WO2023249776A1 - Mesures de capacité simulées pour un apprentissage de reconnaissance d'expression faciale - Google Patents

Mesures de capacité simulées pour un apprentissage de reconnaissance d'expression faciale Download PDF

Info

Publication number
WO2023249776A1
WO2023249776A1 PCT/US2023/022799 US2023022799W WO2023249776A1 WO 2023249776 A1 WO2023249776 A1 WO 2023249776A1 US 2023022799 W US2023022799 W US 2023022799W WO 2023249776 A1 WO2023249776 A1 WO 2023249776A1
Authority
WO
WIPO (PCT)
Prior art keywords
simulated
facial expression
human face
facial
capacitance measurements
Prior art date
Application number
PCT/US2023/022799
Other languages
English (en)
Inventor
Jouya JADIDIAN
Calin CRISTIAN
Petre-Alexandru ARION
Original Assignee
Microsoft Technology Licensing, Llc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US17/809,438 external-priority patent/US20230419722A1/en
Application filed by Microsoft Technology Licensing, Llc filed Critical Microsoft Technology Licensing, Llc
Publication of WO2023249776A1 publication Critical patent/WO2023249776A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/174—Facial expression recognition
    • G06V40/176—Dynamic expression
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T13/00—Animation
    • G06T13/20—3D [Three Dimensional] animation
    • G06T13/40—3D [Three Dimensional] animation of characters, e.g. humans, animals or virtual beings
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

  • a computing device may be programmed to detect and respond to movements of a human user, such as by recognizing the user’s facial expressions.
  • a wearable device equipped with suitable radio frequency (RF) antennas may generate an e-field in proximity to the user’s body. While generating the e-field, electrical conditions at each RF antenna may vary depending on the distance between the RF antenna and a nearby conductive surface. For instance, the capacitance between the RF antenna and the user’s skin may vary as the distance between them changes. This can enable the device to monitor electrical conditions at the RF antennas to detect and characterize the user’s movements.
  • RF radio frequency
  • a neural network may be used to output a predicted facial expression of a human user based at least on RF antenna capacitance measurements as input data. Due to the wide variety of facial expressions a user can make, and the wide range of diversity between the faces of different individuals, it is difficult to train a neural network for this purpose.
  • a method for training a neural network for facial expression recognition includes recognizing a plurality of digital human face models. For each of the plurality of digital human face models, a plurality of simulated facial expressions are simulated. Simulated capacitance measurements for an array of simulated radio frequency (RF) antennas are found for each of the plurality of simulated facial expressions. The simulated capacitance measurements for each simulated facial expression are provided as input training data to a neural network configured to output facial expression parameters based on input capacitance measurements.
  • RF radio frequency
  • FIG. 1 schematically illustrates recognizing the facial expression of a human user.
  • FIG. 2 schematically shows an example head-wearable device.
  • FIG. 3 schematically illustrates measuring the capacitance between radio frequency (RF) antennas and the skin of a human user.
  • RF radio frequency
  • FIG. 4 illustrates an example method for training a neural network for facial expression recognition.
  • FIG. 5 schematically depicts a plurality of digital human face models.
  • FIG. 6 schematically illustrates simulating a plurality of simulated facial expressions for a digital human face model.
  • FIGS. 7A-7C schematically illustrate finding simulated capacitance measurements for an array of simulated RF antennas.
  • FIG. 8 schematically illustrates distributing RF simulations for different digital human face models to compute nodes of a distributed computing system.
  • FIG. 9 schematically illustrates distributing RF simulations for different simulated facial expressions to compute nodes of a distributed computing system.
  • FIG. 10 schematically illustrates providing simulated capacitance measurements to a neural network as input training data.
  • FIG. 11 schematically illustrates outputting predicted facial expression parameters based on simulated capacitance measurements.
  • FIG. 12 schematically shows an example computing system.
  • a computing device may detect movement of a human user in various ways.
  • a computing device may include or interface with one or more radio frequency (RF) antennas configured to expose a body surface of the human user to an e-field - e.g., by driving the RF antennas with a high frequency alternating current.
  • RF radio frequency
  • the computing device may obtain information regarding the position and/or movements of the user’s body, as electrical conditions at each RF antenna can vary based at least on the proximity between the RF antenna and the user. This may beneficially enable detection and classification of specific movements of the user’s body, such as specific facial expressions, in a manner that is beneficially agnostic to different skin tones, skeletal structures, and ambient lighting conditions.
  • an array of RF antennas in proximity to a user’s face may be used to recognize the user’s current facial expression. For instance, as part of a user’s face moves (e.g., lips, nose, cheeks, eyebrows, eyelids), the capacitance between the user’s skin and a nearby RF antenna of the device may change.
  • a computing system may implement a neural network configured to output predictions about the user’s facial expression, based at least in part on input capacitance measurements corresponding to different RF antennas.
  • the present disclosure is directed to techniques for training a neural network for human facial recognition using input training data generated by simulated RF antennas and simulated digital human face models. More particularly, for each of a plurality of digital human face models, a computing system simulates a plurality of different facial expressions, and then simulates interactions between the digital human face model and an array of simulated RF antennas. In this manner, the computing system generates sets of simulated capacitance measurements corresponding to the simulated RF antennas, which can be used as input training data for training a neural network. The trained neural network can then be used to predict the facial expressions of real-world human users based on measurements from real-world RF antennas.
  • the system can beneficially generate simulated capacitance measurements that are more representative of the diversity exhibited by real-world human users. This may have the technical effect of improving human-computer interaction, as devices implementing the trained neural network may more accurately predict the facial expressions of real human users. Furthermore, the techniques described herein may beneficially reduce the associated costs of training a neural network to accurately predict human facial expressions, by allowing training data to be generated more quickly and by reducing manual effort on the part of human developers.
  • the techniques described herein may preserve the privacy of individual human users, as the neural network may beneficially be trained using little to no input training data that is generated from real humans. Furthermore, use of the trained neural network at runtime to predict the facial expressions of real human users may beneficially preserve user privacy, as the capacitance values measured for real RF antennas may lack the specificity required to uniquely identify real humans. In other words, while the collected capacitance measurements are useable to predict the facial expression of a human user, they generally do not provide sufficient detail to accurately distinguish any particular human from other humans that may have similar facial features.
  • FIG. 1 schematically illustrates an example scenario in which the facial expression of a real-world human user is recognized.
  • FIG. 1 shows a human user 100 in a real -world environment 102.
  • the user is wearing a head-wearable device 104, which is configured to collect data relating to the user’s face 106.
  • Such data is useable to output a predicted facial expression 108 of the user, shown in FIG. 1 as being applied to a digital representation 110 of the user’s face.
  • head-wearable device 104 may include a plurality of RF antennas configured to expose the surface of the user’s face to an e-field. By monitoring electrical conditions at each of the plurality of RF antennas, the device may assess the positions and/or movements of different aspects of the user’s face, enabling the user’s facial expression to be predicted.
  • Head-wearable device 104 may, for example, take the form of a head-mounted display device configured to present augmented and/or virtual reality experiences to the user’s eyes.
  • the head-wearable device may in some cases include one or more near-eye displays configured to present computer-generated imagery to the user.
  • headwearable device 104 is a non-limiting example.
  • any suitable computing device may be used to collect data useable for predicting the facial expression of a human user, provided that the computing device includes or interfaces with a suitable set of RF antennas.
  • Such a device may have any suitable capabilities, hardware configuration, and form factor.
  • headwearable device 104 may in some cases be implemented as computing system 1200 described below with respect to FIG. 12.
  • a user’s facial expression may in some cases be predicted by one or more computing devices other than the device that actually monitors electrical conditions at the RF antennas.
  • head-wearable device 104 may collect capacitance measurements for on-board RF antennas, then transmit such measurements to one or more other computing devices (e.g., a local computer and/or remote server) configured to predict the user’s facial expression based on the capacitance measurements.
  • the user’s facial expression may be predicted partially or entirely by the same device that includes the RF antennas - e.g., the predicted facial expression of user 100 may be output entirely by head-wearable device 104.
  • the predicted facial expression of the human user is shown as if applied to a digital representation 110 of a human head.
  • digital representation 110 may be intended for graphical rendering and display to other users, such as in a shared virtual experience (e.g., teleconference, video game).
  • the rendered representation of the human may be updated substantially in real-time, improving the immersion of the user and/or any other users in the virtual experience.
  • the digital representation may be specifically generated to resemble human user 100 - e.g., based on real-world images captured of the human user.
  • the digital representation need not resemble the human user, but rather may have any other suitable appearance, such as a generic human representation, a representation created by the user, a cartoonish or non-human appearance, etc.
  • the predicted facial expression of a human user need not be graphically rendered or displayed at all. Rather, the predicted facial expression of a human user may be expressed as any suitable computer data structure and used for any number of suitable purposes besides representing the human user in a virtual experience. As one example, a human user may provide inputs to a computing device using facial expressions (e.g., winking to control a software application). Thus, it may be beneficial to collect data useable to predict the facial expression of a human user, even if the predicted facial expression is not displayed.
  • FIG. 2 schematically shows head-wearable device 104 in more detail.
  • the headwearable device has a similar form factor to a pair of eyeglasses, having a wearable frame, a left eyepiece 200L, and a right eyepiece 200R.
  • the head-wearable device may in some cases be implemented as a head-mounted display device, and thus either or both of the left and right eyepieces may in some cases be configured to present computer-generated imagery to a user’s eyes - e.g., as part of a virtual or augmented reality experience.
  • head-wearable device 104 is a non-limiting example.
  • an alternate head-wearable device may have a form factor similar to a pair of goggles (e.g., secured by a strap around the user’s head), or a helmet equipped with a visor.
  • a head-wearable device need not include any display componentry, and need not include any components disposed in front of a user’ s eyes.
  • any suitable electronic device including suitable RF antennas may be used to collect capacitance measurements for the RF antennas indicative of a user’s facial expression, regardless of the specific form factor of the device, and regardless of which other functions the device may perform.
  • such a device need not have a form factor that is wearable on a human head.
  • head-wearable device 104 includes a plurality of RF antennas 202A-202H disposed along the left and right eyepieces. It will be understood that the specific arrangement shown in FIG. 2 is non-limiting. Rather, a device may include and/or interface with any suitable number of RF antennas. The plurality of RF antennas may each have any suitable positions relative to one another, the computing device, and the human user.
  • the RF antennas may be implemented using any suitable hardware.
  • the RF antennas may be implemented as transparent glass antennas, which may beneficially enable the RF antennas to be integrated into a near-eye display.
  • the RF antennas may be constructed from any suitable non-transparent materials.
  • the RF antennas may have any suitable appearance, and may or may not be visible to the human user and/or an outside observer.
  • any or all of the plurality of RF antennas may be driven to influence electrical conditions in the vicinity of a human user.
  • Each individual RF antenna may in some cases expose a different part of the human user’s body surface to an e-field.
  • one or more RF antennas may generate an e-field in the vicinity of the user’s eyes, while one or more other RF antennas generate an e-field in the vicinity of the nose, and so on to achieve a desired coverage of the user’s face. This may enable the computing device to detect movements of the user’s face, and thereby recognize the user’s facial expression.
  • proximity of the conductive human skin in a near-field region relative to the plurality of RF antennas may disturb an e-field generated by driving the plurality of RF antennas with drive signals, thereby influencing characteristics of the circuitry at one or more of the plurality of RF antennas.
  • movement of conductive skin near an RF antenna e.g., caused by movement of muscles under the skin
  • detected changes in electrical conditions at the plurality of RF antennas from one time frame to another may be used to evaluate movements of the human user between the two time frames.
  • movement of the human user may change characteristics of a circuit that the user is also part of.
  • the system may be described as an oscillator having a resonant frequency that is sensitive to changes in parasitic capacitance.
  • the change in frequency of the oscillator may be caused by a change in capacitive loading at the user’s body surface, and this may be affected by movements of the user.
  • Detected changes in electrical conditions at any of the RF antennas e.g., a change in voltage
  • FIG. 3 which shows an example computing device 300 that includes a plurality of RF antennas 302A-302E.
  • Each of the plurality of RF antennas is exposing a part of a human user 304 to an e-field 306.
  • the part of the human user 304 generically represents any suitable part of a human user’s body (e.g., the user’s face).
  • the computing device collects capacitance measurements 308A-308E for each of the plurality of RF antennas. It will be understood that the capacitance measurements may be determined or calculated in any suitable way, and derived from any suitable raw data measured from the RF antennas - e.g., changes in voltage, current, amplitude, and/or signal phase.
  • the drive signals applied to the plurality of RF antennas to generate the e-field may have any suitable characteristics.
  • the plurality of RF antennas may each be driven to generate the e-field using drive signals having a same voltage and phase. In other examples, however, drive signals having different voltage and/or phase may be applied to two or more of the plurality of RF antennas.
  • the specific frequencies used to drive the RF antennas, and the electrical characteristics of the larger circuit as a whole, may be tuned to achieve a desired level of sensitivity and power draw.
  • an RF antenna exposing conductive human skin to an e-field may cause capacitive loading of the human skin. This may result in flow of complex or real current between the RF antenna and human user depending on the specific circuit design, the frequency of the drive signal, and the proximity of the human skin.
  • operation of the system may be characterized by different signal response curves corresponding to capacitive, inductive, and resonance modes for any particular RF antenna.
  • the behavior of the system may transition between each of these signal response curves depending on the current amount of capacitance between the RF antenna and the human skin, influenced by the proximity of the human user to the RF antenna.
  • the slope of each signal response curve is based at least in part on the Q-factor, where a higher Q-factor results in a steeper curve, and therefore a greater signal response for a particular change in capacitance.
  • the circuit may beneficially be tuned such that each RF antenna primarily operates in the capacitive mode, which is characterized by relatively low power draw as compared to the resonance and inductive modes. However, as the distance between the RF antenna and human skin changes, a relatively large change in signal may be observed as the circuit transitions from the capacitive curve to the resonance curve, enabling the movement of the human user to be detected with a relatively high confidence.
  • FIG. 4 illustrates an example method 400 for training a neural network for facial expression recognition.
  • Method 400 may be implemented by any suitable computing system of one or more computing devices - e.g., head-wearable device 104.
  • Any computing device(s) implementing method 400 may have any suitable capabilities, hardware configuration, and form factor. In some cases, steps of method 400 may be distributed between two or more devices of a distributed computing system - e.g., via a client/server model. Any or all steps of method 400 may be repeated or looped at any suitable time and for any suitable reason. In some examples, method 400 may be implemented by computing system 1200 described below with respect to FIG. 12.
  • method 400 includes recognizing a plurality of digital human face models.
  • a “digital human face model” refers to any suitable computer data structure that represents the three- dimensional shape of a human face.
  • a digital human face model may be encoded using any suitable file format for three-dimensional models.
  • “Recognizing” the plurality of digital human face models may generally refer to loading the digital human face models into computer memory of one or more different computing devices. For instance, the digital human face models may be accessed from computer storage, received from an external source (e.g., transmitted by another party, loaded from a database), or generated on-the-fly.
  • “recognizing” may also include the act of generating the human face model using a computer simulator, for example a three-dimensional modeling application that outputs a data structure representing the three-dimensional shape of a human face.
  • FIG. 5 schematically illustrates a plurality of digital human face models, including models 500A- C.
  • the plurality of digital human face models may include any suitable number of models, including more than three models.
  • Each digital human face model may differ according to one or more facial shape parameters, which may define positions and/or sizes of anatomical features of each digital human face model. More particularly, facial shape parameters may define the shape, size, position, and general appearance (e.g., skin tone) of any portion of the face model, such as the eyes, nose, lips, ears, brow, cheeks, etc.
  • the plurality of digital human face models may be representative of a wide variety of different human facial appearances - e.g., consistent with different ethnic backgrounds and sexual dimorphic characteristics.
  • the facial shape parameters may be expressed using any suitable format.
  • the facial shape parameters may be specified by the computer data structure used to encode the digital human face model - e.g., the facial shape parameters may take the form of coefficients or variables that can be altered to change the sizes/shapes/positions/appearances of portions of a human face model.
  • each digital human face model 500A-C is indicated by different faces having different shapes, sizes, and facial features.
  • a digital human face model need not be graphically rendered for display. Rather, each digital human face model may be expressed as a set of computer data that can be interpreted and manipulated (e.g., to simulate various facial expressions) without being displayed for viewing by a real human user.
  • the digital human face models may have any suitable source.
  • the plurality of digital human face models may be generated based at least in part on a plurality of images of real- world human faces. For example, such images may be collected from human subjects who consent to images of their faces being used in neural network training. Additionally, or alternatively, the plurality of digital human face models may be computer-generated models that do not correspond to any real humans.
  • a human developer and/or automated software application may generate digital human face models - e.g., by starting with one or more base models, and altering the facial shape parameters for the one or more base models in a random or targeted manner. In cases where the facial shape parameters are varied randomly, anatomical constraints may be used to prevent generation of faces that would not be anatomically possible.
  • method 400 includes, for each of the plurality of digital human face models, simulating a plurality of simulated facial expressions.
  • FIG. 6 again shows digital human face model 500A, along with three simulated facial expressions 600A-600C applied to the digital human face model.
  • any suitable number of different facial expressions may be simulated for each digital human face model, including more than three different expressions.
  • the different simulated facial expressions are represented in FIG. 6 by faces forming the simulated expressions, it will be understood that a “simulated facial expression” of a digital human face model may be generated and manipulated without being graphically rendered for display.
  • each simulated facial expression may be encoded as a plurality of facial expression parameters.
  • the simulated facial expressions 600 A-C are encoded by facial expression parameters 602A-C.
  • the facial expression parameters may take the form of any suitable computer data useable to expresses the facial expression applied to a digital human face model. For instance, by varying one or more of the facial expression parameters, the position or general state of one or more facial features of the digital human face model may be changed - e.g., one or more facial expression parameters may govern the position of an eyebrow of the digital human face model.
  • a facial expression parameter may define a movement or offset of a particular facial feature from its default or neutral position/state.
  • one facial expression parameter may define an upwards movement of an eyebrow from its neutral position to a raised position.
  • the facial expression parameters may be standardized from one digital human face model to another - e.g., a given set of facial expression parameters applied to two different digital human face models may cause the two models to form relatively similar facial expressions.
  • the plurality of different simulated facial expressions may be simulated in any suitable way.
  • the plurality of simulated facial expressions may be pre-defined facial expressions each having pre-defined facial expression parameters.
  • human developers may define a set of expressions to be simulated for each of the plurality of digital human face models. This can beneficially enable the developers to ensure that simulated capacitance measurements for relatively common expressions (e.g., smiling, frowning) are represented in the input training data.
  • any or all of the plurality of simulated facial expressions may be simulated by randomly varying the facial expression parameters for the digital human face model to generate the simulated facial expressions.
  • the random variations applied to the facial expression parameters may be constrained by one or more anatomical constraints to avoid simulating facial expressions that would be difficult or impossible for a human face to perform.
  • the state of one or more facial expression parameters e.g., indicating that the face’s mouth is closed
  • may limit the range of values that can be used for one or more other expression parameters e.g., to prevent the tongue having a position outside of the closed mouth.
  • method 400 includes, for each of the plurality of simulated facial expressions, finding simulated capacitance measurements for an array of simulated RF antennas.
  • FIGS. 7A-7C show simulated facial expression 600A of digital human face model 500A.
  • FIG. 7A also shows an array of simulated RF antennas, some of which are labeled as RF antennas 700A-D. For instance, these may be simulated versions of the physical RF antennas schematically depicted in FIG. 2.
  • the RF antennas are integrated into a simulated head-wearable device 702, which may be a simulated version of head-wearable device 104. This may enable the system to obtain simulated capacitance measurements that would be consistent with actual capacitance measurements measured by real RF antennas.
  • finding the simulated capacitance measurements may include calculating the simulated capacitance measurements based at least on simulated interactions between an e-field of the simulated RF antennas and the digital human face model.
  • This is schematically illustrated with respect to FIG. 7B, showing a side profile of digital human face model 500A.
  • the digital human face model is an entire simulated human head, although it will be understood that this need not be the case. Rather, in some examples, the digital human face model may represent a human face while omitting other portions of the human head.
  • simulated RF antennas of the simulated head-wearable device are generating a simulated e-field 704 in proximity to the digital human face model 500A.
  • the system calculates a set of simulated capacitance measurements 706A.
  • These simulated capacitance measurements may be calculated in any suitable way.
  • the system may calculate the simulated capacitance measurements as though the surface of the digital human face model is a perfect electrical conductor. This may beneficially improve the robustness of the trained neural network to scenarios where the electrical conductivity of a user’ s body surface varies - e.g., due to ambient humidity, temperature, or weather conditions such as rainfall.
  • the system may simulate interactions between the e-field and one or more human tissue layers - e.g., using suitable specific absorption rate (SAR) models corresponding to human skin, fat, bone, etc.
  • SAR specific absorption rate
  • the simulated capacitance measurements may be calculated according to Maxwell’s equations.
  • the computing system may find the simulated capacitance measurements based at least in part on a distance between each RF antenna and a surface of the digital human face model (e.g., a nearest surface to the RF antenna). For example, this may include determining the length of a ray traced from a simulated RF antenna to the surface of the digital human face model (e.g., a ray may be traced from simulated RF antenna 700A of FIG. 7A to the surface of digital human face model 500A). More particularly, a plurality of rays may be radially traced from multiple points on the surface of a simulated antenna patch.
  • Any rays that reach a grounded geometry e.g., surface of a face model
  • the calculated distances for each terminated ray may then be collected into an array for each separate antenna and aggregated into a set of scalars. It has been observed that the resulting set of values correlates strongly with simulated capacitance measurements calculated for the same geometry via simulated e-field interactions.
  • the system may find the simulated capacitance measurements based on a simulated spatial arrangement of the array of simulated RF antennas relative to the digital human face model. For instance, this simulated spatial arrangement may be similar to the real spatial arrangement of RF antennas integrated into a physical device used to predict the facial expression of real-world human users. Furthermore, in some cases, the computing system may simulate two or more different spatial arrangements of the array of simulated RF antennas relative to the digital human face model for each simulated facial expression.
  • simulating the two or more different spatial arrangements for the array of simulated RF antennas may include simulating two or more different positions of the head- wearable device relative to the digital human face model.
  • FIG. 7C This is schematically illustrated with respect to FIG. 7C, again showing a side profile of digital human face model 500A.
  • the position of simulated headwearable device 702 has changed. Specifically, the head-wearable device is simulated at a higher angle relative to a transverse plane through the digital human face model.
  • the system finds another set of simulated capacitance measurements 706B for the changed position of the simulated headwearable device. Simulating two or more different spatial arrangements of the array of RF antennas (e.g., two different positions of a head-wearable device) can improve the accuracy of the predictions output by the trained neural network.
  • two different human users may wear a head-wearable device in different positions (e.g., angled higher or lower relative to a transverse plane, positioned closer to or further from the user’s eyes), depending on their head shape/size and comfort level.
  • the same user may wear a head-wearable device in different positions while in use - e.g., the device may shift or tilt relative to the user’s face while in use.
  • the system may simulate different possible numbers and/or arrangements of RF antennas.
  • the computing system may generate simulated capacitance measurements corresponding to different potential versions or iterations of the physical hardware that will be used to predict the facial expressions of real human users. Different simulation passes can, for instance, cover scenarios where one particular RF antenna is bigger or smaller, has a different position with respect to the spatial array of RF antennas, or is omitted entirely. In this manner, the computing system and/or human developers may determine that some potential arrangements of RF antennas give more useful capacitance measurements, and this can guide future hardware development.
  • Simulating a plurality of different facial expressions for each of a plurality of digital human face models, as well as finding simulated capacitance measurements for each simulated facial expression, can include a significant expenditure of computational resources.
  • any or all of the simulation operations described herein may in some cases be distributed between two or more nodes of a distributed computing system to enable simulations to be performed in parallel.
  • a “compute node” can refer to a distinct physical computing device, such as an individual server computer. Additionally, or alternatively, a “compute node” can refer to a logical entity, such as a virtual machine implemented on a server. Thus, a single computing device may in some cases instantiate two or more different compute nodes. In some cases, a compute node may be implemented as computing system 1200 described below with respect to FIG. 12.
  • FIG. 8 schematically shows a distributed computing system 800 that includes at least four compute nodes 802A-802D.
  • a distributed computing system may include any suitable number of compute nodes, and can potentially include hundreds, thousands, or more different compute nodes.
  • the different compute nodes may communicate over any suitable computer network - e.g., a local network or a wide-access network such as the Internet.
  • simulations for two or more of the digital human face models are distributed between two or more different compute nodes of the distributed computing system.
  • compute nodes 802B performs RF simulations 804A for a first digital human face model (referred to as model 1). This may include simulating a plurality of different facial expressions for the digital human face model, and finding simulated capacitance measurements for each simulated facial expression, as described above.
  • Compute nodes 802C and 802D perform RF simulations 804B and 804C for different digital human face models (e.g., model 2 and model 3).
  • compute nodes 802B-802D generate sets of simulated capacitance measurements 806A-806C, which are transmitted to compute node 802A.
  • Compute node 802A may use the simulated capacitance measurements as neural network input training data 808, as will be described in more detail below.
  • compute node 802A receives, from a plurality of compute nodes (e.g., nodes 802B-D), simulated capacitance measurements based at least on a simulated spatial arrangement of an array of simulated RF antennas relative to at least one digital human face model, the digital human face model having at least one simulated facial expression.
  • each of compute nodes 802B-802D perform simulations for a different digital human face model.
  • the same compute node may perform simulations for two or more different digital human face models, and/or simulations for the same digital human face model may be performed by two or more different compute nodes.
  • system 900 includes at least four compute nodes 902A-902D.
  • simulations for two or more different simulated facial expressions of a digital human face model are distributed between two or more compute nodes of the distributed computing system.
  • compute node 902B performs RF simulations 904A for a first simulated facial expression (referred to as expression 1).
  • Compute nodes 902C and 902D perform RF simulations 904B and 904C for different simulated facial expressions (e.g., expressions 2 and 3) for the same digital human face model.
  • compute nodes 902B-902D generate sets of simulated capacitance measurements 906A-906C, which are transmitted to compute node 902A.
  • Compute node 902A may use the simulated capacitance measurements as neural network input training data 908, as will be described in more detail below.
  • method 400 includes providing the simulated capacitance measurements for each simulated facial expression as input training data to a neural network configured to output facial expression parameters based on input capacitance measurements.
  • the neural network may be implemented in any suitable way, using any suitable machine learning (ML) techniques.
  • ML machine learning
  • suitable ML technologies that may be used to implement and train the neural network will be described below with respect to FIG. 12.
  • FIG. 10 schematically illustrates providing simulated capacitance measurements to a neural network as input training data.
  • a neural network 1000 receives a set of input training data 1002, which includes various simulated capacitance measurements 1004. Based on such measurements, the neural network outputs predicted facial expression parameters 1006. In other words, for a given set of input capacitance measurements, the neural network may output a set of facial expression parameters predicted to be consistent with the input capacitance measurements.
  • the neural network may be configured to output facial expression parameters for two or more different digital human face models of the plurality of digital human face models.
  • the simulated capacitance measurements 1004 included in the set of input training data may include capacitance measurements simulated for two or more different digital human face models.
  • different neural networks may be trained to output facial expression parameters for each digital human face model of the plurality of digital human face models.
  • neural network 1000 may correspond to one specific digital human face model (e.g., model 500A), and the simulated capacitance measurements used to train the neural network may be simulated based on that model.
  • Other neural networks may be trained for other digital human face models.
  • a real device attempting to predict the facial expression of a human user may first determine which of the plurality of different digital human face models the human user is most similar to. From there, the device may provide real capacitance measurements for the user’s face to the appropriate neural network. As one example, the device may use an image of the user’s face to determine which of the plurality of digital human face models is most similar to the real user’s face (e.g., which of the digital human face models is the nearest neighbor to the user’s face).
  • the neural network may be trained in any suitable way.
  • the simulated capacitance measurements provided as input training data may be labeled with simulated facial expression parameters for a simulated facial expression that the capacitance measurements correspond to.
  • the computing system generates a set of simulated capacitance measurements 706A for simulated facial expression 600A of digital human face model 500B.
  • the simulated expression parameters that encode simulated facial expression 600A may be used as ground truth labels when simulated capacitance measurements 706A are provided to the neural network as input training data.
  • method 400 includes, at 410, training the neural network to reduce an error between the output facial expression parameters and the simulated facial expression parameters labeling the input training data.
  • a set of input capacitance measurements may result in an output set of facial expression parameters.
  • backpropagation may be used to adjust parameters of the neural network, such that future passes will result in output facial expression parameters that are more similar to the ground truth labels - e.g., the simulated facial expression parameters used to encode the facial expression for which the input capacitance measurements were simulated.
  • FIG. 11 schematically illustrates the process of outputting predicted facial expression parameters given a set of input simulated capacitance measurements.
  • the operations illustrated in FIG. 11 may be in the process of training a neural network, such as neural network 1000 in FIG. 10.
  • FIG. 11 shows an example set 1100 of simulated capacitance measurements simulated for various RF antennas of a simulated head-wearable device.
  • seven simulated RF antennas are used, although it will be understood that this is non-limiting.
  • the various simulated RF antennas may correspond to different specific regions of a human face (e.g., a digital human face model).
  • the “LEB” RF antenna may correspond to a left eyebrow of the digital human face model
  • the “NOSE” RF antenna corresponds to a nose of the digital human face model, and so on.
  • the specific values shown in FIG. 11 are non-limiting, and that any number of different measurements may be simulated for each RF antenna.
  • the different rows of the set 1100 of simulated capacitance measurements may correspond to different simulated facial expressions, different simulated spatial arrangements of the plurality of RF antennas, different sequential time frames, different simulation iterations of the same scenario, etc.
  • min-max normalization is applied to the set 1100 of simulated capacitance measurements to give a normalized set 1102 of simulated capacitance measurements, although this is non-limiting. In general, any suitable data normalization, filtering, or other processing may optionally be applied to the set of simulated capacitance measurements.
  • the system Based at least on the normalized set 1102 of simulated capacitance measurements, the system outputs a predicted set 1104 of facial expression parameters. As discussed above, these may take any suitable form depending on the implementation. In FIG. 11 only six facial expression parameters are shown, although it will be understood that this is non-limiting. Rather, the system may output any suitable number of different facial expression parameters. In cases where multiple different simulated capacitance measurements are provided for the same RF antenna, the system may in some cases output multiple different values for each facial expression parameter.
  • the methods and processes described herein may be tied to a computing system of one or more computing devices.
  • such methods and processes may be implemented as an executable computer-application program, a network-accessible computing service, an application-programming interface (API), a library, or a combination of the above and/or other compute resources.
  • API application-programming interface
  • FIG. 12 schematically shows a simplified representation of a computing system 1200 configured to provide any to all of the compute functionality described herein.
  • Computing system 1200 may take the form of one or more personal computers, network-accessible server computers, tablet computers, home-entertainment computers, gaming devices, mobile computing devices, mobile communication devices (e.g., smart phone), virtual/augmented/mixed reality computing devices, wearable computing devices, Internet of Things (loT) devices, embedded computing devices, and/or other computing devices.
  • Computing system 1200 includes a logic subsystem 1202 and a storage subsystem 1204.
  • Computing system 1200 may optionally include a display subsystem 1206, input subsystem 1208, communication subsystem 1210, and/or other subsystems not shown in FIG. 12.
  • Logic subsystem 1202 includes one or more physical devices configured to execute instructions.
  • the logic subsystem may be configured to execute instructions that are part of one or more applications, services, or other logical constructs.
  • the logic subsystem may include one or more hardware processors configured to execute software instructions. Additionally, or alternatively, the logic subsystem may include one or more hardware or firmware devices configured to execute hardware or firmware instructions.
  • Processors of the logic subsystem may be single-core or multi-core, and the instructions executed thereon may be configured for sequential, parallel, and/or distributed processing. Individual components of the logic subsystem optionally may be distributed among two or more separate devices, which may be remotely located and/or configured for coordinated processing. Aspects of the logic subsystem may be virtualized and executed by remotely-accessible, networked computing devices configured in a cloudcomputing configuration.
  • Storage subsystem 1204 includes one or more physical devices configured to temporarily and/or permanently hold computer information such as data and instructions executable by the logic subsystem. When the storage subsystem includes two or more devices, the devices may be collocated and/or remotely located. Storage subsystem 1204 may include volatile, nonvolatile, dynamic, static, read/write, read-only, random-access, sequential-access, location-addressable, file-addressable, and/or content-addressable devices. Storage subsystem 1204 may include removable and/or built-in devices. When the logic subsystem executes instructions, the state of storage subsystem 1204 may be transformed - e.g., to hold different data.
  • logic subsystem 1202 and storage subsystem 1204 may be integrated together into one or more hardware-logic components.
  • Such hardware-logic components may include program- and application-specific integrated circuits (PASIC / ASICs), program- and application-specific standard products (PSSP / ASSPs), system-on-a-chip (SOC), and complex programmable logic devices (CPLDs), for example.
  • PASIC / ASICs program- and application-specific integrated circuits
  • PSSP / ASSPs program- and application-specific standard products
  • SOC system-on-a-chip
  • CPLDs complex programmable logic devices
  • the logic subsystem and the storage subsystem may cooperate to instantiate one or more logic machines.
  • the term “machine” is used to collectively refer to the combination of hardware, firmware, software, instructions, and/or any other components cooperating to provide computer functionality.
  • “machines” are never abstract ideas and always have a tangible form.
  • a machine may be instantiated by a single computing device, or a machine may include two or more sub-components instantiated by two or more different computing devices.
  • a machine includes a local component (e.g., software application executed by a computer processor) cooperating with a remote component (e.g., cloud computing service provided by a network of server computers).
  • the software and/or other instructions that give a particular machine its functionality may optionally be saved as one or more unexecuted modules on one or more suitable storage devices.
  • Machines may be implemented using any suitable combination of state-of-the-art and/or future machine learning (ML), artificial intelligence (Al), and/or natural language processing (NLP) techniques.
  • techniques that may be incorporated in an implementation of one or more machines include support vector machines, multi-layer neural networks, convolutional neural networks (e.g., including spatial convolutional networks for processing images and/or videos, temporal convolutional neural networks for processing audio signals and/or natural language sentences, and/or any other suitable convolutional neural networks configured to convolve and pool features across one or more temporal and/or spatial dimensions), recurrent neural networks (e.g., long short-term memory networks), associative memories (e.g., lookup tables, hash tables, Bloom Filters, Neural Turing Machine and/or Neural Random Access Memory), word embedding models (e.g., GloVe or Word2Vec), unsupervised spatial and/or clustering methods (e.g., nearest neighbor algorithms, topological data analysis, and/or k-means
  • the methods and processes described herein may be implemented using one or more differentiable functions, wherein a gradient of the differentiable functions may be calculated and/or estimated with regard to inputs and/or outputs of the differentiable functions (e.g., with regard to training data, and/or with regard to an objective function).
  • a gradient of the differentiable functions may be calculated and/or estimated with regard to inputs and/or outputs of the differentiable functions (e.g., with regard to training data, and/or with regard to an objective function).
  • Such methods and processes may be at least partially determined by a set of trainable parameters. Accordingly, the trainable parameters for a particular method or process may be adjusted through any suitable training procedure, in order to continually improve functioning of the method or process.
  • Non-limiting examples of training procedures for adjusting trainable parameters include supervised training (e.g., using gradient descent or any other suitable optimization method), zeroshot, few-shot, unsupervised learning methods (e.g., classification based on classes derived from unsupervised clustering methods), reinforcement learning (e.g., deep Q learning based on feedback) and/or generative adversarial neural network training methods, belief propagation, RANSAC (random sample consensus), contextual bandit methods, maximum likelihood methods, and/or expectation maximization.
  • a plurality of methods, processes, and/or components of systems described herein may be trained simultaneously with regard to an objective function measuring performance of collective functioning of the plurality of components (e.g., with regard to reinforcement feedback and/or with regard to labelled training data). Simultaneously training the plurality of methods, processes, and/or components may improve such collective functioning.
  • one or more methods, processes, and/or components may be trained independently of other components (e.g., offline training on historical data).
  • display subsystem 1206 may be used to present a visual representation of data held by storage subsystem 1204. This visual representation may take the form of a graphical user interface (GUI).
  • GUI graphical user interface
  • Display subsystem 1206 may include one or more display devices utilizing virtually any type of technology.
  • display subsystem may include one or more virtual-, augmented-, or mixed reality displays.
  • input subsystem 1208 may comprise or interface with one or more input devices.
  • An input device may include a sensor device or a user input device. Examples of user input devices include a keyboard, mouse, touch screen, or game controller.
  • the input subsystem may comprise or interface with selected natural user input (NUI) componentry. Such componentry may be integrated or peripheral, and the transduction and/or processing of input actions may be handled on- or off-board.
  • NUI componentry may include a microphone for speech and/or voice recognition; an infrared, color, stereoscopic, and/or depth camera for machine vision and/or gesture recognition; a head tracker, eye tracker, accelerometer, and/or gyroscope for motion detection and/or intent recognition.
  • communication subsystem 1210 may be configured to communicatively couple computing system 1200 with one or more other computing devices.
  • Communication subsystem 1210 may include wired and/or wireless communication devices compatible with one or more different communication protocols.
  • the communication subsystem may be configured for communication via personal-, local- and/or wide-area networks.
  • the methods and processes disclosed herein may be configured to give users and/or any other humans control over any private and/or potentially sensitive data.
  • data Whenever data is stored, accessed, and/or processed, the data may be handled in accordance with privacy and/or security standards.
  • users or other stakeholders may designate how the data is to be used and/or stored.
  • user data Whenever user data is collected for any purpose, the user data may only be collected with the utmost respect for user privacy (e.g., user data may be collected only when the user owning the data provides affirmative consent, and/or the user owning the data may be notified whenever the user data is collected).
  • the user may opt-in and/or opt-out of data collection at any time. After data has been collected, users may issue a command to delete the data, and/or restrict access to the data. All potentially sensitive data optionally may be encrypted and/or, when feasible, anonymized, to further protect user privacy. Users may designate portions of data, metadata, or statistics/results of processing data for release to other parties, e.g., for further processing.
  • Data that is private and/or confidential may be kept completely private, e.g., only decrypted temporarily for processing, or only decrypted for processing on a user device and otherwise stored in encrypted form.
  • Users may hold and control encryption keys for the encrypted data.
  • users may designate a trusted third party to hold and control encryption keys for the encrypted data, e.g., so as to provide access to the data to the user according to a suitable authentication protocol.
  • the ML and/or Al components may make decisions based at least partially on training of the components with regard to training data. Accordingly, the ML and/or Al components may be trained on diverse, representative datasets that include sufficient relevant data for diverse users and/or populations of users. In particular, training data sets may be inclusive with regard to different human individuals and groups, so that as ML and/or Al components are trained, their performance is improved with regard to the user experience of the users and/or populations of users.
  • ML and/or Al components may additionally be trained to make decisions so as to minimize potential bias towards human individuals and/or groups.
  • Al systems when Al systems are used to assess any qualitative and/or quantitative information about human individuals or groups, they may be trained so as to be invariant to differences between the individuals or groups that are not intended to be measured by the qualitative and/or quantitative assessment, e.g., so that any decisions are not influenced in an unintended fashion by differences among individuals and groups.
  • ML and/or Al components may be designed to provide context as to how they operate, so that implementers of ML and/or Al systems can be accountable for decisions/assessments made by the systems.
  • ML and/or Al systems may be configured for replicable behavior, e.g., when they make pseudo-random decisions, random seeds may be used and recorded to enable replicating the decisions later.
  • data used for training and/or testing ML and/or Al systems may be curated and maintained to facilitate future investigation of the behavior of the ML and/or Al systems with regard to the data.
  • ML and/or Al systems may be continually monitored to identify potential bias, errors, and/or unintended outcomes.
  • a method for training a neural network for facial expression recognition comprises: recognizing a plurality of digital human face models; and for each of the plurality of digital human face models: simulating a plurality of simulated facial expressions; for each of the plurality of simulated facial expressions, finding simulated capacitance measurements for an array of simulated radio frequency (RF) antennas; and providing the simulated capacitance measurements for each simulated facial expression as input training data to a neural network configured to output facial expression parameters based on input capacitance measurements.
  • the input training data is labeled with simulated facial expression parameters for the simulated facial expression.
  • the method further comprises training the neural network to reduce an error between the output facial expression parameters and the simulated facial expression parameters labeling the input training data.
  • the simulated capacitance measurements are found based on a simulated spatial arrangement of the array of simulated RF antennas relative to the digital human face model with the simulated facial expression.
  • finding the simulated capacitance measurements includes calculating the simulated capacitance measurements based at least on simulated interactions between an e-field of the simulated RF antennas and the digital human face model.
  • the simulated capacitance measurements are calculated according to Maxwell’s equations.
  • finding the simulated capacitance measurements includes estimating each simulated capacitance measurement based at least on a length of a ray traced from a simulated RF antenna to a surface of the digital human face model.
  • the method further comprises, for each simulated facial expression, simulating two or more different spatial arrangements for the array of simulated RF antennas relative to the digital human face model.
  • the array of simulated RF antennas are integrated into a simulated head-wearable device, and wherein simulating the two or more different spatial arrangements for the array of simulated RF antennas includes simulating two or more different positions of the head-wearable device relative to the digital human face model.
  • the plurality of different human face models are generated based at least on a plurality of images of real-world human faces.
  • the plurality of simulated facial expressions are each encoded as a plurality of facial expression parameters, and wherein the plurality of simulated facial expressions are a plurality of pre-defined facial expressions each having pre-defined facial expression parameters.
  • the plurality of simulated facial expressions are each encoded as a plurality of facial expression parameters, and wherein simulating the plurality of simulated facial expressions includes randomly varying the facial expression parameters for the digital human face model to generate each simulated facial expression.
  • each of the plurality of digital human face models differ according to one or more facial shape parameters.
  • the one or more facial shape parameters define positions and/or sizes of anatomical facial features of the plurality of digital human face models.
  • different neural networks are trained to output facial expression parameters for each digital human face model of the plurality of digital human face models.
  • the neural network is configured to output facial expression parameters for two or more digital human face models of the plurality of digital human face models.
  • simulations for two or more of the digital human face models are distributed between two or more compute nodes of a distributed computing system.
  • simulations for two or more different simulated facial expressions of a digital human face model are distributed between two or more compute nodes of a distributed computing system.
  • a computing system comprises: a communications subsystem; a logic subsystem; and a storage subsystem holding instructions executable by the logic subsystem to: receive, from a plurality of compute nodes via the communications subsystem, simulated capacitance measurements based on a simulated spatial arrangement of an array of simulated radio frequency (RF) antennas relative to a digital human face model with a simulated facial expression; and provide the simulated capacitance measurements to a neural network configured to output facial expression parameters based on input capacitance measurements.
  • RF radio frequency
  • a method for training a neural network for facial expression modeling comprises: recognizing a plurality of digital human face models; and for each of the plurality of digital human face models: simulating a plurality of simulated facial expressions; for each of the plurality of simulated facial expressions, finding simulated capacitance measurements for an array of simulated radio frequency (RF) antennas based on a simulated spatial arrangement of the array of simulated RF antennas relative to the digital human face model with the simulated facial expression; providing the simulated capacitance measurements for each simulated facial expression as input training data to a neural network configured to output facial expression parameters based on input capacitance measurements, the input training data labeled with simulated facial expression parameters for the simulated facial expression; and training the neural network to reduce an error between the output facial expression parameters and the simulated facial expression parameters labeling the input training data.
  • RF radio frequency

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Human Computer Interaction (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Image Processing (AREA)

Abstract

Un procédé d'apprentissage d'un réseau neuronal pour une reconnaissance d'expression faciale comprend la reconnaissance d'une pluralité de modèles de visage humain numériques. Pour chacun de la pluralité de modèles de visage humain numériques, une pluralité d'expressions faciales simulées sont simulées. Des mesures de capacité simulées pour un réseau d'antennes radiofréquence (RF) simulées sont trouvées pour chacune de la pluralité d'expressions faciales simulées. Les mesures de capacité simulées pour chaque expression faciale simulée sont fournies en tant que données d'apprentissage d'entrée à un réseau neuronal configuré pour délivrer en sortie des paramètres d'expression faciale sur la base de mesures de capacité d'entrée.
PCT/US2023/022799 2022-06-24 2023-05-19 Mesures de capacité simulées pour un apprentissage de reconnaissance d'expression faciale WO2023249776A1 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
ROA202200361 2022-06-24
RO202200361 2022-06-24
US17/809,438 2022-06-28
US17/809,438 US20230419722A1 (en) 2022-06-24 2022-06-28 Simulated capacitance measurements for facial expression recognition training

Publications (1)

Publication Number Publication Date
WO2023249776A1 true WO2023249776A1 (fr) 2023-12-28

Family

ID=86851584

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2023/022799 WO2023249776A1 (fr) 2022-06-24 2023-05-19 Mesures de capacité simulées pour un apprentissage de reconnaissance d'expression faciale

Country Status (1)

Country Link
WO (1) WO2023249776A1 (fr)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9830507B2 (en) * 2011-03-28 2017-11-28 Nokia Technologies Oy Method and apparatus for detecting facial changes
US20230162531A1 (en) * 2021-11-22 2023-05-25 Microsoft Technology Licensing, Llc Interpretation of resonant sensor data using machine learning

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9830507B2 (en) * 2011-03-28 2017-11-28 Nokia Technologies Oy Method and apparatus for detecting facial changes
US20230162531A1 (en) * 2021-11-22 2023-05-25 Microsoft Technology Licensing, Llc Interpretation of resonant sensor data using machine learning

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
AWR DESIGN ENVIRONMENT: "Design Example: ADAS Automotive Radar System", 19 September 2018 (2018-09-19), XP093065498, Retrieved from the Internet <URL:https://www.youtube.com/watch?v=oDXIJTwowaE> [retrieved on 20230719] *
CARLA SIMULATOR: "CARLA Talks 2020 - Sensors in CARLA", 10 June 2020 (2020-06-10), XP093065497, Retrieved from the Internet <URL:https://www.youtube.com/watch?v=T8qCSet8WK0> [retrieved on 20230719] *
KOPACZKA MARCIN ET AL: "An automated method for realistic face simulation and facial landmark annotation and its application to active appearance models", 2016 SIXTH INTERNATIONAL CONFERENCE ON IMAGE PROCESSING THEORY, TOOLS AND APPLICATIONS (IPTA), IEEE, 12 December 2016 (2016-12-12), pages 1 - 6, XP033043821, DOI: 10.1109/IPTA.2016.7820979 *
MAIER F MICHAEL ET AL: "Environment perception simulation for radar stimulation in automated driving function testing", ELEKTROTECHNIK UND INFORMATIONSTECHNIK, SPRINGER VERLAG, WIEN, AT, vol. 135, no. 4, 28 June 2018 (2018-06-28), pages 309 - 315, XP036551860, ISSN: 0932-383X, [retrieved on 20180628], DOI: 10.1007/S00502-018-0624-5 *

Similar Documents

Publication Publication Date Title
US10496898B2 (en) State detection using machine-learning model trained on simulated image data
CN109255830B (zh) 三维人脸重建方法和装置
US10007866B2 (en) Neural network image classifier
CN111298445B (zh) 目标账号检测方法、装置、电子设备及存储介质
US20220172518A1 (en) Image recognition method and apparatus, computer-readable storage medium, and electronic device
US10587776B2 (en) Electronic device and method for controlling the electronic device
CN111340013B (zh) 人脸识别方法、装置、计算机设备及存储介质
US20210073563A1 (en) Depth-based object re-identification
US11094074B2 (en) Identification of transparent objects from image discrepancies
CN110462645A (zh) 具有更新能力的传感器数据处理器
CN112734873B (zh) 对抗生成网络的图像属性编辑方法、装置、设备及介质
US11763135B2 (en) Concept-based adversarial generation method with steerable and diverse semantics
CN110728319B (zh) 一种图像生成方法、装置以及计算机存储介质
CN111539903B (zh) 训练人脸图像合成模型的方法和装置
CN108509904A (zh) 用于生成信息的方法和装置
US20230419722A1 (en) Simulated capacitance measurements for facial expression recognition training
CN114140841A (zh) 点云数据的处理方法、神经网络的训练方法以及相关设备
US11720168B1 (en) Inferred body movement using wearable RF antennas
WO2023249776A1 (fr) Mesures de capacité simulées pour un apprentissage de reconnaissance d&#39;expression faciale
CN117011449A (zh) 三维面部模型的重构方法和装置、存储介质及电子设备
Gang et al. Skeleton-based action recognition with low-level features of adaptive graph convolutional networks
CN111821688A (zh) 虚拟现实游戏画面处理方法及相关设备
US20230282031A1 (en) Pose prediction for articulated object
EP4246459A1 (fr) Génération de données d&#39;entraînement et/ou de test d&#39;un système de reconnaissance faciale pour une fiabilité améliorée
US20240193870A1 (en) Content creation platform for xr devices

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23731842

Country of ref document: EP

Kind code of ref document: A1