageously, it further includes a DISP display device, for example of screen type, which is used by a DOC user to view the physiological measurements received and decoded by 300'.

[0107] An example of the implementation of a method for segmenting a plurality of input data according to an embodiment of the invention is now presented in relation to Figure 6, in the form of a flowchart. Advantageously, this method is implemented by the aforementioned device 100 or 100'.

[0108] We denote here by SQ1(t), SQ2(t), ..., SQN(t), with N a non-zero integer, N time sequences of data acquired by N sensors placed around the object, subject or scene of interest at Nt successive time instants, with Nt a non-zero integer, for example the scene SC or the patient SB of figures 4 and 5. We assume that the N sequences are obtained in 60. In 61, the weight values ​​of a segmentation module SEG1 (resp. SEG1'), to be applied to the input data, are determined. For example, the SEG1 module (resp. SEG1') is an artificial intelligence module capable of learning, from the input data SQl(t), SQ2(t), ..., SQN(t), those that are useful to a PROCI processing device (resp. PROC1') located downstream (the output of the SEG1 module (resp. SEG1') is connected to the input of the PROCI processing device (resp. PROC1')) and previously configured to apply a predetermined processing to the plurality of input data, and those that are not.According to the invention, the segmentation module SEG1 (resp. SEG1') is structured to apply weights to the plurality of input data, the value of which is determined during this learning process. Advantageously, the segmentation module SEG1 (resp. SEG1') comprises as many weights as there are input data. These weights have, for example, floating-point values ​​between 0 and 1, and their application (multiplication) to the value of an input data has the effect of keeping that data "on" or "off" before it is processed by the PROCI (resp. PROC1') processing device.

[0109] For example, in a particular configuration of the SEG1 (resp. SEG1') automatic segmentation module, these weights are distributed across successive layers associated with each of the N input data sequences at each of the Nt time points of each sequence. For example, when the input sequences are image sequences, the SEG1 module comprises N x Nt layers, with one layer per image or view and per time point t, whose dimensions correspond respectively to those of the images in the N sequences.

[0110] According to the invention, learning is based on a first performance criterion of the segmentation module SEG1 (resp. SEG1'), which is an optimization criterion for the quantity of data to be presented to the processing device PROC1 (resp. PROC1'), but also on a second performance criterion, specific to the processing device PROCI (resp. PROC1') itself, which is an optimization criterion for the quality of the processing. This takes into account the combination of the two successive processing steps and the satisfaction of these two performance criteria. ensure that the pruning of input data by the segmentation module SEG1 (resp. SEG1') does not come at the expense of the performance of the processing device PROCI (resp. PROC1').

[0111] For example, the first criterion includes minimizing the amplitude of the cumulative weights of the layers of the segmentation module SEG1, SEG1'. In one variant, it includes minimizing the number of weight values ​​above a given threshold in said layers or maximizing the number of weight values ​​below said threshold. In yet another variant, it includes minimizing the entropy of the input data at the output of the segmentation module SEG1, SEG1', that is, after the weight values ​​of said layers have been applied. In another variant, it includes minimizing the amplitude of the cumulative values ​​of the input data stored in the subset of data to be processed. In yet another variant, it includes minimizing the number of input data points stored in the subset of input data to be processed.According to yet another variant, it includes minimizing the entropy of the input data values ​​stored in the subset of input data to be processed. Of course, the first criterion can also include a combination of several of these different criteria.

[0112] In version 62, SGI segmentation information for the input data is determined from the weight values ​​defined for the SEG1 (or SEG1') module. For an input data item, the corresponding SGI segmentation information is set to a first or second value distinct from the first, depending on the corresponding weight value. For example, the first value is greater than the second value. The weight value is compared to a given threshold. When the weight value is greater than the threshold, the segmentation information is set to the first value (which means "useful," therefore "on"), and to the second value (which means "useless," therefore "off") otherwise.For example, for image input data with dimensions width W x height H, the segmentation information associated with this image includes a segmentation map of the same dimensions W x H. Therefore, pixels are valued at a gray level corresponding to "white" when they correspond to data deemed "useful," and at a gray level corresponding to "black" otherwise. Alternatively, the first value is equal to 1 and the second is zero. The SGI segmentation information is stored in memory M1.

[0113] In 63, a subset of data to be processed, USS, is obtained by applying segmentation information to the plurality of input data SQ1, SQ2, ..., SQN, and retaining only those data associated with a segmentation information valued at the first value. For example, the subset of data to be processed, USS, is stored in memory M1 and can be provided to an encoder.

[0114] Optionally, according to a particular embodiment of the invention, the PROCI processing device (resp. PROC1') also includes an artificial intelligence module having a layered structure comprising processing weights. assumes that it has been previously trained, using a training set and in a way known per se, to process input data of the type of N temporal sequences SQ1(t), SQ2(t), ..., SQN(t), according to the criterion of maximizing a measure of processing quality. We therefore assume that the values ​​of its processing weights have already been determined during this prior training.

[0115] According to a first embodiment of the invention, the processing weights of the PROCI processing device (resp. PROC1') are fixed, that is to say that their values ​​are not modified by determination 61.

[0116] According to a second embodiment of the invention, determination 61 also applies to the PROCI (resp. PROC1') processing device and has the effect of modifying the values ​​of its processing weights. The modified processing weight values ​​MW are obtained in 64. They are provided to an encoder or stored in memory M1.

[0117] Examples of implementation of the invention in specific use cases will be detailed below in relation to figures 9 to 11.

[0118] An example of the implementation of a method for encoding a plurality of input data according to an embodiment of the invention is now presented in relation to Figure 7, in the form of a flowchart. Advantageously, this method is implemented by the aforementioned device 200 or 200'.

[0119] In step 70, the device 200, 200' obtains N time sequences of input data SQ1(t), SQ2(t), ..., SQN(t), where N is a non-zero integer, acquired by N sensors placed around the object, subject, or scene of interest. In step 71, it obtains a segmentation of this data performed by the segmentation method according to the invention just described in relation to Figure 6. It is assumed that the device 200, 200' obtains at least the segmentation information SGI and the subset of data to be processed USS provided by the segmentation method according to the invention. Advantageously, it also obtains modified weight values ​​MW intended for the processing device PROCI, PROC1'.

[0120] It should be noted that the device 100, 100' may or may not be integrated into the encoding device 200, 200'. In the first case, the device 200, 200' directly implements the segmentation process according to the invention. In the second case, it commands the segmentation of the input data to the device 100, 100' and receives the segmented data in response.

[0121] In 72, device 200, 200' encodes the obtained SGI, USS data in a conventional, known manner. In 73, the encoded data is stored in memory M2 in a data file F D or transmitted in an STR data stream to a remote device, for example the terminal device UE, UE' of Figures 4 and 5.

[0122] An example of the implementation of a method for decoding data acquired by a plurality of sensors, according to an embodiment of the invention, is now presented in the form of a flowchart in relation to Figure 8. Advantageously, this method is implemented by the aforementioned device 300 or 300'.

[0123] In 1980, the coded data is obtained, for example, from an M3 memory or They are received via a communication network, such as the RC network shown in Figures 4 and 5.

[0124] According to the invention, the coded data includes at least the SGI segmentation information and the USS subset of data to be processed, the SGI segmentation information enabling the identification of the data to be processed from the USS subset in the N input data sequences expected by the decoder.

[0125] In step 81, the encoded data is decoded in a method known per se. A subset of decoded USSD data to be processed and decoded SGID segmentation information are obtained. Optionally, in a particular embodiment, decoded modified MWD processing weight values ​​intended for the PROC2 (or PROC2') processing device are also obtained.

[0126] In step 82, the decoded SGID segmentation information and the decoded USSD subset of data to be processed are used to construct segmented decoded data sequences SQ1DS(t), SQ2DS(t), ..., SQNDS(t) for presentation to the processing device PROC2, PROC2'. If modified decoded weight values ​​(MWD) have been obtained, they are first applied to a processing device PROC2, PROC2' with a structure and initial configuration similar to the encoder-side processing device PROCI, PROC1'. Once configured, the segmented decoded data sequences SQ1DS(t), SQ2DS(t), ..., SQNDS(t) are provided to the device PROC2, PROC2' for processing.

[0127] We now turn our attention to describing a first example of implementation of the invention, in relation to figures 9 and 10.

[0128] As in figures 1 and 4, we consider a scene SC captured by a set of N cameras Cl, C2, ..., CN. Each camera observes the scene from a given viewpoint and generates at Nt successive times t an image or view of texture Tl, T2, ..., TN corresponding to this viewpoint.

[0129] The processing applied to the input data here corresponds to view synthesis. Specifically, it involves synthesizing an additional view associated with a separate viewpoint PVS, distinct from the N viewpoints of the N cameras. The synthesis device in question is configured to take as input all or part of the N images T1(t), T2(t), ..., TN(t) and their intrinsic (focal length, sensor resolution, number of color components, etc.) and extrinsic (position in space, orientation, etc.) parameters from the N cameras, as well as the coordinates of the scene's additional viewpoint PVS, and to produce as output the additional image or view TS. In this example, it is assumed that the synthesis device SYNT1, SYNT2 comprises an IBRNet neural network. Such a network has a structure of M successive layers L'1, L'2, ..., L'M, with M a non-zero integer, connected together such that the output of an upstream layer is presented as the input of the next downstream layer. Each layer has configuration weights WjL'i, with i an integer between 1 and M and j an integer between 1 and a number of processing weights associated with layer i, whose values ​​are determined during prior training.

[0130] In relation to Figure 9, we first describe the operations implemented on the encoder side, that is to say at the level of the ES server equipment of Figure 4. According to the invention, the automatic segmentation module SEG1 also has the structure of a layered neural network, which are also called relevance layers, because the values ​​of its weights are intended to indicate in the end a level of relevance or usefulness of an input data for its subsequent processing by the synthesis device SYNT1.

[0131] According to this embodiment of the invention, the automatic segmentation module comprises as many LI, L2, ..., LN layers as there are input images Tl(t), T2(t), ..., TN(t) at a given time t. For simplicity, in the following, we will be at this given time t and we will denote by Tl, T2, ..., TN the corresponding sequence of input images.

[0132] Advantageously, each layer Li, with i ranging from 1 to N, has the same dimensions as the corresponding input image or view. For example, if the input images are WxH pixels in size, the layer Li is also WxH in size; that is, it comprises WxH weights, each weight being a unique multiplier of the scalar (grayscale) or vector (RGB or YUV triplet) value associated with the corresponding pixel in the input image. Advantageously, the multiplier is a floating-point number.

[0133] Alternatively, the weights of each layer Li take binary values. If the weight bit has the value 1, the associated pixel is transmitted without change to the input of the processing device SYNT1, otherwise the associated pixel is not transmitted to the processing device SYNT1 or it is transmitted with a predefined value (for example -1) to indicate that it is not useful.

[0134] The neural network of the SEG1 segmentation module is then subjected to learning, so as to assign appropriate values ​​to the weights of these different layers so that it performs the expected task (here, the synthesis of an additional view) satisfactorily.

[0135] According to this embodiment of the invention, it is actually a neural network CRN1 combining the N layers L1, L2, ..., LN of the segmentation module SEG1 and the M layers L'1, L'2, ..., L'M of the processing device PROC1 that is subjected to this training. To this end, two distinct performance criteria, here called loss functions, are used. The first criterion concerns a loss related to the task of the neural network of the synthesis device SYNT1, and measures the quality of the synthesized image. The second criterion concerns a loss related to the segmentation of the input data by the neural network of the module SEG1, and measures the cumulative amplitude of the weight values ​​of its layers L1, ..., LN. A global loss function is defined that takes these two criteria into account. An example is detailed below.

[0136] In practice, the learning process is similar to that described in the document by Wang et al. cited earlier. Briefly, this learning works as follows: for each value j in 1, ..., N, the combined network CRN1 is asked to generate the view Tj from the other views. Since we have the original view Tj associated with the From PVj's perspective, it is possible to evaluate the quality of the synthesis performed by the PROC1 device, for example, by calculating a difference (e.g., based on the mean squared error) between the synthesized view TSj and the original view Tj. This difference is then used to calculate a gradient that is backpropagated to the Lj layer of the SEG1 module in a manner known per se. Note that when the synthesis device is not based on an artificial intelligence method, the calculated gradient is simply propagated to the Lj layer.

[0137] For example, the global loss function to be minimized is the following (where || .|| is the L2 norm), based on the square root of the sum of the squares of the coordinates of a vector): A is a user-defined parameter that determines a choice between minimizing the size of segmented areas (high A) or view synthesis performance (low A). It is a positive real number, which can be greater than or less than 1. PLj denotes a vector concatenation of the WxH weights of the N layers Li.

[0139] According to a first option, the weights of the neural network of the SYNT1 synthesis device are fixed and are not modified by the learning of the SEG1 module. In other words, only the weights of the LI, L2, ..., LN layers of the SEG1 module are optimized in the sense of the loss function.

[0140] Alternatively, the weights of the network layers of the SYNT1 processing device can vary. Advantageously, they are optimized during the training of the SEG1 module, which becomes a training of the combination of the SEG1 module and the SYNT1 device. In this case, the modified values ​​MWL'1, MWL'2, ..., MWL'M of the SYNT1 device's processing weights are obtained and stored for transmission, for example, to the terminal equipment UE shown in Figure 4.

[0141] Once the learning (61) is completed, the SGI segmentation information of the input data T1, T2, ..., TN is determined in 62.

[0142] According to this embodiment of the invention, each layer LI, L2, ..., LN of the module SEG1 is transformed into a segmentation map SGI1, SGI2, ..., SGIN of the associated view T1, T2, ..., TN. It should be noted that in this example, a segmentation map of a given image or view is a map with the same dimensions as that image, and whose elements or pixels are assigned a first value, for example, the value 1, meaning that the associated pixel in the image Tj is useful for the subsequent task of the processing device PROC1, and the value 0 meaning that the associated pixel in the image Tj is not useful for this processing device.

[0143] Advantageously, when the weight values ​​of the LI, L2, ..., LN layers are floating-point real numbers, for example between 0 and 1, the corresponding value of the segmentation map is determined by comparing each weight to a given threshold, for example equal to 0.01. If the weight value is less than this threshold, then the corresponding pixel of the segmentation map is set to 0; otherwise, this pixel is set to 1.

[0144] Alternatively, when the weight values ​​are binary, they are directly copied into the corresponding SGIi segmentation map (the SGIi map is equal to the weight map of the Li layer).

[0145] The views T1, T2, ..., TN are then encoded (72) by the 200 encoding device using the SGI, SG2, ..., SGN segmentation cards. Advantageously, a multi-view video encoding method is used, preferably the MIV standard. Alternatively, other standards such as MV-HEVC or 3D HEVC can be used.

[0146] According to the MIV standard, at a given time t, portions of views T1(t), T2(t), ..., TN(t), called patches, are arranged in a large common image called an atlas. Typically, these portions correspond, for example, to the result of a prior segmentation of the views acquired by the N cameras, designed to retain only the non-redundant information between views. For example, a portion of an object in the scene that would be visible in a given view and that is transmitted in a patch of that view no longer needs to be transmitted in a patch of another view.

[0147] The atlases corresponding to the different moments in the video are then encoded like classic 2D videos, using, for example, the HEVC video coding standard. The HEVC IV standard provides a syntax that allows the decoder to identify the patches within the atlases and rearrange them in the original views. The views decoded and then reconstructed by the decoder are therefore partial.

[0148] According to the invention, the division into patches is carried out, for each view Tj, on the basis of the segmentation map SGIj produced from the values ​​assigned to the weights of the automatic segmentation module at the end of the learning (61).

[0149] In this regard, it should be noted that there are several ways to create patches from the same segmentation map. For example, one can create as many patches as there are groups of connected pixels with the "useful" value in the segmentation map under consideration. Alternatively, one can impose a minimum limit for creating a patch from such a group of connected pixels. For example, if a group of segmented pixels contains fewer than a first threshold (for example, 100 pixels), it will not be considered for creating a new patch, but an attempt will be made to attach it to a neighboring group of segmented pixels until this limit is exceeded. According to another variation, it is also possible to simply omit to consider groups of segmented pixels whose number of pixels is less than a second threshold, which is lower than the first, and for example, equal to 5.

[0150] The patches thus created for a given view Ti(t) are incorporated into an atlas along with other patches possibly originating from other views Tj(t) associated with the same time t. In order to reconstruct the partial views on the decoder side, it is necessary to transmit, and therefore encode, the following information:

[0151] - the view to which the patch is associated, - the coordinates of the patch in the atlas, - an occupancy map of the patch in the atlas.

[0152] It is therefore noted that the MIV standard already allows the consideration of view segmentation maps in the coding scheme of a multiview video so that only the input data of the subset of data to be processed USS are coded and transmitted to a receiver (UE).

[0153] Once the atlases are filled with all the patches identified in all the views, the atlases are coded by a classic encoder, for example conforming to the HEVC standard.

[0154] In one embodiment, the configuration of the neural network of the SYNT1 synthesis device is fixed.

[0155] In another embodiment, the configuration of the neural network of the SYNT1 synthesis device is modified by learning, and the modified values ​​MWL'i, MWL'2, ..., MWL'N of the processing weights of the different layers L'1, L'2, ..., L'M of the neural network of the SYNT1 synthesis device are also encoded (for example, with the MPEG NNR encoding standard, from the English "Neural Net Representation"), specified in Part 17 of the MPEG-7 standard, for transmission to the corresponding SYNT2 synthesis device on the receiver side (UE). It should be noted that here several modified processing weight values ​​MWL'i can be transmitted for layer L'i, and that this set of MWL'i values ​​can take the form of a map or an image whose dimensions are identical to those of the layer L'i in question.

[0156] It is assumed that the encoded data is transmitted to a receiving device, for example the UE terminal equipment in Figure 4, via the RC communication network, for example in the form of an STR encoded data stream or an F encoded data file D .

[0157] They are received by the decoding device 300 which, according to this embodiment, performs a decoding of the coded data relating to the complete or partial views T1, T2, ..., TN in accordance with the MIV standard.

[0158] This decoding process produces the following decoded data: - a series or sequence of atlases, -for each atlas, representative data of the patches it contains and, for each of these patches: - the view to which the patch is associated, - the coordinates of the patch in the atlas, and - a map showing the patch's occupation in the atlas.

[0159] From these decoded data, the segmented decoded views T1DS, T2DS, ..., TNDS are reconstructed (82), either completely or partially (depending on the information contained in the patches).

[0160] In the embodiment in which modified MWL'l, MWL'2, ..., MWL'M weight values ​​are encoded, they are also decoded and then used to update the configuration of the neural network layers of the SYNT2 processing device.

[0161] Next, the N decoded T1DS, T2DS, ..., TNDS views are presented to the neural network corresponding to the PROC2 device.

[0162] It is assumed that a user UT wishes to synthesize an additional view according to the additional viewpoint PVS. In this case, according to the invention, the segmented decoded views T1DS, T2DS, ..., TNDS are presented to the input of the synthesis device SYNT2 (some layers of which L'i may have been updated with the modified weight values ​​MWL'1, MWL'2, ..., MWL'M, according to the embodiment considered). The coordinates of the PVS viewpoint desired by the user UT are also entered. The SYNT2 device finally produces the synthesized additional view TSj'.

[0163] We now turn to describing a second example of implementation of the invention, in relation to figures 11 and 12.

[0164] In relation to Figures 5 and 11, we consider a patient SB surrounded by a plurality of sensors SI, S2, ..., Si, ..., SN of various types (ECG, EEG, CT scan, MRI, chemical analyses, X-ray, ...). Each sensor acquires at least one data point T1(Y), T2(Y), ..., Ti(Y), ..., TN(Y) of the type physiological measurement, image, etc.

[0165] Furthermore, it is assumed that this data is then processed by several neural networks RI, R2, ..., Rk, ..., RK, where K is a non-zero integer less than or equal to N, each specialized in a diagnostic or diagnostic support task. For example, the set of these networks is considered to form the DIAG1 processing system in Figure 5. Each of these networks is configured to take as input a subset of the N input data T1(Y), ..., Ti(y), ..., TN(Y) and to produce as output either diagnostic information Di(Y) (e.g., presence of a tumor, suspicion of epilepsy, etc.) or diagnostic support information (e.g., a segmentation, i.e., a geographical delimitation of an area likely to contain a tumor, etc.). The subset of input data to be presented to each neural network Rk is selected by an algorithm external to the neural network in question and may include some or all of the Ti(Y) data.Examples of Rk neural networks adapted for the diagnosis of particular pathologies are given in the paper by Zhang et al., entitled “Multi-channel deep convolutional neural networks for multi-classifying thyroid disease”, published by the website https: / / arxiv.org / abs / 2203.03627, in March 2022, in the paper by Dehghani et al., entitled “Joint brain tumor segmentation from multi MR sequences through a deep convolutional neural network”, published by the website https: / / arxiv.org / abs / 2203.03338, in March 2022, and in the paper by Salafian et al., entitled “CNN-Aided Factor Graphs with Estimated Mutual Information Features for Seizure Detection”, published by the website https: / / arxiv.org / abs / 2203.05950, in March 2022.

[0166] The processing to be applied to the input data of each of the Rk networks therefore corresponds here to a diagnostic aid processing.

[0167] In this example, we assume that each processing device DIAG11, DIAG12, ..., DIAG1K comprises a neural network in layers RI, R2, ..., RK. Each of these networks Rk has a structure of successive layers, connected to each other such that the output of an upstream layer is presented as the input of the next downstream layer. Each layer has configuration weights whose values ​​are adjusted during prior learning.

[0168] In relation to Figure 11, we first describe the operations implemented on the encoder side, that is to say at the level of the server equipment ES' in Figure 5. In this example, we consider K segmentation modules SEG11', SEGlk', ..., SEG1K', each presenting the structure of a layered neural network, or relevance layers, because the values ​​of its weights are intended to indicate in the end a level of relevance or usefulness of an input data for its subsequent processing by the corresponding neural network RI, R2, ..., RK of the processing device concerned DIAG11, DIAGlk, ..., DIAG1K.

[0169] According to this embodiment of the invention, each segmentation module SEGlk' comprises as many layers LI, L2, ..., LN as there are input data intended for the network Rk. According to a variant not shown, the segmentation layers can be shared so that the segmentation module SEG1 has one layer Li per type of input data Ti, and the corresponding segmented data is addressed to each of the networks Rk that require it. For example, in Figure 11, the network Rk, which receives data T3, T8, and T9, is connected as input to layers L3(y), L8(y), and L9(y).

[0170] Advantageously, each layer Lik(y) of the SEGlk' module has the same size as the associated input data. For example, if the input data is of size H parameters, the layer Li(y) is also of size H weights, each weight being a unique multiplier coefficient associated with a data point (this data point having, for example, a scalar or vector value).

[0171] In one particular embodiment, the weights of each layer Lik(y) of the SEGlk' segmentation module take binary values. If the weight bit has the value 1, the associated data or parameter is transmitted to the Rk network; otherwise, it is not transmitted. Note that, in the latter case, a predefined value (for example, -1) is transmitted to the DIAGlk processing device.

[0172] According to the invention, network learning is performed by combining a number of layers Li(Y) and the layers (not shown) of the networks Rk, using a loss function that combines the loss related to the task of the networks Rk (namely, the quality of the diagnosis) and a loss representative of the cumulative amplitude of the weights of the layers Ci(Y). The network obtained by combining the input layer(s) Li(Y) with the network Rk is called CRk. Before performing this network learning of the segmentation modules SEGlk', the input data sequence T1(y), T2(y)...(TN(y)) is presented to each of the networks Rk so as to obtain original diagnoses Dk(y).

[0173] Next, the learning process, which is self-explanatory, works as follows. For each value k in 1, ..., K, we will try to maximize the diagnostic capability of the CRk network from the data. Since we have the original diagnosis Dk(y), it is possible to calculate a difference between this original diagnosis and the output of the CRk network (for example, the numerical diagnostic error). This difference is used to cal- to calculate a gradient which is back-propagated in the weights of the combined network CRk to be optimized, according to the known formula.

[0174] Destk denotes the value of the diagnosis generated by CRk(Y) and Dk(y) the original diagnosis. The loss function to be minimized (where || .|| is the L2 norm) can be expressed as follows: | | Destk-Dk(y) | | + A. | | PLi | |

[0175] Or A is a user-defined parameter that determines a choice between minimizing the size of segmented areas (high A) or diagnostic performance (low A), and APLi is a vector concatenation of the weights of K layers Li present in the neural network of the SEGlk' segmentation module input to the DIAGlk processing device. These are positive real numbers that can be greater or less than 1.

[0176] In a first embodiment, the weights of the Rk network contained in the CRk neural network are fixed during training, only the weights of the Li layers of the SEGlk' segmentation module network are optimized in the sense of the loss function explained above.

[0177] In a second embodiment, the weights of the layers of the Rk network are also modified by learning. Their modified values ​​(not shown) are stored in memory.

[0178] Each Lik layer of the SEGlk' network resulting from the training is then transformed into an SGIi segmentation map of the associated data. A segmentation map is a map of the same size as the data, containing for each parameter of the data the value 0 or 1, where 0 means that the associated parameter in the data is not used for the subsequent diagnostic task, and the value 1 means the opposite.

[0179] Advantageously, when the weight values ​​of the Li layers are floating-point real numbers, for example between 0 and 1, the determination of the corresponding value of the segmentation map is done by comparing each weight to a given threshold, for example equal to 0.01. If the weight value is less than this threshold, then the corresponding data in the segmentation map is set to 0, otherwise it is set to 1.

[0180] Alternatively, when the weight values ​​are binary, they are directly copied into the corresponding SGIi segmentation map (the SGIi map is equal to the weight map of the Ci layer).

[0181] The segmented parts of the input data Tl(y), T2(y), ..., TN(y) are then encoded; that is, the USS dataset to be processed is obtained by applying the SGIi segmentation maps obtained to each of the N input data sequences. For example, a standard such as MPEG NNR or gzip is used.

[0182] It is also necessary to encode the segmentation map, for example in the form of a series of binary values ​​indicating for each parameter or measurement of the Ti sequence considered whether it is coded or not.

[0183] The coded data is stored in memory and / or transmitted in a data file or data stream.

[0184] In relation to Figure 12, we now consider a receiver of the STR-encoded data stream or the F-encoded data file D , for example the terminal equipment UE' in Figure 5, said receiver performing a decoding of the encoded data received in the STR stream or in the F file D The segmentation maps SGD1, SGD2, and SGDN are decoded, indicating which parameters of the N sequences T1, T2, ..., TN were transmitted and are therefore contained within the subset of data to be processed. Then, the values ​​of these parameters (contained within the encoded subset of data to be processed) are decoded using a decoding method adapted and reciprocal to the one used during encoding (MPEG NNR decoding or gzip decoding). The resulting subset of data to be processed is decoded as USSD.

[0185] The segmented decoded data sequences T1DS, T2DS, ..., TNDS are then reconstructed respectively (in step 82, by CONST1, CONST2, ..., CONSTK) for each of the diagnostic devices DIAG21, DIAG22, ..., DIAG2K, whether complete (all parameters of the data sequence can be reconstructed) or partial (the transmitted parameters only allow reconstruction of part of the sequence) from the decoded USSD data subset to be processed and the decoded segmentation maps SGD1, SGD2, ..., SGDK. Optionally, when the decoded data includes modified weight values ​​for the Rk networks, these modified weight values ​​are injected into the relevant Rk networks.

[0186] We now perform a diagnostic aid using the DIAG2 device shown in Figure 5. In this example, a user (here, a practitioner) wishes to make a diagnosis using an Rk network from Figure 11, for example, the R2 network, and some of the reconstructed data sequences, for example, T3D, T8D, and T9D. The segmented data T1DS, T2DS, ..., TNDS reconstructed during decoding (possibly partial) are entered into the Ci layers, which are themselves connected to the input of the CRk network of the DIAG2 device. The Rk network can then produce its diagnosis using only the segmented data.

[0187] We now present, in relation to Figure 13, an example of the hardware structure of a device 100 for segmenting a plurality of input data acquired by sensors, comprising a module for determining weight values ​​to be applied to the plurality of input data before processing by at least one processing device configured to produce a processing result according to a criterion for maximizing a quality measure of the processing result of the input data, said weight values ​​being determined according to said criterion and another criterion for minimizing a quantity of input data to be processed, a module for determining segmentation information for said plurality of input data, said segmentation information for said data being assigned a first value or a second value distinct from the first, depending on said weights, and a module for obtaining a subset of data to to process by applying the determined segmentation information to said plurality of input data, the subset of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the first value.

[0188] The term "module" can refer to a software component as well as a hardware component or a set of hardware and software components, a software component itself corresponding to one or more computer programs or subprograms or more generally to any element of a program capable of implementing a function or a set of functions.

[0189] More generally, such a device 100 comprises a random access memory 103 (for example, RAM), a processing unit 102 equipped, for example, with a processor, and controlled by a computer program Pgl, representative of the modules of the aforementioned device 100, stored in a read-only memory 101 (for example, ROM or a hard drive). At initialization, the code instructions of the computer program are, for example, loaded into the random access memory 103 before being executed by the processor of the processing unit 102. The random access memory 103 may also contain segmentation information and the subset of useful data.

[0190] Figure 13 illustrates only one particular way, among several possible ways, of implementing device 100, 100' so that it performs the steps of the segmentation process as detailed above, in relation to Figures 6, 9, and 11 in its various embodiments. Indeed, these steps can be performed interchangeably on a reprogrammable computing machine (a PC, a DSP processor, or a microcontroller) executing a program comprising a sequence of instructions, or on a dedicated computing machine (for example, a set of logic gates such as an FPGA or an ASIC, or any other hardware module).

[0191] In the case where device 100 is made with a reprogrammable computing machine, the corresponding program (i.e. the sequence of instructions) may be stored in a removable storage medium (such as, for example, an SD card, a USB key, a CD-ROM or a DVD-ROM) or not, this storage medium being readable partially or totally by a computer or a processor.

[0192] Also presented, in relation to Figure 14, is an example of the material structure of a coding device 200, 200' for a plurality of input data according to the invention, comprising at least one module for obtaining segmentation information for the plurality of input data and a subset of data to be processed by applying said segmentation information to the plurality of input data, said segmentation comprising the determination of weight values ​​to be applied to the plurality of input data before processing by at least one processing device previously configured to produce a processing result according to a maximization criterion for a processing quality measure, said weight values ​​being determined according to said criterion and another minimi- sation of a quantity of input data to be processed, a said segmentation information of a said data being valued at a first or a second value distinct from the first, depending on said weight values, the subset of data to be processed comprising the data of the plurality of input data associated with a segmentation information equal to the first value, and a coding module for the segmentation information and the subset of data to be processed.

[0193] The term "module" can refer to a software component as well as a hardware component or a set of hardware and software components, a software component itself corresponding to one or more computer programs or subprograms or more generally to any element of a program capable of implementing a function or a set of functions.

[0194] More generally, such a device 200, 200' comprises a random access memory 203 (for example, RAM), a processing unit 202 equipped, for example, with a processor, and controlled by a computer program Pg2, representing the segmentation and coding modules, stored in a read-only memory 201 (for example, ROM or a hard drive). At initialization, the code instructions of the computer program are, for example, loaded into the random access memory 203 before being executed by the processor of the processing unit 202. The random access memory 203 can also contain the coded information.

[0195] Figure 14 illustrates only one particular way, among several possible ways, of implementing device 200, 200' so that it carries out the steps of the coding process as detailed above, in relation to Figures 7, 9, and 11 in its various embodiments. Indeed, these steps can be carried out interchangeably on a reprogrammable computing machine (a PC, a DSP processor, or a microcontroller) executing a program comprising a sequence of instructions, or on a dedicated computing machine (for example, a set of logic gates such as an FPGA or an ASIC, or any other hardware module).

[0196] In the case where device 200, 200' is made with a reprogrammable computing machine, the corresponding program (i.e. the sequence of instructions) may be stored in a removable storage medium (such as for example an SD card, a USB key, a CD-ROM or a DVD-ROM) or not, this storage medium being readable partially or totally by a computer or a processor.

[0197] Finally, in relation to Figure 15, we present an example of the hardware structure of a device 300, 300' for decoding coded data, comprising a coded data decoding module including segmentation information for a plurality of data acquired by sensors, called input data, and a subset of data to be processed, said segmentation information for said input data being valued at a first value or a second value distinct from the first, said subset of data to be processed having been obtained by applying said segmentation information to the plurality of input data, the subset of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the first value, a construction module for constructing a plurality of segmented input data decoded from the subset of decoded data to be processed and the decoded segmentation information, and a supply module for providing the plurality of segmented input data decoded to a processing device configured to produce a processing result based on a criterion for maximizing a processing quality measure.

[0198] The term "module" can refer to a software component as well as a hardware component or a set of hardware and software components, a software component itself corresponding to one or more computer programs or subprograms or more generally to any element of a program capable of implementing a function or a set of functions.

[0199] More generally, such a device 300, 300' comprises a random access memory 303 (for example, RAM), a processing unit 302 equipped, for example, with a processor, and controlled by a computer program Pg3, representative of the aforementioned modules, stored in a read-only memory 301 (for example, ROM or a hard drive). At initialization, the code instructions of the computer program are, for example, loaded into the random access memory 303 before being executed by the processor of the processing unit 302.

[0200] Figure 15 illustrates only one particular way, among several possible ways, of implementing the device 300, 300' so that it performs the decoding process steps as detailed above, in relation to Figures 8, 10, and 12 in its various embodiments. Indeed, these steps can be performed interchangeably on a reprogrammable computing machine (a PC, a DSP processor, or a microcontroller) executing a program comprising a sequence of instructions, or on a dedicated computing machine (for example, a set of logic gates such as an FPGA or an ASIC, or any other hardware module).

[0201] In the case where device 300, 300' is made with a reprogrammable computing machine, the corresponding program (i.e. the sequence of instructions) may be stored in a removable storage medium (such as for example an SD card, a USB key, a CD-ROM or a DVD-ROM) or not, this storage medium being readable partially or totally by a computer or a processor.

[0202] The invention described above, in its various embodiments, offers numerous advantages. It enables the selection, from among a plurality of data representing a scene, object, or subject, before their transmission over a communication network, of those data points that will be truly useful for processing this plurality of data by a processing device in a receiving device. To this end, the invention employs machine learning techniques, which it cleverly implements to specifically train an automatic segmentation module within a transmitting device to segment the plurality of input data in such a way as to minimize the amount of data to be transmitted while maximizing the quality of the processing.

Claims

Demands

1. A method for segmenting a plurality of data acquired by sensors, referred to as input data, said method comprising: - the determination (61) of weight values ​​to be applied to the plurality of input data before processing by at least one processing device (PROCI, SYNT1, DIAG1) configured to produce a processing result according to an optimization criterion of the quality of the processing result of the input data, said weight values ​​being determined according to said criterion and another optimization criterion of a quantity of input data to be processed, - the determination (62) of segmentation information (SGI) of said plurality of input data, each segmentation information of each data item being valued at a first value or a second value distinct from the first, depending on said weights, and - obtaining (63) a subset of data to be processed by applying the segmentation information determined to said plurality of input data, the subset of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the first value.

2. Segmentation method according to the preceding claim, characterized in that the determination (61) includes learning said weight values ​​from said plurality of input data, said learning being carried out by backpropagation of a gradient of a loss function combining the two criteria.

3. A segmentation method according to any one of claims 1 and 2, characterized in that, said processing device (PROCI, SYNT1, DIAG1) comprising weights, said processing weights, the values ​​of said processing weights having been previously determined according to the criterion of optimizing the quality of the result of processing the input data, the method further comprises the determination (64) of modified values ​​of said processing weights.

4. A method for segmenting a plurality of data according to any one of claims 2 and 3, characterized in that said plurality of input data comprises a plurality of views acquired by a plurality of cameras, said view comprising pixels, said weights being included in a plurality of layers, said layer being associated with said view and comprising said weight per pixel, and in that the segmentation information comprises a plurality of segmentation maps, said map being associated with said view.

5. A method for segmenting a plurality of data according to any one of claims 2 and 3, characterized in that said plurality of input data comprises a plurality of sequences of measurement data acquired by a plurality of sensors, said weights being contained in a plurality of layers, each layer being associated with each measurement data sequence and comprising each weight per measurement data, and in that the segmentation information comprises a plurality of segmentation sequences, each segmentation sequence being associated with each measurement data sequence.

6. A method for encoding a plurality of data acquired by sensors, called input data, characterized in that it comprises: - obtaining (71) segmentation information (SGI) of the plurality of input data and a subset of data (USS) to be processed by applying said segmentation information to the plurality of input data, said segmentation including the determination of weight values ​​to be applied to the plurality of input data before processing by at least one processing device (PROCI, SYNT1, DIAG1) previously configured to produce a processing result according to a criterion for optimizing the quality of the processing, said weight values ​​being determined according to said criterion and a criterion for optimizing the quantity of input data to be processed, each segmentation information of each data being assigned a first or second value distinct from the first, according to said weight values,the subset (USS) of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the first value, and, - the coding (72) of the segmentation information and the subset of data to be processed.

7. A coding method according to the preceding claim, characterized in that the process further comprises: - obtaining (71) modified weight values ​​(MW), said weights being intended to be applied to the input data by the processing device (PROCI, SYNT1, DIAG1), the values ​​of said weights previously determined according to said criterion of optimization of the quality of the result of processing the input data and for the processing of the plurality of input data, having been modified according to said criteria, for the processing of the subset of data to be processed, - the coding (72) of said modified weight values.

8. A method for decoding coded data, characterized in that it comprises: - the decoding (81) of coded data, said coded data comprising segmentation information (SGI) of a plurality of data acquired by sensors, called input data, a subset (USS) of data to be processed by a processing device configured to apply weights to the plurality of decoded input data and to produce a processing result according to an optimization criterion of a processing quality, said segmentation information of said input data being valued at a first value or a second value distinct from the first, said subset of data to be processed having been obtained by applying said segmentation information to the plurality of input data, the subset of data to be processed comprising the data of the plurality of input data associated with said segmentation information equal to the first value, said coded data further comprising modified values ​​of said weights, said modified values ​​having been determined for the processing of the plurality of segmented input data,based on the optimization criterion of processing quality and the optimization criterion of the quantity of data in the subset of data to be processed, - the construction (82) of a plurality of segmented input data (SQ1DS(t), SQ2DS(t), ..., SQNDS(t)) decoded from the decoded subset of data to be processed and the decoded segmentation information, and - the provision (83) of the plurality of segmented input data decoded and of the modified values ​​of said weights to the processing device.

9. A signal carrying coded data, said coded data comprising segmentation information for a plurality of data acquired by sensors, said input data, a subset of data to be processed obtained by applying said segmentation information to said plurality of input data, said segmentation information for said input data being valued at a first value or a second value distinct from the first, the subset of data to be processed comprising the data from the plurality of input data associated with segmentation information equal to the first value, said subset of data to be processed being intended to be decoded and then used to reconstruct a plurality of segmented input data decoded from the decoded segmentation information, for the purpose of processing said plurality of segmented input data by a processing device,configured to produce a processing result based on a processing quality optimization criterion, by applying weights to the plurality of said input data, said coded data further including modified values ​​of said weights, said modified values ​​having been determined for the, processing of the plurality of segmented input data, according to the optimization criterion of a quality of processing and an optimization criterion of a quantity of data of the subset of data to be processed, and intended to be used by said processing device to update said weights before the processing of the plurality of segmented decoded reconstructed input data.

10. Device (100) for segmenting a plurality of data acquired by sensors, referred to as input data, said device being configured to implement: - the determination of weight values ​​to be applied to the plurality of input data before processing by at least one processing device (PROCI, SYNT1, DIAG1) configured to produce a processing result according to an optimization criterion of the quality of the processing result of the input data, said weight values ​​being determined according to said criterion and an optimization criterion of the quantity of input data to be processed, - the determination of segmentation information for said plurality of input data, each segmentation piece of said data being assigned a first value or a second value distinct from the first, depending on said weights, and - obtaining a subset of data to be processed by applying the segmentation information determined to said plurality of input data, the subset of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the first value.

11. Device (200) for encoding a plurality of data acquired by sensors, called input data, characterized in that it is configured to implement: - obtaining segmentation information (SGI) for the plurality of input data and a subset of data to be processed by applying said segmentation information to the plurality of input data, said segmentation including the determination of weight values ​​to be applied to the plurality of input data before processing by at least one processing device (PROCI, SYNT1, DIAG1) previously configured to produce a processing result according to a criterion for optimizing the quality of the processing, said weight values ​​being determined according to said criterion and another criterion for optimizing the quantity of input data to be processed, each segmentation information for each data being assigned a first or second value distinct from the first, according to said weight values,the subset of data to be processed comprising the data from the plurality of input data associated with a segmentation information equal to the, first value, and - the coding of segmentation information and the subset of data to be processed.

12. A device (300) for decoding encoded data, characterized in that it is configured to implement: - the decoding of coded data, said coded data comprising segmentation information (SGI) of a plurality of data acquired by sensors, said input data, a subset (USS) of data to be processed by a processing device configured to apply weights to the plurality of decoded input data and to produce a processing result according to an optimization criterion of a processing quality, said segmentation information of said input data being valued at a first value or a second value distinct from the first, said subset of data to be processed having been obtained by applying said segmentation information to the plurality of input data, the subset of data to be processed comprising the data of the plurality of input data associated with said segmentation information equal to the first value,said coded data further including modified values ​​of said weights, said modified values ​​having been determined for the processing of the plurality of segmented input data, according to the criterion of optimizing the quality of the processing and a criterion of optimizing the quantity of data in the subset of data to be processed, - the construction of a plurality of segmented input data (SQ1DS(t), SQ2DS(t), ..., SQNDS(t)) decoded from the subset of decoded data to be processed and the decoded segmentation information, and - the provision of the plurality of segmented input data decoded and the modified values ​​of said weights to the processing device.

13. System (S), characterized in that it comprises: - a segmentation device (100) for a plurality of data, according to claim 10, configured to produce a plurality of segmented data; - a device for encoding (200) the plurality of segmented data according to claim 11; a device for decoding (300) encoded data according to claim 12; and - a processing device (PROC2, SYNT2, DIAG2) configured to process a plurality of segmented decoded data reconstructed by the decoding device (300).

14. Product computer program comprising program code instructions for implementing a method according to any one of claims 1 to 8, when executed by a processor.

Images