Oral cavity piece brush head size selection

Abstract

The invention relates to a computer-implemented method of selecting the size of a mouthpiece brush head (10) for an individual. The computer-implemented method comprises: receiving (500) data associated with an individual; generating (502) potential parameter vectors of the individual jaw bones (310) using a plurality of machine learning models based on the received data; measuring (504) a size of the jaw bone based on the potential parameter vector of the jaw bone; and selecting (506) a size of the mouthpiece brushhead based on the measured size of the jaw bone.

Description

Technical Field

[0001] The subject matter of this disclosure relates to computer-implemented methods for selecting the size of an oral brush head for an individual, computer-implemented methods for training a machine learning model to generate a latent parameter vector of an individual's jawbone based on a face image, and computer-implemented methods for training a machine learning model to generate a latent parameter vector of an individual's head based on a face image, and also to transient or non-transitory computer-readable media. Background Technology

[0002] Successful oral care (OHC) products (e.g., single-piece toothbrushes) are manufactured in a way that ensures a perfect fit to an individual's jawbone shape. However, jawbone shape and size vary considerably among different populations. For example, the length of an adult's jawbone can vary by approximately 20 millimeters. When the concept of "one size fits all" doesn't work, multiple sizes, such as "small," "medium," and "large," can be envisioned. This presents a challenge for end-users in deciding whether "one size fits all" is suitable for them, or which "small," "medium," or "large" size to choose. Furthermore, for hygiene reasons, end-users find it difficult to try on different mouthpieces in a store (similar to trying on clothes). This problem is amplified by the high price of (technically complex) single-piece toothbrushes.

[0003] The purpose of this disclosure is to improve existing technology. Summary of the Invention

[0004] According to a first aspect of the invention, a computer-implemented method is provided for selecting the size of an oral component brush head for an individual. The computer-implemented method includes: receiving data associated with the individual; generating a latent parameter vector of the individual's jawbone using multiple machine learning models based on the received data; measuring the size of the jawbone based on the latent parameter vector; and selecting the size of the oral component brush head based on the measured jawbone size. Using multiple machine learning models in this manner enables easy selection of the oral component without requiring intraoral measurement of the jawbone.

[0005] In one embodiment, the data includes an image of an individual's face, and the generation of a latent parameter vector for the jawbone using multiple machine learning models includes: using a first machine learning model among the multiple machine learning models, using the image of the individual's face as input, to generate a latent parameter vector for the individual's head; and using a second machine learning model among the multiple machine learning models, to convert the latent parameter vector for the head into a latent parameter vector for the jawbone.

[0006] In one embodiment, the data includes an image of an individual's face, and generating a latent parameter vector of the jawbone using multiple machine learning models includes: using a third machine learning model among the multiple machine learning models, taking the image of the individual's face as input, to generate a latent parameter vector of the individual's jawbone.

[0007] In one embodiment, the data includes metadata, and wherein generating a latent parameter vector of an individual's jawbone based on a face in a received image using one or more machine learning models includes: generating the latent parameter vector of the jawbone by feeding the metadata into a third machine learning model using a fourth machine learning model among a plurality of machine learning models.

[0008] In one embodiment, the data includes metadata, and generating a latent parameter vector of an individual's jawbone based on a face in a received image using one or more machine learning models includes: generating a latent parameter vector of the individual's head using a fifth machine learning model among one or more machine learning models, using the metadata as input; and converting the latent parameter vector of the head into a latent parameter vector of the jawbone using a second machine learning model among one or more machine learning models.

[0009] In one embodiment, generating a latent parameter vector of the jawbone using multiple machine learning models includes fusing two or more latent parameter vectors of the jawbone using an aggregator machine learning model of the multiple machine learning models; each vector corresponds to a specific output from one of the multiple machine learning models.

[0010] In one embodiment, measuring the size of the jawbone based on its latent parameter vector includes: inputting the latent parameter vector of the jawbone into a sixth machine learning model to directly obtain the size of the jawbone.

[0011] In one embodiment, measuring the size of the jawbone based on its latent parameter vector includes: inputting the latent parameter vector of the jawbone into a seventh machine learning model to obtain a three-dimensional model of the jawbone; and measuring the size of the jawbone based on the obtained three-dimensional model of the jawbone.

[0012] In one embodiment, selecting the size of the oral component brush head based on the measured jawbone dimensions includes: inputting the measured dimensions into an eighth machine learning model to obtain a selection of the oral component size.

[0013] In one embodiment, selecting the size of the oral component brush head based on the measured jawbone dimensions includes inputting the measured dimensions into a rule-based mathematical model to obtain a selection of the oral component size.

[0014] In one embodiment, the dimension is at least one of the following: the length of the jawbone; the angle between the molars, incisors and canines; and the thickness of the molars.

[0015] According to one aspect of the invention, a computer-implemented method is provided for training a machine learning model (MLM) to generate latent parameter vectors of an individual's jawbone based on facial images. The computer-implemented method includes: receiving multiple images and multiple jawbone scans; registering a three-dimensional jawbone mesh with each scan in the multiple jawbone scans; deconstructing the registered mesh using principal component analysis to obtain latent parameter vectors of the jawbone; inputting the multiple facial images into the machine learning model to generate corresponding latent parameter vectors of the jawbone; calculating the error between the generated latent parameter vectors and the obtained latent parameter vectors; and updating the parameters of the machine learning model to minimize the error. For example, the latent parameter vectors of the jawbone can be used as input to a machine learning model trained to derive one or more measurements of the jawbone.

[0016] According to one aspect of the present invention, a computer-implemented method is provided for training a machine learning model (MLM) to generate latent parameter vectors of an individual's head based on faces from images. The computer-implemented method includes: receiving multiple images and multiple facial scans; registering a three-dimensional head mesh with each scan in the multiple facial scans; deconstructing the registered mesh using principal component analysis to obtain latent parameter vectors of the head; inputting the multiple facial images into the machine learning model to generate corresponding latent parameter vectors of the head; calculating the error between the generated latent parameter vectors and the obtained latent parameter vectors; and updating the parameters of the machine learning model to minimize the error.

[0017] According to one aspect of the invention, a transient or non-transitory computer-readable medium is provided, on which instructions are stored, which, when executed by one or more processors, cause one or more processors to perform a computer-implemented method of any of the foregoing aspects or embodiments.

[0018] These and other aspects of the invention will be apparent from and illustrated by reference to the embodiments described below. Attached Figure Description

[0019] The embodiments of the invention can be best understood with reference to the accompanying drawings, in which:

[0020] Figure 1 The oral cavity brush head is shown;

[0021] Figure 2 Mobile and remote devices for implementing the computer-implemented methods described herein are shown;

[0022] Figure 3 A block diagram of a computer-implemented method for selecting the size of an oral component brush head for an individual, according to one or more embodiments, is shown;

[0023] Figure 4 A block diagram of a computer implementation method for training multiple machine learning models according to one or more embodiments is shown;

[0024] Figure 5 A flowchart is shown, summarizing a computer-implemented method for an individual to select the size of a dental brush head;

[0025] Figure 6 A flowchart summarizing a computer-implemented method for training a machine learning model (MLM) to generate latent parameter vectors of an individual's jawbone based on facial images is shown; and

[0026] Figure 7 A flowchart is shown summarizing a computer-implemented method for training a machine learning model (MLM) to generate a vector of latent parameters for an individual's head based on facial features from an image. Detailed Implementation

[0027] At least some of the exemplary embodiments described herein can be constructed, partially or entirely, using dedicated hardware. Terms such as “component,” “module,” or “unit” as used herein may include, but are not limited to, hardware devices such as circuit devices in the form of discrete or integrated components, field-programmable gate arrays (FPGAs), or application-specific integrated circuits (ASICs) that perform a particular task or provide associated functionality. In some embodiments, the described elements may be configured to reside on a tangible, persistent, addressable storage medium and may be configured to execute on one or more processors. In some embodiments, these functional elements may include, by way of example, components such as software components, object-oriented software components, class components and task components, processes, functions, attributes, procedures, subroutines, program code segments, drivers, firmware, microcode, circuit devices, data, databases, data structures, tables, arrays, and variables. Although exemplary embodiments have been described with reference to the components, modules, and units discussed herein, such functional elements may be combined into fewer elements or separated into additional elements. Various combinations of optional features have been described herein, and it should be understood that the described features can be combined in any suitable combination. In particular, features of any example embodiment may be suitably combined with features of any other embodiment, unless such combinations are mutually exclusive. Throughout this specification, the terms "comprising" or "comprises" mean to include the specified components (one or more), but do not exclude the presence of other components.

[0028] refer to Figure 1 The oral component toothbrush 100 (also referred to herein as the "oral component brush head" for the sake of brevity) is shaped to conform to the individual's jawbone. As used herein, the term jawbone refers to the dental arch and gums of the teeth, in which the individual possesses both a maxilla and a mandible.

[0029] refer to Figure 2 The computer-implemented method described herein can be embodied using user equipment 200 and remote equipment 202. The user equipment can be a mobile device. The mobile device includes a user interface 204, which includes a display and user input functionality. The mobile device also includes an imaging device 206. Figure 2 In the mobile device shown, the imaging device is a 2D camera. In other user devices, the imaging device could be a 3D scanning device used to capture a 3D scan of an individual.

[0030] Remote device 202 is configured to receive data from mobile device 200. The remote device includes storage 208 and one or more processors 210. Remote device 202 can be used to execute the computer-implemented methods described herein and transmit data back to mobile device 200, or the computer-implemented methods can be executed directly on mobile device 200. More specifically, the computer-implemented methods described herein can be stored as instructions as electronic data on storage 208 and executed by one or more processors 210, causing one or more processors 210 to perform the computer-implemented methods described herein. In this way, the storage can be a non-transitory computer-readable medium. Alternatively, a transient computer-readable medium can include instructions and be downloaded to storage 208.

[0031] refer to Figure 3 This paper illustrates a computer-implemented method for selecting the size of an oral brush head for an individual, the method including a first step of receiving data associated with the individual.

[0032] These data include images of individuals' faces. These images include a smiling image 300 and a neutral expression image 302. The smiling image 300 shows an individual displaying a happy smile and shows the individual's teeth. The neutral expression is the user's expression when their teeth are not visible.

[0033] The data also includes metadata 304. Metadata 304 includes parameters about the individual, such as weight, height, age, and gender. Metadata 304 may also include other information, such as dental records.

[0034] This method may include using multiple machine learning models (MLMs) to generate a vector of latent parameters for an individual's jawbone based on the received data. For example, the machine learning model could be a neural network.

[0035] Various methods exist for obtaining the latent parameter vector of an individual's jawbone, as described below.

[0036] When the data is either image 300 or 302, a first machine learning model 306 of the multiple machine learning models generates a latent parameter vector of the individual's head from an image of the individual's face used as input. Then, a second machine learning model 308 of the multiple machine learning models is used to transform the latent parameter vector of the head into a latent parameter vector of the jawbone 310.

[0037] In other embodiments, the latent parameter vector of the jawbone can be obtained directly without transformation using a second machine learning model. This can be achieved by using a third machine learning model 312, one of multiple machine learning models, to generate the latent parameter vector of the individual's jawbone using the individual's face as input. In other words, either image 300 or 302 can be input into the third machine learning model 312 to directly obtain the latent parameter vector of the jawbone.

[0038] When the data is metadata or includes metadata, a fourth machine learning model 314 can be used among multiple machine learning models. The potential parameter vector of the jawbone can be obtained by inputting the metadata into the fourth machine learning model 314.

[0039] Similarly, the latent parameter vector of the jawbone can be obtained by inputting metadata into a fifth machine learning model 316 among multiple machine learning models. The fifth machine learning model 316 can be trained to generate a latent parameter vector of an individual's head using metadata as input. The latent vector of the head is then used as input to a second machine learning model 308 to convert it into a latent parameter vector of the jawbone.

[0040] Next, an aggregator machine learning model of multiple machine learning models is used to fuse two or more latent parameter vectors of the jawbone, where each vector corresponds to the output of one of the machine learning models. More specifically, the latent parameter vectors of the jawbone obtained based on the smiling image 300, the neutral expression image 302, and the metadata 304 are fused using an aggregator machine learning model 305. Numerically, the aggregator machine learning model can be a ninth machine learning model.

[0041] The output from the aggregator machine learning model will be a vector of latent parameters of an individual jawbone.

[0042] Next, the computer-implemented method involves measuring the size of the jawbone based on a latent parameter vector. The jawbone size can be, for example, the length of the dental arch, the height of the teeth, and the thickness of one or more teeth. More specifically, the size can be at least one of the following: jawbone length (e.g., the length of the dental arch), the angle between the molars, incisors, and canines, and the thickness of the molars. These dimensions are typically required at a minimum to determine the size of the oral cavity brush head. This can include inputting the latent parameter vector of the jawbone into a sixth machine learning model 318 to directly obtain the jawbone size.

[0043] In other embodiments, this can be accomplished by first inputting the latent parameter vector of the jawbone into a seventh machine learning model 319 to obtain a three-dimensional model of the jawbone, and then measuring the size of the jawbone based on the obtained three-dimensional model. The measurement can be performed using a rule-based model 321. The rule-based model can be a first rule-based model.

[0044] As a result of these methods, the measured size of the jawbone was obtained as 320.

[0045] Next, the computer-implemented method involves selecting the size of the oral cavity brush head based on the measured dimensions of the jawbone. This can be achieved in one of several different ways.

[0046] In one embodiment, the measured jawbone dimensions 320 are fed into an eighth machine learning model 324 to select the size 322 of the oral cavity component. In other embodiments, the measured dimensions may be fed into a rule-based mathematical model to select the size of the oral cavity component. The rule-based mathematical model may be a second rule-based model 326.

[0047] The training of various machine learning models is described below.

[0048] refer to Figure 4 There exists a method to obtain the potential parameter vectors of a 3D head and a 3D jawbone for training a first machine learning model, a second machine learning model, a third machine learning model, a fourth machine learning model, and a fifth machine learning model.

[0049] First, a method for obtaining the potential parameter vectors of the jawbone and head is described.

[0050] This method requires 3D scans of the individual's head and jawbone. The scans are also captured along with images of the individual for training first, third, fourth, and fifth machine learning models.

[0051] Number of scans for different individuals It needs to be large enough, for example, greater than 100 scans. The actual number of scans can exceed 1000. For each scan, we register a universal 3D template. , ,in Using the first A template for registration in each scan. A general 3D template may include a set of 2D landmarks or a dense mesh of 3D landmarks. The general mesh is deformed to fit the 3D scan surface for each scan, so that the general mesh is registered with that scan. Then, the registration mesh for each scan is compared with the average mesh. Perform rigid alignment. Apply principal component analysis (PCA) to obtain a linear decomposition of the alignment model.

[0052] Linear decomposition yields two things. First, there exists a vector of PCA coefficients, which encodes the latent parameters of a 3D scan of the head. This is the latent parameter vector of the head. Second, there exists a PCA basis. The PCA basis comprises the main 3D shape components. The PCA basis can be used to extract a 3D head model from the latent parameter vector of the head. It can also be used to encode new head scans.

[0053] This method can also be used for 3D scanning of the jawbone to obtain the potential parameter vector of the jawbone.

[0054] Figure 4 The method described above is for training a second machine learning model 308. The latent parameter vector 402 of the jawbone and the latent parameter vector 404 of the head are derived from the method described above. The jawbone encoder 406 and jawbone decoder 408 can be neural networks. Similarly, the head encoder 410 and head decoder 412 can be neural networks. The registered 3D model of the jawbone 414 obtained above can be input into the jawbone encoder 406, and the jawbone decoder 408 can generate a corresponding registered 3D jawbone model 415 from the obtained latent parameter vector 402 of the jawbone. The jawbone encoder and jawbone decoder can be trained to reduce the jawbone loss function 416, which is the loss between the registered 3D jawbone models.

[0055] It should be noted that the encoder-decoder neural network generalizes the previously described PCA decomposition, where the latent parameters in the "bottleneck" generalize the PCA coefficients. Therefore, in the case of PCA, the encoder or decoder becomes a linear matrix multiplication.

[0056] Similarly, the registered 3D head model 418 can be input into the head encoder 410, and the head decoder 412 can generate a corresponding registered 3D head model 420 from the resulting latent parameter vector 404 of the head. The head encoder 410 and head decoder 412 can be trained to reduce the head loss function 422, which is the loss between the registered 3D head models.

[0057] The second machine learning model 308 is trained to use conventional machine learning techniques to transform the latent parameter vector of the jawbone into the latent parameter vector of the head, and vice versa.

[0058] Because an individual's image is paired with the individual's scan and metadata, it is possible to train a first, third, fourth, and fifth machine learning model.

[0059] For example, a first machine learning model is trained to generate latent parameter vectors for an individual's head based on faces from images. This method includes obtaining latent parameter vectors for the head from head scans as described above. This can be summarized as receiving multiple images and multiple facial scans; registering a 3D head mesh with each scan in the multiple facial scans; and deconstructing the registered mesh using principal component analysis to obtain the latent parameter vectors for the head. Next, multiple faces are input into the first machine learning model to generate corresponding latent parameter vectors for the head. Then, the error between the generated latent parameter vectors and the obtained latent parameter vectors is calculated. Finally, the parameters of the first machine learning model are updated to minimize the error.

[0060] Similarly, a third machine learning model is trained to generate latent parameter vectors for an individual's jawbone based on facial images. This method involves obtaining the latent parameter vectors of the jawbone from jawbone scans as described above. This can be summarized as receiving multiple images and multiple facial scans (e.g., jawbone scans); registering a 3D head mesh with each scan in the multiple facial scans; and deconstructing the registered mesh using principal component analysis to obtain the latent parameter vectors of the jawbone. Next, multiple faces are input into a first machine learning model to generate corresponding latent parameter vectors for the jawbone. Then, the error between the generated latent parameter vectors and the obtained latent parameter vectors is calculated. Finally, the parameters of the third machine learning model are updated to minimize the error.

[0061] Similarly, the fourth and fifth machine learning models can be trained in the same way, but instead of receiving facial images, they receive metadata and use those as input.

[0062] The other machine learning models described in this paper can be trained using conventional machine learning techniques with the aforementioned related inputs and outputs.

[0063] refer to Figure 5 A computer-implemented method for selecting the size of an oral component brush head for an individual can be summarized as including: receiving 500 data associated with the individual; generating 502 latent parameter vectors of the individual's jawbone using multiple machine learning models based on the received data; measuring 504 the size of the jawbone based on the latent parameter vectors of the jawbone; and selecting 506 the size of the oral component brush head based on the measured size of the jawbone.

[0064] refer to Figure 6 A computer-implemented method for training an MLM to generate latent parameter vectors of an individual's jawbone based on facial images can be summarized as follows: receiving over 600 images and multiple facial scans; registering a 3D jawbone mesh with each scan in the multiple facial scans 602; deconstructing the registered mesh using principal component analysis 604 to obtain latent parameter vectors of the jawbone; inputting multiple facial images 606 into the machine learning model to generate corresponding latent parameter vectors of the jawbone; calculating 608 the error between the generated latent parameter vector and the obtained latent parameter vector; and updating 610 the parameters of the machine learning model to minimize the error.

[0065] refer to Figure 7 A computer-implemented method for training an MLM to generate latent parameter vectors of an individual's head based on facial images can be summarized as follows: receiving over 700 images and multiple facial scans; registering a 3D head mesh with each scan in the multiple facial scans 702; deconstructing the registered mesh using principal component analysis 704 to obtain the latent parameter vectors of the head; inputting multiple facial images 706 into the machine learning model to generate corresponding latent parameter vectors of the head; calculating 708 the error between the generated latent parameter vectors and the obtained latent parameter vectors; and updating 710 the parameters of the machine learning model to minimize the error.

[0066] Although the invention has been described in detail with reference to the accompanying drawings and the foregoing description, such descriptions should be regarded as illustrative or exemplary, and not restrictive; the invention is not limited to the disclosed embodiments.

[0067] Those skilled in the art, in practicing the claimed invention, can understand and implement other variations of the disclosed embodiments by studying the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite articles "a" or "an" do not exclude multiple. A single processor or other unit can perform the functions of several items listed in the claims. The fact that certain measures are listed in mutually different dependent claims does not indicate that a combination of these measures cannot be used advantageously. No reference numerals in the claims should be construed as limiting the scope.

Claims

1. A computer-implemented method for selecting the size of an oral cavity brush head (10) for an individual, the computer-implemented method comprising: Receive (500) data associated with the individual; Based on the received data, a potential parameter vector of the jawbone (310) of the individual is generated (502) using multiple machine learning models; The dimensions of the jawbone are measured (504) based on the latent parameter vector of the jawbone; and The size of the oral cavity brush head is selected (506) based on the measured size of the jawbone.

2. The computer-implemented method of claim 1, wherein the data includes an image of the individual's face, and wherein generating the latent parameter vector of the jawbone using the plurality of machine learning models comprises: Using the first machine learning model (306) of the plurality of machine learning models, and using the image of the individual's face as input (300, 302), a latent parameter vector of the individual's head is generated; as well as The potential parameter vector of the head is converted into the potential parameter vector of the jawbone using the second machine learning model (308) of the plurality of machine learning models.

3. The computer-implemented method of claim 1, wherein the data includes an image of the individual's face, and generating the latent parameter vector of the jawbone using the plurality of machine learning models comprises: Using the third machine learning model (312) of the plurality of machine learning models, the image of the individual's face is used as input to generate a potential parameter vector of the individual's jaw.

4. A computer-implemented method according to any one of the preceding claims, wherein the data includes metadata, and wherein generating a latent parameter vector of the jawbone of the individual based on the face in the received image using one or more machine learning models comprises: Using the fourth machine learning model (314) of the plurality of machine learning models, the potential parameter vector of the jawbone is generated by inputting the metadata into the fourth machine learning model.

5. A computer-implemented method according to any one of claims 1 to 3, wherein the data includes metadata, and wherein generating a latent parameter vector of the jawbone of the individual based on the face in the received image using one or more machine learning models comprises: Using the fifth machine learning model (316) of the one or more machine learning models, and using the metadata as input, a potential parameter vector of the individual's head is generated; as well as The potential parameter vector of the head is converted into the potential parameter vector of the jawbone using a second machine learning model from the one or more machine learning models.

6. The computer-implemented method according to any one of the preceding claims, wherein generating the latent parameter vector of the jawbone using the plurality of machine learning models comprises: The aggregator machine learning model (305) among the plurality of machine learning models is used to fuse two or more latent parameter vectors of the jawbone; Each vector corresponds to the output of one of the multiple machine learning models.

7. A computer-implemented method according to any one of the preceding claims, wherein measuring the size of the jawbone based on the latent parameter vector of the jawbone comprises: The potential parameter vector of the jawbone is input into the sixth machine learning model (318) to directly obtain the size of the jawbone.

8. A computer-implemented method according to any one of claims 1 to 6, wherein measuring the size of the jawbone based on the latent parameter vector of the jawbone comprises: Input the latent parameter vector of the jawbone into the seventh machine learning model (319) to obtain a three-dimensional model of the jawbone; as well as The dimensions of the jawbone are measured based on the obtained three-dimensional model of the jawbone.

9. A computer-implemented method according to any one of the preceding claims, wherein selecting the size of the oral cavity brush head based on the measured dimensions of the jawbone comprises: The measured dimensions are input into the eighth machine learning model (324) to obtain a selection of the size of the oral cavity component.

10. The computer-implemented method according to any one of claims 1 to 8, wherein selecting the size of the oral cavity brush head based on the measured size of the jawbone comprises: The measured dimensions are input into a rule-based mathematical model (326) to obtain a selection of the size of the oral cavity component.

11. A computer-implemented method according to any one of the preceding claims, wherein the dimension is at least one of the following: the length of the jawbone; the angle between the molars, incisors and canines; and the thickness of the molars.

12. A computer-implemented method for training a machine learning model (MLM) to generate a latent parameter vector of an individual's jawbone based on a face from an image, the computer-implemented method comprising: Receives (600) multiple images and multiple facial scans; The three-dimensional jawbone mesh is registered with each scan in the multiple facial scans (602). The registered mesh was deconstructed using principal component analysis (604) to obtain the potential parameter vector of the jawbone; The multiple facial images are input (606) into a machine learning model to generate a corresponding latent parameter vector for the jawbone; Calculate the error between the latent parameter vector generated by (608) and the obtained latent parameter vector; and Update (610) the parameters of the machine learning model to minimize the error.

13. A computer-implemented method for training a machine learning model (MLM) to generate a latent parameter vector of an individual's head based on a face from an image, the computer-implemented method comprising: Receives (700) multiple images and multiple facial scans; Register the 3D head mesh with each scan in the multiple facial scans (702). Principal component analysis is used to deconstruct the registered mesh (704) to obtain the potential parameter vector of the head; The multiple facial images are input (706) into a machine learning model to generate a corresponding latent parameter vector for the head; Calculate the error between the latent parameter vector generated by (708) and the obtained latent parameter vector; and Update (710) the parameters of the machine learning model to minimize the error.

14. A transient or non-transitory computer-readable medium having instructions stored thereon, which, when executed by one or more processors, cause the one or more processors to perform the computer-implemented method of any one of the preceding claims.

Images