WO2025007739A1 - 一种数据处理方法、装置、电子设备、计算机可读介质 - Google Patents

一种数据处理方法、装置、电子设备、计算机可读介质 Download PDF

Info

Publication number
WO2025007739A1
WO2025007739A1 PCT/CN2024/099890 CN2024099890W WO2025007739A1 WO 2025007739 A1 WO2025007739 A1 WO 2025007739A1 CN 2024099890 W CN2024099890 W CN 2024099890W WO 2025007739 A1 WO2025007739 A1 WO 2025007739A1
Authority
WO
WIPO (PCT)
Prior art keywords
key point
trajectory
image
sequence
prediction
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2024/099890
Other languages
English (en)
French (fr)
Inventor
陈光启
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Zitiao Network Technology Co Ltd
Original Assignee
Beijing Zitiao Network Technology Co Ltd
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
Application filed by Beijing Zitiao Network Technology Co Ltd filed Critical Beijing Zitiao Network Technology Co Ltd
Publication of WO2025007739A1 publication Critical patent/WO2025007739A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/761—Proximity, similarity or dissimilarity measures

Definitions

  • Embodiments of the present disclosure relate to a data processing method, device, electronic device, and computer-readable medium.
  • Key point prediction is a computer vision task, and key point prediction is used to locate certain key points (such as the key point of the corner of the eye, etc.) in an image data or an image sequence (such as video data, etc.).
  • the present disclosure provides a data processing method, device, electronic device, and computer-readable medium.
  • the present disclosure provides a data processing method, the method comprising:
  • the key point prediction sequence includes a prediction trajectory of at least one key point
  • the key point label sequence includes a label trajectory of the at least one key point
  • the trajectory similarity characterization data is used to represent the difference between the predicted trajectory of the key point and the label trajectory of the key point in terms of trajectory length and/or trajectory direction. degree of similarity;
  • the key point prediction performance corresponding to the key point prediction sequence is determined according to the trajectory similarity characterization data.
  • the image sequence includes a first image and a second image, a time sequence corresponding to the first image in the image sequence is earlier than a time sequence corresponding to the second image in the image sequence, and a time sequence corresponding to the first image in the image sequence is adjacent to a time sequence corresponding to the second image in the image sequence;
  • the predicted trajectory of the key point includes the predicted position corresponding to the key point in the first image and the predicted position corresponding to the key point in the second image;
  • the label track of the key point includes the label position corresponding to the key point in the first image and the label position corresponding to the key point in the second image;
  • the trajectory similarity characterization data is determined according to the length similarity characterization data.
  • the image sequence includes a first image and a second image, a time sequence corresponding to the first image in the image sequence is earlier than a time sequence corresponding to the second image in the image sequence, and a time sequence corresponding to the first image in the image sequence is adjacent to a time sequence corresponding to the second image in the image sequence;
  • the predicted trajectory of the key point includes the predicted position corresponding to the key point in the first image and the predicted position corresponding to the key point in the second image;
  • the label track of the key point includes the label position corresponding to the key point in the first image and the label position corresponding to the key point in the second image;
  • the trajectory similarity characterization data is determined according to the direction similarity characterization data.
  • the trajectory similarity representation data includes length similarity representation data and/or direction similarity representation data; the length similarity representation data is used to indicate the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in trajectory length; the direction similarity representation data is used to indicate the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in trajectory direction.
  • determining the key point prediction performance corresponding to the key point prediction sequence according to the trajectory similarity representation data includes:
  • the key point prediction performance corresponding to the key point prediction sequence is determined according to the prediction stability characterization data corresponding to the key point prediction sequence.
  • the method before determining the key point prediction performance corresponding to the key point prediction sequence according to the prediction stability characterization data corresponding to the key point prediction sequence, the method further includes:
  • the determining, according to the prediction stability characterization data corresponding to the key point prediction sequence, the key point prediction performance corresponding to the key point prediction sequence comprises:
  • the key point prediction performance corresponding to the key point prediction sequence is determined according to the prediction stability characterization data corresponding to the key point prediction sequence and the prediction accuracy characterization data corresponding to the key point prediction sequence.
  • the step of acquiring a key point prediction sequence of an image sequence includes:
  • the key point prediction model is used to perform key point prediction processing on the image sequence to obtain the image sequence.
  • the key point prediction sequence of the image sequence
  • the method further includes:
  • the key point prediction model is updated according to the key point prediction performance corresponding to the key point prediction sequence.
  • the present disclosure provides a data processing device, comprising:
  • a data acquisition unit is configured to acquire a key point prediction sequence of an image sequence and a key point label sequence corresponding to the image sequence;
  • the key point prediction sequence includes a prediction trajectory of at least one key point;
  • the key point label sequence includes a label trajectory of the at least one key point;
  • a first determining unit is configured to determine, for any of the key points, trajectory similarity characterization data between the predicted trajectory of the key point and the label trajectory of the key point; the trajectory similarity characterization data is used to indicate the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in terms of trajectory length and/or trajectory direction;
  • the second determining unit is configured to determine the key point prediction performance corresponding to the key point prediction sequence according to the trajectory similarity representation data.
  • the present disclosure provides an electronic device, the device comprising: a processor and a memory;
  • the memory is configured to store instructions or computer programs
  • the processor is configured to execute the instructions or computer programs in the memory so that the electronic device performs the data processing method provided by the present disclosure.
  • the present disclosure provides a computer-readable medium, in which instructions or computer programs are stored.
  • the instructions or computer programs are executed on a device, the device executes the data processing method provided by the present disclosure.
  • the present disclosure provides a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, wherein the computer program contains program codes for executing the data processing method provided by the present disclosure.
  • FIG1 is a flow chart of a data processing method provided by an embodiment of the present disclosure.
  • FIG2 is a schematic diagram of a predicted trajectory and a label trajectory having a corresponding relationship provided by an embodiment of the present disclosure
  • FIG3 is a schematic diagram of the structure of a data processing device provided by an embodiment of the present disclosure.
  • FIG. 4 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure.
  • the predicted position of the key point will usually be distributed around the label position of the key point, and if the jitter is weaker (for example, there is almost no jitter), the predicted position of the key point is closer to the label position of the key point, so that the trajectory formed by the predicted position corresponding to the key point in multiple image data and the trajectory formed by the label position corresponding to the key point in multiple image data are more similar (for example, the trajectory lengths are closer, the trajectory directions are more similar, etc.); if the jitter is stronger (that is, more unstable), the predicted position of the key point deviates from the label position of the key point, so that the difference between the trajectory formed by the predicted position corresponding to the key point in multiple image data and the trajectory formed by the label position corresponding to the key point in multiple image data is greater (for example, the trajectory length difference is larger, the trajectory direction is more dissimilar, etc.).
  • the present disclosure provides a data processing method, which includes: first obtaining a key point prediction sequence of an image sequence and a key point label sequence corresponding to the image sequence, so that the key point prediction sequence includes a prediction trajectory of at least one key point, and the key point label sequence includes a label trajectory of at least one key point; then, for any key point, determining the trajectory similarity characterization data between the prediction trajectory of the key point and the label trajectory of the key point, so that the trajectory similarity characterization data can represent the degree of similarity between the prediction trajectory of the key point and the label trajectory of the key point in trajectory length and/or trajectory direction, so that the trajectory similarity characterization data can better represent the degree of jitter of the prediction trajectory of the key point relative to the label trajectory of the key point; finally, based on the trajectory similarity characterization data corresponding to all key points, determining the key point prediction performance corresponding to the key point prediction
  • the present disclosure does not limit the execution subject of the above data processing method.
  • the data processing method provided in the embodiment of the present disclosure can be applied to a device with data processing function such as a terminal device or a server.
  • the data processing method provided in the embodiment of the present disclosure can also be implemented with the help of a data communication process between different devices (for example, a terminal device and a server, two terminal devices, or two servers).
  • the terminal device can be a smart phone, a computer, a personal digital assistant (PDA) or a tablet computer.
  • PDA personal digital assistant
  • the server can be an independent server, a cluster server or a cloud server.
  • Figure 1 is a flow chart of a data processing method provided by the embodiment of the present disclosure.
  • S1 Obtain a key point prediction sequence of an image sequence and a key point label sequence corresponding to the image sequence; the key point prediction sequence includes a prediction trajectory of at least one key point; the key point label sequence includes a label trajectory of at least one key point.
  • the image sequence is used to record a plurality of image data arranged in a certain order (eg, time sequence); and the present disclosure does not limit the implementation method of the image sequence, for example, it can be implemented using video data.
  • the key point prediction sequence of an image sequence is used to describe the change of the key point prediction position in the image sequence; and the present disclosure does not limit the key point prediction sequence.
  • the key point prediction sequence of the image sequence may include the key point prediction sequence of the first image data.
  • the key point prediction result of the nth image data is used to describe the predicted distribution of the key points in the nth image data; and the present disclosure does not limit the key point prediction result of the nth image data.
  • the key point prediction result of the nth image data may include predicted positions of multiple key points, and the predicted position of the key point is used to describe the predicted position of the key point in the nth image data.
  • n is a positive integer
  • N is a positive integer
  • N represents the number of image data in the image sequence.
  • the key point prediction sequence can not only represent the key point distribution according to the image data dimension (for example, the key point prediction sequence includes the key point prediction results of each image data in the image sequence), but also represent the key point distribution according to the key point dimension (for example, the key point prediction sequence includes the predicted position changes of each key point in the image sequence).
  • the key point prediction sequence of the image sequence may include the prediction trajectory of the at least one key point, so that the key point prediction sequence can represent the predicted position change of each key point in the image sequence.
  • the prediction trajectory of the key point is used to describe the predicted position change of the key point in the image sequence; and the present disclosure does not limit the prediction trajectory of the key point, for example, when the image sequence is the sequence of ⁇ 1st image data, 2nd image data, ..., Nth image data ⁇ , the prediction trajectory of the key point may include the predicted position corresponding to the key point under the 1st image data (for example, x 0 shown in FIG. 2 ), the predicted position corresponding to the key point under the 2nd image data (for example, x 1 shown in FIG. 2 ), the predicted position corresponding to the key point under the 3rd image data (for example, x 2 shown in FIG. 2 ), ...
  • n is a positive integer
  • n ⁇ N N is a positive integer
  • N represents the number of image data in the image sequence.
  • x 0 ⁇ x 1 ⁇ x 2 ⁇ x 3 represents a predicted trajectory of a key point
  • x 0 represents the predicted position corresponding to the key point under the first image data
  • x 1 represents the predicted position corresponding to the key point under the second image data
  • x 2 represents the predicted position corresponding to the key point under the third image data
  • x 3 represents the predicted position corresponding to the key point under the fourth image data
  • g 0 ⁇ g 1 ⁇ g 2 ⁇ g 3 represents the label trajectory of the key point
  • g 0 represents the label position corresponding to the key point in the first image data
  • g 1 represents the label position corresponding to the key point in the second image data
  • g 2 represents the label position corresponding to the key point in the third image data
  • g 3 represents the label position corresponding to the key point in the fourth image data.
  • the present disclosure does not limit the method of representing the position (for example, the "label position” or the “predicted position” above).
  • it can be implemented using any existing or future position representation method (for example, coordinates, etc.).
  • the present disclosure does not limit the method for obtaining the above-mentioned key point prediction sequence.
  • it can be implemented by any existing or future method that can perform key point prediction processing on an image sequence (for example, with the help of a pre-built machine learning model with key point prediction function, etc.).
  • the process of acquiring the above key point prediction sequence can be specifically as follows: after acquiring the image sequence, performing key point prediction processing on the image sequence to obtain the key point prediction sequence of the image sequence.
  • the process of acquiring the above key point prediction sequence can be specifically as follows: after acquiring the image sequence, performing key point prediction processing on the image sequence using a key point prediction model to obtain the key point prediction sequence of the image sequence.
  • the key point prediction model is used to perform key point prediction processing on the input data of the key point prediction model; and the present disclosure does not limit the key point prediction model.
  • the key point label sequence corresponding to the image sequence is used to describe the actual position change of the key points in the image sequence; and the present disclosure does not limit the key point label sequence.
  • the key point label sequence corresponding to the image sequence may include the key point label result of the 1st image data, the key point label result of the 2nd image data, ... (and so on), and the key point label result of the Nth image data.
  • the key point label result of the nth image data is used to describe the actual distribution of the key points in the nth image data; and the present disclosure does not limit the key point label result of the nth image data.
  • the key point label result of the nth image data may include the label positions of multiple key points, and the label position of the key point is used to describe the actual position of the key point in the nth image data.
  • n is a positive integer
  • n ⁇ N is a positive integer
  • N represents the number of image data in the image sequence.
  • the key point label sequence corresponding to the image sequence when the above image sequence is used to describe at least one key, the key point label sequence corresponding to the image sequence may include at least one key point label track.
  • the label track of the key point is used to describe the actual position change of the key point in the above image sequence; and the present disclosure does not limit the label track of the key point.
  • the image sequence is a sequence of ⁇ the first image data, the second image data, ..., the Nth image data ⁇
  • the label track of the key point may include the label position corresponding to the key point under the first image data (for example, g 0 shown in FIG.
  • the label position corresponding to the key point under the nth image data is used to describe the actual position of the key point in the nth image data.
  • n is a positive integer
  • n ⁇ N N is a positive integer
  • N represents the number of image data in the image sequence.
  • the present disclosure does not limit the method for determining the key point label sequence corresponding to the above image sequence.
  • it can be implemented by manual annotation.
  • the key point prediction sequence of the image sequence and the key point label sequence corresponding to the image sequence are obtained, so that the key point prediction sequence can describe the predicted position changes of the key points in the image sequence (for example, the predicted trajectory of some key points, etc.), and the key point label sequence is used to describe the actual position changes of the key points in the above image sequence (for example, the label trajectory of some key points, etc.), so that the performance (for example, stability, etc.) presented when performing key point prediction processing on the image sequence can be determined based on the differences between the key point prediction sequence and the key point label sequence in certain aspects (for example, trajectory length, etc.).
  • trajectory similarity characterization data For any key point, determine the trajectory similarity characterization data between the predicted trajectory of the key point and the label trajectory of the key point, where the trajectory similarity characterization data is used to represent the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in terms of trajectory length and/or trajectory direction.
  • the trajectory similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point is used to indicate the degree of similarity between the predicted trajectory of the mth key point and the label trajectory of the mth key point, and in particular, can indicate the degree of similarity between the predicted trajectory of the mth key point and the label trajectory of the mth key point in terms of trajectory length and/or trajectory direction.
  • m is a positive integer, m ⁇ the number of key points in the above “at least one key point” (for example, the above image The number of all key points in the sequence, etc.).
  • the present disclosure does not limit the determination process of the above trajectory similarity characterization data. For ease of understanding, some situations are described below.
  • the present disclosure provides a possible implementation method of the process of determining the above trajectory similarity representation data.
  • the following is an explanation taking the process of determining the above "trajectory similarity representation data between the predicted trajectory of the mth key point and the label trajectory of the mth key point" as an example.
  • Step 11 Calculate the distance between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image, and obtain the predicted length of the trajectory segment of the mth key point between the second image and the first image.
  • the first image is used to represent the image data at the front position in any pair of adjacent image data in the above image sequence; the second image is used to represent the image data at the back position in the pair of adjacent image data.
  • the above “predicted position corresponding to the mth key point in the first image” refers to the position information predicted for the mth key point using the first image, so that the “predicted position corresponding to the mth key point in the first image” can represent the predicted position of the mth key point in the first image.
  • predicted position corresponding to the mth key point in the second image refers to the position information predicted for the mth key point using the second image, so that the “predicted position corresponding to the mth key point in the second image” can represent the predicted position of the mth key point in the second image.
  • the distance between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image is used to describe the trajectory length between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image; and the embodiments of the present disclosure do not limit the calculation process of the “distance between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image”, for example, it can be implemented using any existing or future method that can calculate the distance between two points.
  • the above “predicted length of the trajectory segment of the m-th key point between the second image and the first image” is used to describe the predicted trajectory length of the m-th key point between the second image and the first image.
  • the predicted trajectory of the mth key point above may include the predicted position corresponding to the mth key point under the first image data, the predicted position corresponding to the mth key point under the second image data, ...
  • the predicted position corresponding to the mth key point under the Nth image data, and the distance between the predicted position corresponding to the mth key point under the second image data and the predicted position corresponding to the mth key point under the first image data can be determined as the predicted length of the trajectory segment of the mth key point between the second image data and the first image data;
  • the distance between the predicted position corresponding to the mth key point under the third image data and the predicted position corresponding to the mth key point under the second image data is determined as the predicted length of the trajectory segment of the mth key point between the third image data and the second image data; ...
  • the distance between the predicted position corresponding to the mth key point under the Nth image data and the predicted position corresponding to the mth key point under the N-1th image data is determined as the predicted length of the trajectory segment of the mth key point between the Nth image data and the N-1th image data, so that the predicted length of the trajectory segment presented by the mth key point in any pair of adjacent image data in the image sequence can be obtained.
  • Step 12 Calculate the distance between the label position corresponding to the mth key point in the second image and the label position corresponding to the mth key point in the first image, and obtain the label length of the trajectory segment of the mth key point between the second image and the first image.
  • label position corresponding to the mth key point in the first image refers to the position information pre-marked for the mth key point in the first image, so that the “label position corresponding to the mth key point in the first image” can indicate the actual position of the mth key point in the first image.
  • the label position corresponding to the mth key point in the second image refers to the position information pre-marked for the mth key point in the second image, so that the “label position corresponding to the mth key point in the second image” can indicate the actual position of the mth key point in the second image.
  • the distance between the label position corresponding to the mth key point in the second image and the label position corresponding to the mth key point in the first image is used to describe the trajectory length between the label position corresponding to the mth key point in the second image and the label position corresponding to the mth key point in the first image; and the embodiments of the present disclosure do not limit the calculation process of the “distance between the label position corresponding to the mth key point in the second image and the label position corresponding to the mth key point in the first image”.
  • the implementation method of the calculation process is similar to the implementation method of the above “the distance between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image”. For the sake of brevity, it will not be repeated here.
  • the length of the trajectory segment label of the m-th key point between the second image and the first image is used to describe the actual trajectory length of the m-th key point between the second image and the first image.
  • the label trajectory of the mth key point above may include the label position corresponding to the mth key point under the 1st image data, the label position corresponding to the mth key point under the 2nd image data, ...
  • the label position corresponding to the mth key point under the Nth image data, and the distance between the label position corresponding to the mth key point under the 2nd image data and the label position corresponding to the mth key point under the 1st image data can be determined as the trajectory segment label length of the mth key point between the 2nd image data and the 1st image data; the distance between the label position corresponding to the mth key point under the 3rd image data and the label position corresponding to the mth key point under the 4th ...
  • the distance between the label position corresponding to the mth key point under the first image data and the label position corresponding to the mth key point under the second image data is determined as the label length of the trajectory segment of the mth key point between the third image data and the second image data; ... (and so on); the distance between the label position corresponding to the mth key point under the Nth image data and the label position corresponding to the mth key point under the N-1th image data is determined as the label length of the trajectory segment of the mth key point between the Nth image data and the N-1th image data, so that the label length (that is, the actual length) of the trajectory segment presented by the mth key point in any pair of adjacent image data in the image sequence can be obtained.
  • Step 13 Determine the length similarity representation data between the predicted trajectory of the m-th key point and the labeled trajectory of the m-th key point based on the difference between the predicted length of the corresponding trajectory segment and the label length of the trajectory segment, so that the length similarity representation data can represent the degree of similarity between the predicted trajectory of the m-th key point and the labeled trajectory of the m-th key point in terms of trajectory length.
  • the “corresponding predicted length of trajectory segment and trajectory segment label length” are used to describe the predicted length and actual length of the trajectory segment presented by the same key point in the same pair of adjacent image data.
  • the “corresponding predicted length of trajectory segment and trajectory segment label length” may include: the above “the predicted length of trajectory segment of the mth key point between the n+1th image data and the nth image data” and the above “the trajectory segment label length of the mth key point between the n+1th image data and the nth image data”; wherein n is a positive integer, n+1 ⁇ N.
  • the corresponding relationship is predetermined, for example, the corresponding relationship may specifically include: the corresponding relationship between the above “the trajectory segment prediction length of the mth key point between the n+1th image data and the nth image data” and the above “the trajectory segment label length of the mth key point between the n+1th image data and the nth image data”.
  • n is a positive integer, n+1 ⁇ N.
  • the present disclosure does not limit the implementation of the above step 13.
  • it can be implemented using the following formula (1).
  • E represents the length similarity characterization data between the predicted trajectory of a key point and the label trajectory of the key point;
  • xi represents the predicted position corresponding to the key point under the i+1th image data;
  • xi -1 represents the predicted position corresponding to the key point under the i-th image data;
  • 2 represents the predicted length of the trajectory segment of the key point between the i+1th image data and the i-th image data;
  • gi represents the label position corresponding to the key point under the i+1th image data;
  • 2 represents the predicted length of the trajectory segment of the key point between the i+1th image data and the i-th image data;
  • N represents the number of image data in the above image sequence.
  • Step 14 Determine the trajectory similarity representation data between the predicted trajectory of the mth key point and the label trajectory of the mth key point based on the length similarity representation data between the predicted trajectory of the mth key point and the label trajectory of the mth key point.
  • the present disclosure is not limited to the implementation method of the above step 14.
  • it can be specifically: directly determine the length similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point as the trajectory similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, so that the trajectory similarity characterization data can at least represent the degree of similarity between the predicted trajectory of the mth key point and the label trajectory of the mth key point in terms of trajectory length.
  • the process of determining the trajectory similarity characterization data between the predicted trajectory of the key point and the label trajectory of the key point can be specifically as follows: first, calculate the distance between the predicted position corresponding to the key point in the second image and the predicted position corresponding to the key point in the first image to obtain the predicted length of the trajectory segment, and calculate the distance between the label position corresponding to the key point in the second image and the label position corresponding to the key point in the first image to obtain the label length of the trajectory segment; then, according to The length similar
  • the present disclosure provides a possible implementation method of the process of determining the above trajectory similarity characterization data.
  • the following is an explanation taking the process of determining the above "trajectory similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point" as an example.
  • Step 21 Determine the predicted direction representation data of the trajectory segment of the mth key point between the second image and the first image based on the direction representation data between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image.
  • the above “directional characterization data between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image” is used to describe the trajectory direction between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image.
  • the present disclosure does not limit the implementation of the above “the direction representation data between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image”.
  • it can be implemented by any existing or future method that can represent the direction between two positions (such as a position vector, a unit vector, a straight line slope, or a straight line inclination angle, etc.).
  • the following five examples are used for explanation.
  • the above step 21 can be specifically as follows: subtract the predicted position corresponding to the mth key point in the first image from the predicted position corresponding to the mth key point in the second image to obtain a position vector, and regard the position vector as the above "the directional representation data between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image".
  • the above step 21 can be specifically as follows: first, subtract the predicted position corresponding to the mth key point in the first image from the predicted position corresponding to the mth key point in the second image to obtain a position vector; then, using the position vector, determine a unit vector so that the unit vector can describe the direction described by the position vector; and then regard the unit vector as the above "the direction representation data between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image".
  • the above step 21 can be specifically as follows: perform a straight line fitting on the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image to obtain the slope of the straight line, and regard the slope of the straight line as the above "the directional characterization data between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image".
  • the above step 21 can be specifically as follows: first, perform a straight line fitting on the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image to obtain the slope of the straight line; then convert the slope of the straight line into an inclination angle of the straight line, and regard the inclination angle of the straight line as the above "the direction characterization data between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image".
  • the above step 21 can be specifically as follows: based on the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image, calculate the straight line inclination angle to obtain the straight line inclination angle, and regard the straight line inclination angle as the above "the direction characterization data between the predicted position corresponding to the mth key point in the second image and the predicted position corresponding to the mth key point in the first image".
  • the predicted direction representation data of the trajectory segment of the mth key point between the second image and the first image is used to describe the predicted direction representation data of the trajectory segment of the mth key point between the second image and the first image.
  • the direction of the trajectory is used to describe the predicted direction representation data of the trajectory segment of the mth key point between the second image and the first image.
  • the predicted trajectory of the mth key point above may include the predicted position corresponding to the mth key point under the 1st image data, the predicted position corresponding to the mth key point under the 2nd image data, ...
  • the predicted position corresponding to the mth key point under the Nth image data, and the direction representation data between the predicted position corresponding to the mth key point under the 2nd image data and the predicted position corresponding to the mth key point under the 1st image data can be determined as the position of the mth key point between the 2nd image data and the 1st image data.
  • the predicted direction of the trajectory segment of the m-th key point is determined by the direction representation data between the predicted position corresponding to the m-th key point under the third image data and the predicted position corresponding to the m-th key point under the second image data as the predicted direction of the trajectory segment of the m-th key point between the third image data and the second image data; ...
  • the predicted direction representation data between the predicted position corresponding to the m-th key point under the N-th image data and the predicted position corresponding to the m-th key point under the N-1-th image data are determined as the predicted direction of the trajectory segment of the m-th key point between the N-th image data and the N-1-th image data, so that the predicted direction of the trajectory segment presented by the m-th key point in any pair of adjacent image data in the image sequence can be obtained.
  • Step 22 Determine the label direction representation data of the trajectory segment of the mth key point between the second image and the first image based on the direction representation data between the label position corresponding to the mth key point in the second image and the label position corresponding to the mth key point in the first image.
  • the above “directional representation data between the label position corresponding to the mth key point in the second image and the label position corresponding to the mth key point in the first image” is used to describe the trajectory direction between the label position corresponding to the mth key point in the second image and the label position corresponding to the mth key point in the first image.
  • the implementation method of the above is similar to the implementation method of the above "the directional representation data between the predicted position corresponding to the m-th key point in the second image and the predicted position corresponding to the m-th key point in the first image", and for the sake of brevity, it will not be repeated here.
  • the above “the direction of the trajectory segment label of the mth key point between the second image and the first image” The “characterization data” is used to describe the actual trajectory direction of the m-th key point between the second image and the first image.
  • the label trajectory of the mth key point above may include the label position corresponding to the mth key point under the 1st image data, the label position corresponding to the mth key point under the 2nd image data, ...
  • the label position corresponding to the mth key point under the Nth image data, and the direction representation data between the label position corresponding to the mth key point under the 2nd image data and the label position corresponding to the mth key point under the 1st image data can be determined as the trajectory segment label of the mth key point between the 2nd image data and the 1st image data.
  • label direction determine the direction representation data between the label position corresponding to the m-th key point under the third image data and the label position corresponding to the m-th key point under the second image data as the label direction of the trajectory segment of the m-th key point between the third image data and the second image data; ...
  • Step 23 Determine the directional similarity characterization data between the predicted trajectory of the m-th key point and the labeled trajectory of the m-th key point based on the predicted directional characterization data of the above trajectory segment and the labeled directional characterization data of the above trajectory segment, so that the directional similarity characterization data can represent the degree of similarity between the predicted trajectory of the m-th key point and the labeled trajectory of the m-th key point in the trajectory direction.
  • step 23 does not limit the implementation method of step 23.
  • it can be specifically as follows: first, based on the predicted direction of the trajectory segment of the mth key point between the nth image data and the n+1th image data, and the label direction of the trajectory segment of the mth key point between the nth image data and the n+1th image data, determine the direction difference representation data of the mth key point between the nth image data and the n+1th image data, so that the direction difference representation data can represent the direction deviation of the predicted direction of the trajectory segment relative to the label direction of the trajectory segment, n is a positive integer, n+1 ⁇ N; then calculate the average value of all the direction difference representation data corresponding to the mth key point as the direction similarity table between the predicted trajectory of the mth key point and the label trajectory of the mth key point. Collect data.
  • the present disclosure does not limit the calculation method of the "direction difference characterization data" in the above paragraph.
  • the "direction difference characterization data" can be determined with the help of the inner product calculation formula.
  • the "direction difference characterization data” can be directly implemented with the help of a difference calculation formula or a calculation formula for the angle between the slopes of two straight lines.
  • the "direction difference characterization data” can be directly implemented with the help of a difference calculation formula.
  • Step 24 Determine trajectory similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point based on the directional similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point.
  • the present disclosure does not limit the implementation method of the above step 24.
  • it can specifically be: directly determine the directional similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point as the trajectory similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, so that the trajectory similarity characterization data can at least express the degree of similarity between the predicted trajectory of the mth key point and the label trajectory of the mth key point in the trajectory direction.
  • the process of determining the trajectory similarity representation data between the predicted trajectory of the key point and the label trajectory of the key point can be specifically as follows: first, based on the direction representation data between the predicted position corresponding to the key point in the second image and the predicted position corresponding to the key point in the first image, determine the trajectory segment prediction direction representation data, and based on the direction representation data between the label position corresponding to the key point in the second image and the label position corresponding to the key point in the first image, determine the trajectory segment label direction representation data;
  • Case 3 In some application scenarios, in order to better improve the effect of determining the trajectory similarity, the similarity between the two trajectories can be determined by comparing the trajectory directions and trajectory lengths of the two trajectories.
  • the present disclosure provides a possible implementation of the process of determining the above trajectory similarity characterization data, which may specifically include the following steps 31 and 32.
  • Step 31 Determine, based on the predicted trajectory of the mth key point and the label trajectory of the mth key point, length similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, and direction similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point.
  • m is a positive integer, m ⁇ the number of key points in the above “at least one key point”.
  • Step 32 Determine the trajectory similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point based on the length similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, and the direction similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, so that the trajectory similarity characterization data can represent the degree of similarity between the predicted trajectory of the mth key point and the label trajectory of the mth key point in terms of trajectory length and trajectory direction.
  • step 32 may specifically be: combining the length similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, and the direction similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, to obtain the trajectory similarity characterization data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, so that the trajectory similarity characterization data includes the length similarity characterization data and the direction similarity characterization data.
  • the above step 32 can be specifically as follows: weighted summation of the length similarity representation data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, and the direction similarity representation data between the predicted trajectory of the mth key point and the label trajectory of the mth key point, to obtain the trajectory similarity representation data between the predicted trajectory of the mth key point and the label trajectory of the mth key point.
  • the length similarity characterization data and the direction similarity characterization data can be first determined based on the predicted trajectory of the key point and the label trajectory of the key point; and then, based on the length similarity characterization data and the direction similarity characterization data, the trajectory similarity characterization data between the predicted trajectory of the key point and the label trajectory of the key point can be determined, so that the trajectory similarity characterization data can at least represent the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in terms of trajectory length and trajectory direction.
  • the trajectory similarity representation data between the predicted trajectory of the key point and the label trajectory of the key point includes length similarity representation data and/or direction similarity representation data;
  • the length similarity representation data is used to indicate the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in trajectory length;
  • the direction similarity representation data is used to indicate the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in trajectory direction.
  • the trajectory similarity characterization data between the predicted trajectory of the key point and the label trajectory of the key point can be calculated, so that the trajectory similarity characterization data can represent the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in terms of trajectory length and/or trajectory direction, so that the performance (for example, stability, etc.) presented when performing key point prediction processing on the image sequence can be determined based on the trajectory similarity characterization data.
  • the key point prediction performance corresponding to the key point prediction sequence is used to describe the performance (eg, stability, etc.) presented when the key point prediction sequence is obtained by predicting the above image sequence.
  • the present disclosure does not limit the implementation method of S3 above.
  • it can specifically be: directly performing some statistical analysis (such as average value, maximum value, median value, etc.) on the trajectory similarity representation data corresponding to all key points to obtain the key point prediction performance corresponding to the key point prediction sequence above.
  • the present disclosure also provides a possible implementation of S3 above, Specifically, it may include the following steps 41 and 42.
  • Step 41 Determine the prediction stability characterization data corresponding to the above key point prediction sequence based on the above trajectory similarity characterization data, so that the prediction stability characterization data is used to represent the prediction stability presented when the key point prediction sequence is predicted for the above image sequence.
  • the present disclosure does not limit the implementation method of the above step 41.
  • the step 41 can specifically be: collecting the trajectory similarity characterization data corresponding to all key points to obtain the predicted stability characterization data corresponding to the above key point prediction sequence, so that the predicted stability characterization data includes the trajectory similarity characterization data corresponding to all key points.
  • the above step 41 can also be: directly determine the average value of the trajectory similarity characterization data corresponding to all key points as the prediction stability characterization data corresponding to the above key point prediction sequence, so that the prediction stability characterization data is used to represent the stability presented when the key point prediction sequence is predicted for the above image sequence.
  • Step 42 Determine the key point prediction performance corresponding to the key point prediction sequence according to the prediction stability characterization data corresponding to the key point prediction sequence.
  • the present disclosure does not limit the implementation method of the above step 42.
  • the step 42 can specifically be: directly determining the predicted stability characterization data as the key point prediction performance corresponding to the key point prediction sequence.
  • the above step 42 can specifically be: determining the average value between all elements in the data set as the predicted stability characterization data corresponding to the key point prediction sequence.
  • the present disclosure also provides a possible implementation of the key point prediction performance determination process corresponding to the key point prediction sequence above (that is, step 42 above), which can specifically include the following steps 421-422.
  • Step 421 According to the key point prediction sequence of the above image sequence and the key point label sequence corresponding to the image sequence, determine the prediction accuracy characterization data corresponding to the key point prediction sequence, so that the prediction The accuracy characterization data is used to represent the prediction accuracy presented when the key point prediction sequence is predicted for the image sequence.
  • the present disclosure does not limit the implementation of the above step 421.
  • it can be implemented by any existing or future method that can calculate the prediction accuracy with the help of prediction information and standard information (for example, a method of calculating the similarity between the prediction information and the standard information).
  • a method of calculating the similarity between the prediction information and the standard information for example, two examples are provided below for explanation.
  • the key point prediction sequence of the image sequence includes the prediction position corresponding to the mth key point under the 1st image data, the prediction position corresponding to the mth key point under the 2nd image data, ... (and so on), and the prediction position corresponding to the mth key point under the Nth image data, m is a positive integer, m ⁇ the number of key points in the above "at least one key point”;
  • the key point label sequence corresponding to the image sequence includes the label position corresponding to the mth key point under the 1st image data, the label position corresponding to the mth key point under the 2nd image data, ... (and so on), and the label position corresponding to the mth key point under the Nth image data, m is a positive integer, m ⁇ the number of key points in the above "at least one key point”.
  • the above step 421 can be specifically as follows: first, calculate the distance between the predicted position corresponding to the mth key point in the nth image data and the label position corresponding to the mth key point in the nth image data, and obtain the position prediction loss corresponding to the mth key point in the nth image data, where n is a positive integer, n ⁇ N, and m is a positive integer, m ⁇ the number of key points in the above "at least one key point”; then, the position prediction losses corresponding to all key points in all image data are averaged (or summed) to obtain the prediction accuracy characterization data corresponding to the above key point prediction sequence.
  • the above step 421 can be specifically as follows: first, calculate the distance between the predicted position corresponding to the mth key point in the nth image data and the label position corresponding to the mth key point in the nth image data, and obtain the position prediction loss corresponding to the mth key point in the nth image data, where n is a positive integer, n ⁇ N, and m is a positive integer, m ⁇ the number of key points in the above "at least one key point”; then, collect the position prediction losses corresponding to all key points in all image data to obtain the prediction accuracy characterization data corresponding to the above key point prediction sequence.
  • step 421 does not limit the execution time of step 421 above, but only needs to ensure that the execution time of step 421 is The execution time of step 421 is later than the execution time of S1 above, and the execution time of step 421 is ensured to be earlier than the execution time of step 422 below.
  • Step 422 Determine the key point prediction performance corresponding to the key point prediction sequence according to the prediction stability characterization data corresponding to the above key point prediction sequence and the prediction accuracy characterization data corresponding to the key point prediction sequence.
  • the present disclosure does not limit the implementation method of the above step 422.
  • the step 422 can specifically be: directly performing weighted sum processing on the prediction stability characterization data corresponding to the key point prediction sequence and the prediction accuracy characterization data corresponding to the key point prediction sequence to obtain the key point prediction performance corresponding to the key point prediction sequence.
  • the above step 422 can be specifically: first calculate the average values between all elements in the two data sets respectively; then perform weighted sum processing on the two average values to obtain the key point prediction performance corresponding to the key point prediction sequence.
  • the above step 422 can be specifically as follows: first, the average value between the position prediction loss corresponding to the mth key point under the n+1th image data and the position prediction loss corresponding to the mth key point under the nth image data is determined as the weighted weight corresponding to the trajectory segment prediction information of the mth key point between the nth image data and the n+1th image data (for example, the trajectory segment prediction length and/or the trajectory segment prediction direction), n is a positive integer, n ⁇ N, m is a positive integer, m ⁇ the number of key points in the above "at least one key point"; then, according to the weighted weight, these trajectory segment prediction information are weighted and summed to obtain the key point prediction performance corresponding to the key point prediction sequence.
  • step 421 to 422 it can be known that in some application scenarios, for an image sequence, after obtaining the key point prediction sequence of the image sequence and the key point label sequence corresponding to the image sequence, it is necessary not only to analyze the prediction stability presented when the key point prediction sequence is predicted for the image sequence based on the key point prediction sequence and the key point label sequence, but also to analyze the prediction accuracy presented when the key point prediction sequence is predicted for the image sequence based on the key point prediction sequence and the key point label sequence, so that the key point prediction performance corresponding to the key point prediction sequence can be determined with the help of these two analysis results in the future.
  • the key point prediction performance is enabled to better represent the prediction effect presented when the key point prediction sequence is predicted for the image sequence.
  • the trajectory similarity characterization data between the predicted trajectory of each key point described by the image sequence and the label trajectory of the key point can be determined based on the key point prediction sequence and the key point label sequence; then, based on these trajectory similarity characterization data, the prediction stability characterization data corresponding to the key point prediction sequence is calculated, so that the prediction stability characterization data is used to represent the prediction stability presented when the key point prediction sequence is predicted for the image sequence; finally, based on the prediction stability characterization data (and the prediction accuracy characterization data corresponding to the key point prediction sequence), the key point prediction performance corresponding to the key point prediction sequence is determined, so that the key point prediction performance can represent the state achieved under at least one evaluation indicator (for example, stability + accuracy) when the key point prediction sequence is predicted for the image sequence, so that the key point prediction performance can better represent the prediction effect presented when
  • the key point prediction sequence of the image sequence and the key point label sequence corresponding to the image sequence are first obtained, so that the key point prediction sequence includes the predicted trajectory of at least one key point, and the key point label sequence includes the label trajectory of at least one key point; then, for any key point, the trajectory similarity characterization data between the predicted trajectory of the key point and the label trajectory of the key point is determined, so that the trajectory similarity characterization data can represent the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in terms of trajectory length and/or trajectory direction, so that the trajectory similarity
  • the characterization data can better express the degree of jitter of the predicted trajectory of the key point relative to the label trajectory of the key point; finally, based on the trajectory similarity characterization data corresponding to all key points, the key point prediction performance corresponding to the key point prediction sequence is determined, so that the key point prediction performance can at least better express the degree of
  • the present disclosure does not limit the application scenario of the data processing method.
  • it can be applied to any application scenario (such as model training and other scenarios) that requires evaluation processing of a key point prediction sequence of an image sequence.
  • an application scenario is used as an example for explanation below.
  • the data processing method provided by the present disclosure may include the following steps 51 to 55.
  • Step 51 Acquire an image sequence and a key point label sequence corresponding to the image sequence; the image sequence is used to describe at least one key point, and the key point label sequence includes a label trajectory of the at least one key point.
  • the present disclosure does not limit the method for obtaining the image sequence in step 51.
  • it can be specifically: randomly extracting an image sequence from the training data set as the image sequence required for the current round of training process.
  • the training data set refers to the data set required to be based on when training a machine learning model; and the learning goal of the machine learning model in the training process is: to perform key point prediction processing on an image sequence.
  • the present disclosure does not limit the implementation method of the training data set.
  • the training data set can include multiple image sequences and key point label sequences corresponding to each image sequence.
  • the present disclosure does not limit the method of obtaining the "key point label sequence corresponding to the image sequence" in step 51. For example, it can be specifically: after randomly extracting an image sequence from the training data set, extract the key point label sequence corresponding to the image sequence from the training data set as the key point label sequence required to be used in the current round of training process.
  • Step 52 perform key point prediction processing on the above image sequence using the key point prediction model to obtain a key point prediction sequence of the image sequence; the key point prediction sequence includes a prediction trajectory of at least one key point in the above.
  • the key point prediction model is used to perform key point prediction processing on the input data of the key point prediction model; and the present disclosure does not limit the key point prediction model.
  • it can be implemented using any existing or future machine learning model that requires learning and training in key point prediction.
  • Step 53 For any key point, determine the trajectory similarity characterization data between the predicted trajectory of the key point and the label trajectory of the key point, where the trajectory similarity characterization data is used to represent the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in terms of trajectory length and/or trajectory direction.
  • step 53 can be found in S2 above, and for the sake of brevity, it will not be repeated here.
  • Step 54 Determine the key point prediction performance corresponding to the key point prediction sequence based on the above trajectory similarity representation data.
  • step 54 can be found in S3 above, and for the sake of brevity, it will not be repeated here.
  • Step 55 Update the key point prediction model according to the key point prediction performance corresponding to the key point prediction sequence above, and return to continue to execute step 51 above and its subsequent steps until the preset stop condition is reached.
  • the present disclosure does not limit the implementation method of the updating process in step 55 above.
  • it can specifically be: first use the key point prediction performance corresponding to the key point prediction sequence above to determine the model loss of the key point prediction model; then update the key point prediction model based on the model loss to obtain an updated key point prediction model, so that the updated key point prediction model has better key point prediction performance, so that the above step 51 and its subsequent steps can be continued to be executed based on the updated key point prediction model to realize the next round of training process for the key point prediction model.
  • the preset stop condition mentioned above refers to the condition that needs to be achieved at the end of the training process for the key point prediction model; and the preset stop condition can be set in advance according to the application scenario.
  • the preset stop condition may specifically include: the model loss of the key point prediction model is lower than a preset first threshold.
  • the preset stop condition may specifically include: the rate of change of the model loss of the key point prediction model is lower than a preset second threshold.
  • the preset stop condition may specifically include: the number of updates of the key point prediction model is higher than a preset third threshold.
  • the key point prediction model is updated by using the key point prediction performance corresponding to the key point prediction sequence determined in the current round of training, so that the updated key point prediction model has better key point prediction performance, so that the above step 51 and its subsequent steps can be continued to be performed based on the updated key point prediction model to achieve the next round of training process for the key point prediction model. In this way, the purpose of performing at least one round of training for the key point prediction model can be achieved.
  • the key point prediction model can be used to perform key point prediction processing on the image sequence to obtain the key point prediction sequence of the image sequence; then, based on the key point prediction sequence of the image sequence and the key point label sequence corresponding to the image sequence, the key point prediction performance corresponding to the key point prediction sequence is determined; finally, based on the key point prediction performance, the key point prediction model is updated so that the updated key point prediction model has better key point prediction performance.
  • the key point prediction performance can better represent the prediction effect presented when the key point prediction sequence is predicted for the image sequence, so that the key point prediction performance can better represent the prediction performance of the key point prediction model, so that the key point prediction performance can better guide the update process of the key point prediction model, and then the key point prediction model obtained by training based on the key point prediction performance has better prediction performance (for example, prediction stability + prediction accuracy), which is conducive to improving the model training effect for the key point prediction model.
  • Figure 3 is a schematic diagram of the structure of a data processing device provided by the embodiment of the present disclosure. It should be noted that for the technical details of the data processing device provided by the embodiment of the present disclosure, please refer to the relevant content of the data processing method above.
  • the data processing device 300 provided in the embodiment of the present disclosure includes:
  • a data acquisition unit 301 is used to acquire a key point prediction sequence of an image sequence and a key point label sequence corresponding to the image sequence;
  • the key point prediction sequence includes a prediction trajectory of at least one key point;
  • the key point label sequence includes a label trajectory of the at least one key point;
  • a first determining unit 302 is used to determine, for any of the key points, trajectory similarity characterization data between the predicted trajectory of the key point and the label trajectory of the key point; the trajectory similarity characterization data is used to indicate the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in terms of trajectory length and/or trajectory direction;
  • the second determining unit 303 is used to determine the key point prediction according to the trajectory similarity characterization data. The prediction performance of the key points corresponding to the test sequence.
  • the image sequence includes a first image and a second image, a time sequence corresponding to the first image in the image sequence is earlier than a time sequence corresponding to the second image in the image sequence, and a time sequence corresponding to the first image in the image sequence is adjacent to a time sequence corresponding to the second image in the image sequence;
  • the predicted trajectory of the key point includes the predicted position corresponding to the key point in the first image and the predicted position corresponding to the key point in the second image;
  • the label track of the key point includes the label position corresponding to the key point in the first image and the label position corresponding to the key point in the second image;
  • the first determination unit 302 is specifically used to: calculate the distance between the predicted position corresponding to the key point in the second image and the predicted position corresponding to the key point in the first image to obtain the predicted length of the trajectory segment; calculate the distance between the label position corresponding to the key point in the second image and the label position corresponding to the key point in the first image to obtain the label length of the trajectory segment; determine the length similarity representation data according to the difference between the predicted length of the trajectory segment and the label length of the trajectory segment; determine the trajectory similarity representation data based on the length similarity representation data.
  • the image sequence includes a first image and a second image, a time sequence corresponding to the first image in the image sequence is earlier than a time sequence corresponding to the second image in the image sequence, and a time sequence corresponding to the first image in the image sequence is adjacent to a time sequence corresponding to the second image in the image sequence;
  • the predicted trajectory of the key point includes the predicted position corresponding to the key point in the first image and the predicted position corresponding to the key point in the second image;
  • the label track of the key point includes the label position corresponding to the key point in the first image and the label position corresponding to the key point in the second image;
  • the first determination unit 302 is specifically used to: determine the trajectory segment prediction direction representation data based on the direction representation data between the predicted position corresponding to the key point in the second image and the predicted position corresponding to the key point in the first image; determine the trajectory segment label direction representation data based on the direction representation data between the label position corresponding to the key point in the second image and the label position corresponding to the key point in the first image; determine the direction similarity representation data based on the trajectory segment prediction direction representation data and the trajectory segment label direction representation data; determine the trajectory similarity representation data based on the direction similarity representation data.
  • the trajectory similarity representation data includes length similarity representation data and/or direction similarity representation data; the length similarity representation data is used to indicate the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in trajectory length; the direction similarity representation data is used to indicate the degree of similarity between the predicted trajectory of the key point and the label trajectory of the key point in trajectory direction.
  • the second determining unit 303 includes:
  • a first determination subunit is used to determine the prediction stability characterization data corresponding to the key point prediction sequence according to the trajectory similarity characterization data;
  • the second determining subunit is used to determine the key point prediction performance corresponding to the key point prediction sequence according to the prediction stability characterization data corresponding to the key point prediction sequence.
  • the data processing device 300 further includes:
  • a third determining unit configured to determine prediction accuracy characterization data corresponding to the key point prediction sequence according to the key point prediction sequence and the key point label sequence;
  • the second determining subunit is specifically used to determine the key point prediction performance corresponding to the key point prediction sequence according to the prediction stability characterization data corresponding to the key point prediction sequence and the prediction accuracy characterization data corresponding to the key point prediction sequence.
  • the data acquisition unit 301 is specifically used to: perform key point prediction processing on the image sequence using a key point prediction model to obtain a key point prediction sequence of the image sequence;
  • the data processing device 300 further includes:
  • a model updating unit is used to update the key point prediction model according to the key point prediction performance corresponding to the key point prediction sequence.
  • the key point prediction sequence of the image sequence and the key point label sequence corresponding to the image sequence are first obtained, so that the key point prediction sequence includes the predicted trajectory of at least one key point, and the key point label sequence includes the label trajectory of at least one key point; then, for any key point, the trajectory similarity characterization data between the predicted trajectory of the key point and the label trajectory of the key point is determined, so that the trajectory similarity characterization data can represent the similarity between the predicted trajectory of the key point and the label trajectory of the key point in terms of trajectory length and/or trajectory direction, so that the trajectory similarity characterization data can better represent the predicted trajectory of the key point relative to the key point.
  • the key point prediction performance corresponding to the key point prediction sequence is determined based on the trajectory similarity characterization data corresponding to all key points, so that the key point prediction performance can at least better represent the degree of jitter of the prediction trajectory of each key point relative to its corresponding label trajectory, so that the key point prediction performance can at least better represent the prediction stability presented when the key point prediction sequence is predicted for the image sequence, and further the key point prediction performance can better represent the prediction performance presented when performing key point prediction processing on an image sequence.
  • an embodiment of the present disclosure also provides an electronic device, which includes a processor and a memory: the memory is used to store instructions or computer programs; the processor is used to execute the instructions or computer programs in the memory, so that the electronic device executes any implementation of the data processing method provided by the embodiment of the present disclosure.
  • the terminal device in the embodiment of the present disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc.
  • mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc.
  • PDAs personal digital assistants
  • PADs tablet computers
  • PMPs portable multimedia players
  • vehicle-mounted terminals such as vehicle-mounted navigation terminals
  • fixed terminals such as digital TVs, desktop computers, etc.
  • the electronic device shown in FIG. 4 is only an example and should not bring any limitation to the functions and scope of use
  • the electronic device 400 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage device 408 to a random access memory (RAM) 403.
  • a processing device 401, the ROM 402, and the RAM 403 are connected to each other via a bus 404.
  • An input/output (I/O) interface 405 is also connected to the bus 404.
  • the following devices may be connected to the I/O interface 405: input devices 406 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 407 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 408 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 409.
  • the communication device 409 may allow the electronic device 400 to communicate wirelessly or wired with other devices to exchange data.
  • FIG. 4 shows an electronic device 400 with various devices, it should be understood that it is not required to implement or have all the devices shown. More or fewer devices may be implemented or have alternatively.
  • an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart.
  • the computer program can be downloaded and installed from the network through the communication device 409, or installed from the storage device 408, or installed from the ROM402.
  • the processing device 401 the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.
  • the embodiments of the present disclosure further provide a computer-readable medium, in which instructions or computer programs are stored.
  • the instructions or computer programs are executed on a device, the device executes any implementation of the data processing method provided by the embodiments of the present disclosure.
  • the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above.
  • Computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
  • a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.
  • a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried.
  • This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above.
  • the computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device.
  • the program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.
  • the client and server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network).
  • HTTP Hyper Text Transfer Protocol
  • Examples of communication networks include a local area network ("LAN”), a wide area network ("WAN”), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.
  • the computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.
  • the computer-readable medium carries one or more programs, and when the one or more programs are executed by the electronic device, the electronic device can execute the method.
  • Computer program code for performing the operations of the present disclosure may be written in one or more programming languages or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages.
  • the program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server.
  • the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).
  • LAN local area network
  • WAN wide area network
  • Internet service provider e.g., AT&T, MCI, Sprint, EarthLink, MSN, GTE, etc.
  • each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function.
  • the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.
  • each square box in the block diagram and/or flow chart, and the combination of the square boxes in the block diagram and/or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
  • the units involved in the embodiments of the present disclosure may be implemented by software or by computer.
  • the unit/module name does not limit the unit itself in some cases.
  • exemplary types of hardware logic components include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.
  • FPGAs field programmable gate arrays
  • ASICs application specific integrated circuits
  • ASSPs application specific standard products
  • SOCs systems on chips
  • CPLDs complex programmable logic devices
  • a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment.
  • a machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium.
  • a machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing.
  • a more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
  • RAM random access memory
  • ROM read-only memory
  • EPROM or flash memory erasable programmable read-only memory
  • CD-ROM portable compact disk read-only memory
  • CD-ROM compact disk read-only memory
  • magnetic storage device or any suitable combination of the foregoing.
  • At least one (item) means one or more, and “plurality” means two or more.
  • “And/or” is used to describe the association relationship of associated objects, indicating that three relationships may exist.
  • a and/or B can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural.
  • the character “/” generally indicates that the objects associated before and after are in an “or” relationship.
  • At least one of the following” or similar expressions refers to any combination of these items, including any combination of single or plural items.
  • At least one of a, b or c can mean: a, b, c, "a and b", “a and c", “b and c", or "a and b and c", where a, b, c can be single or multiple.
  • the steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two.
  • the software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本公开的实施例提供一种数据处理方法、装置、电子设备、计算机可读介质，该方法包括：先获取图像序列的关键点预测序列及其对应的关键点标签序列，以使该关键点预测序列包括至少一个关键点的预测轨迹，并使该关键点标签序列包括这些关键点的标签轨迹；然后，对于任一关键点，确定该关键点的预测轨迹与关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据能够表示该关键点的预测轨迹与关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度；最后，依据所有关键点对应的轨迹相似表征数据，确定关键点预测序列对应的关键点预测性能，以使关键点预测性能能够更好地表示出在针对一个图像序列进行关键点预测处理时所呈现的预测性能。

Description

一种数据处理方法、装置、电子设备、计算机可读介质
本申请要求于2023年7月4日递交的中国专利申请第202310816266.2号的优先权，在此全文引用上述中国专利申请公开的内容以作为本申请的一部分。
技术领域
本公开的实施例涉及一种数据处理方法、装置、电子设备、计算机可读介质。
背景技术
关键点预测是一种计算机视觉任务，而且关键点预测用于针对一个图像数据或者一个图像序列(比如，视频数据等)中某些关键点(比如，眼角这一关键点等)进行定位。
对于一些应用场景(比如，训练一个具有关键点预测功能的机器学习模型等场景)来说，在针对一个图像序列(比如，视频数据)进行关键点预测处理得到关键点预测序列之后，可能需要进一步地评估该关键点预测序列在某些方面(比如，稳定性等方面)下所呈现的状态。
发明内容
本公开提供了一种数据处理方法、装置、电子设备、计算机可读介质。
为了实现上述目的，本公开提供的技术方案如下：
本公开提供一种数据处理方法，所述方法包括：
获取图像序列的关键点预测序列以及所述图像序列对应的关键点标签序列；所述关键点预测序列包括至少一个关键点的预测轨迹；所述关键点标签序列包括所述至少一个关键点的标签轨迹；
对于任一所述关键点，确定所述关键点的预测轨迹与所述关键点的标签轨迹之间的轨迹相似表征数据；所述轨迹相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现 的相似程度；
依据所述轨迹相似表征数据，确定所述关键点预测序列对应的关键点预测性能。
在一种可能的实施方式下，所述图像序列包括第一图像和第二图像，所述第一图像在所述图像序列中对应的时序早于所述第二图像在所述图像序列中对应的时序，所述第一图像在所述图像序列中对应的时序与所述第二图像在所述图像序列中对应的时序相邻；
所述关键点的预测轨迹包括所述关键点在所述第一图像下对应的预测位置和所述关键点在所述第二图像下对应的预测位置；
所述关键点的标签轨迹包括所述关键点在所述第一图像下对应的标签位置和所述关键点在所述第二图像下对应的标签位置；
所述轨迹相似表征数据的确定过程，包括：
计算所述关键点在所述第二图像下对应的预测位置与所述关键点在所述第一图像下对应的预测位置之间的距离，得到轨迹段预测长度；
计算所述关键点在所述第二图像下对应的标签位置与所述关键点在所述第一图像下对应的标签位置之间的距离，得到轨迹段标签长度；
根据所述轨迹段预测长度与所述轨迹段标签长度之间的差值，确定长度相似表征数据；
依据所述长度相似表征数据，确定所述轨迹相似表征数据。
在一种可能的实施方式下，所述图像序列包括第一图像和第二图像，所述第一图像在所述图像序列中对应的时序早于所述第二图像在所述图像序列中对应的时序，所述第一图像在所述图像序列中对应的时序与所述第二图像在所述图像序列中对应的时序相邻；
所述关键点的预测轨迹包括所述关键点在所述第一图像下对应的预测位置和所述关键点在所述第二图像下对应的预测位置；
所述关键点的标签轨迹包括所述关键点在所述第一图像下对应的标签位置和所述关键点在所述第二图像下对应的标签位置；
所述轨迹相似表征数据的确定过程，包括：
依据所述关键点在所述第二图像下对应的预测位置与所述关键点在所述第一图像下对应的预测位置之间的方向表征数据，确定轨迹段预测方向表征 数据；
依据所述关键点在所述第二图像下对应的标签位置与所述关键点在所述第一图像下对应的标签位置之间的方向表征数据，确定轨迹段标签方向表征数据；
根据所述轨迹段预测方向表征数据和所述轨迹段标签方向表征数据，确定所述方向相似表征数据；
依据所述方向相似表征数据，确定所述轨迹相似表征数据。
在一种可能的实施方式下，所述轨迹相似表征数据包括长度相似表征数据和/或方向相似表征数据；所述长度相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹在轨迹长度上所呈现的相似程度；所述方向相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹在轨迹方向上所呈现的相似程度。
在一种可能的实施方式下，所述依据所述轨迹相似表征数据，确定所述关键点预测序列对应的关键点预测性能，包括：
依据所述轨迹相似表征数据，确定所述关键点预测序列对应的预测稳定性表征数据；
根据所述关键点预测序列对应的预测稳定性表征数据，确定所述关键点预测序列对应的关键点预测性能。
在一种可能的实施方式下，所述根据所述关键点预测序列对应的预测稳定性表征数据，确定所述关键点预测序列对应的关键点预测性能之前，所述方法还包括：
根据所述关键点预测序列和所述关键点标签序列，确定所述关键点预测序列对应的预测精确性表征数据；
所述根据所述关键点预测序列对应的预测稳定性表征数据，确定所述关键点预测序列对应的关键点预测性能，包括：
根据所述关键点预测序列对应的预测稳定性表征数据以及所述关键点预测序列对应的预测精确性表征数据，确定所述关键点预测序列对应的关键点预测性能。
在一种可能的实施方式下，所述获取图像序列的关键点预测序列，包括：
利用关键点预测模型对所述图像序列进行关键点预测处理，得到所述图 像序列的关键点预测序列；
所述确定所述关键点预测序列对应的关键点预测性能之后，所述方法还包括：
依据所述关键点预测序列对应的关键点预测性能，对所述关键点预测模型进行更新。
本公开提供了一种数据处理装置，包括：
数据获取单元，被配置为获取图像序列的关键点预测序列以及所述图像序列对应的关键点标签序列；所述关键点预测序列包括至少一个关键点的预测轨迹；所述关键点标签序列包括所述至少一个关键点的标签轨迹；
第一确定单元，被配置为对于任一所述关键点，确定所述关键点的预测轨迹与所述关键点的标签轨迹之间的轨迹相似表征数据；所述轨迹相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度；
第二确定单元，被配置为依据所述轨迹相似表征数据，确定所述关键点预测序列对应的关键点预测性能。
本公开提供了一种电子设备，所述设备包括：处理器和存储器；
所述存储器，被配置为存储指令或计算机程序；
所述处理器，被配置为执行所述存储器中的所述指令或计算机程序，以使得所述电子设备执行本公开提供的数据处理方法。
本公开提供了一种计算机可读介质，所述计算机可读介质中存储有指令或计算机程序，当所述指令或计算机程序在设备上运行时，使得所述设备执行本公开提供的数据处理方法。
本公开提供了一种计算机程序产品，其包括承载在非暂态计算机可读介质上的计算机程序，该计算机程序包含用于执行本公开提供的数据处理方法的程序代码。
附图说明
为了更清楚地说明本公开实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本公开中记载的一些实施例，对于本领域普通技术人员来 讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其它的附图。
图1为本公开实施例提供的一种数据处理方法的流程图；
图2为本公开实施例提供的一种具有对应关系的预测轨迹与标签轨迹的示意图；
图3为本公开实施例提供的一种数据处理装置的结构示意图；
图4为本公开实施例提供的一种电子设备的结构示意图。
具体实施方式
经研究发现，对于一个关键点来说，在抖动情况(也就是，不稳定情况)下，该关键点的预测位置通常会分布在该关键点的标签位置周围，而且如果抖动越弱(比如，几乎没有发生抖动)，则该关键点的预测位置越接近于该关键点的标签位置，以使由该关键点在多个图像数据中对应的预测位置所形成的轨迹与由该关键点在多个图像数据中对应的标签位置所形成的轨迹之间越相似(比如，轨迹长度越接近、轨迹方向越相似等)；如果抖动越强烈(也就是，越不稳定)，则该关键点的预测位置越偏离该关键点的标签位置，以使由该关键点在多个图像数据中对应的预测位置所形成的轨迹与由该关键点在多个图像数据中对应的标签位置所形成的轨迹之间的差异性越大(比如，轨迹长度差距越大、轨迹方向越不相似等)。
基于上述研究发现，为了更好地评估在针对一个图像序列进行关键点预测处理时所呈现的预测性能(比如，稳定性等)，本公开提供了一种数据处理方法，该方法包括：先获取图像序列的关键点预测序列以及该图像序列对应的关键点标签序列，以使该关键点预测序列包括至少一个关键点的预测轨迹，并使得该关键点标签序列包括至少一个关键点的标签轨迹；然后，对于任一关键点，确定该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据能够表示出该关键点的预测轨迹与该关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度，从而使得该轨迹相似表征数据能够更好地表示出该关键点的预测轨迹相对于该关键点的标签轨迹所发生的抖动程度；最后，依据所有关键点对应的轨迹相似表征数据，确定该关键点预测序列对应的关键点预测性能，以使该关键点预测性能至少能够更好地表示出每个关键点的预测轨迹相对于其对应的标签轨迹所发生的 抖动程度，从而使得该关键点预测性能至少能够更好地表示针对该图像序列预测得到该关键点预测序列时所呈现的预测稳定性，进而使得该关键点预测性能能够更好地表示出在针对一个图像序列进行关键点预测处理时所呈现的预测性能。
另外，本公开不限定上文数据处理方法的执行主体，例如，本公开实施例提供的数据处理方法可以应用于终端设备或服务器等具有数据处理功能的设备。又如，本公开实施例提供的数据处理方法也可以借助不同设备(例如，终端设备与服务器、两个终端设备、或者两个服务器)之间的数据通信过程进行实现。其中，终端设备可以为智能手机、计算机、个人数字助理(Personal Digital Assitant，PDA)或平板电脑等。服务器可以为独立服务器、集群服务器或云服务器。
为了使本技术领域的人员更好地理解本公开方案，下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅是本公开一部分实施例，而不是全部的实施例。基于本公开中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本公开保护的范围。
为了更好地理解本公开所提供的技术方案，下面先结合一些附图对本公开提供的数据处理方法进行说明。如图1所示，本公开实施例提供的数据处理方法，包括下文S1-S3。其中，该图1为本公开实施例提供的一种数据处理方法的流程图。
S1：获取图像序列的关键点预测序列以及该图像序列对应的关键点标签序列；该关键点预测序列包括至少一个关键点的预测轨迹；该关键点标签序列包括至少一个关键点的标签轨迹。
其中，图像序列用于记录按照一定顺序(比如，时序)进行排列的多个图像数据；而且本公开不限定该图像序列的实施方式，比如，其可以采用视频数据进行实施。
图像序列的关键点预测序列用于描述该图像序列中的关键点预测位置变化情况；而且本公开不限定该关键点预测序列，比如，当该图像序列为{第1个图像数据，第2个图像数据，……(以此类推)，第N个图像数据}这一序列时，该图像序列的关键点预测序列可以包括该第1个图像数据的关键点预 测结果、该第2个图像数据的关键点预测结果、……(以此类推)、该第N个图像数据的关键点预测结果。其中，第n个图像数据的关键点预测结果用于描述该第n个图像数据中的关键点预测分布情况；而且本公开不限定该第n个图像数据的关键点预测结果，比如，该第n个图像数据的关键点预测结果可以包括多个关键点的预测位置，该关键点的预测位置用于描述该关键点在该第n个图像数据中预测所处位置。n为正整数，n≤N，N为正整数，N表示该图像序列中的图像数据个数。
实际上，对于上文图像序列的关键点预测序列来说，该关键点预测序列不仅可以按照图像数据维度表示关键点分布情况(比如，该关键点预测序列包括该图像序列中各个图像数据的关键点预测结果)，还可以按照关键点维度表示关键点分布情况(比如，该关键点预测序列包括该图像序列中各个关键点的预测位置变化情况)。
基于上段内容可知，在一种可能的实施方式下，当上文图像序列用于描述至少一个关键点时，该图像序列的关键点预测序列可以包括该至少一个关键点的预测轨迹，以使该关键点预测序列能够表示出该图像序列中各个关键点的预测位置变化情况。其中，对于任一关键点来说，该关键点的预测轨迹用于描述该关键点在该图像序列中的预测位置变化情况；而且本公开不限定该关键点的预测轨迹，比如，当该图像序列为{第1个图像数据，第2个图像数据，……，第N个图像数据}这一序列时，该关键点的预测轨迹可以包括该关键点在第1个图像数据下对应的预测位置(比如，图2所示的x0)、该关键点在第2个图像数据下对应的预测位置(比如，图2所示的x1)、该关键点在第3个图像数据下对应的预测位置(比如，图2所示的x2)、……(以此类推)、该关键点在第N个图像数据下对应的预测位置(比如，图2所示的x3)。该关键点在第n个图像数据下对应的预测位置用于描述该关键点在该第n个图像数据中预测所处位置。n为正整数，n≤N，N为正整数，N表示该图像序列中的图像数据个数。
需要说明的是，对于图2来说，x0→x1→x2→x3表示一个关键点的预测轨迹，x0表示该关键点在第1个图像数据下对应的预测位置，x1表示该关键点在第2个图像数据下对应的预测位置，x2表示该关键点在第3个图像数据下对应的预测位置，x3表示该关键点在第4个图像数据下对应的预测位置； g0→g1→g2→g3表示该关键点的标签轨迹，g0表示该关键点在第1个图像数据下对应的标签位置，g1表示该关键点在第2个图像数据下对应的标签位置，g2表示该关键点在第3个图像数据下对应的标签位置，g3表示该关键点在第4个图像数据下对应的标签位置。
还需要说明的是，本公开不限定位置(比如，上文“标签位置”或者上文“预测位置”)的表示方式，例如，其可以利用现有的或者未来出现的任意一种位置表示方式(比如，坐标等)进行实施。
另外，本公开不限定上文关键点预测序列的获取方式，例如，其可以采用现有的或者未来出现的任意一种能够针对一个图像序列进行关键点预测处理的方法(比如，借助预先构建的具有关键点预测功能的机器学习模型等)进行实施。
基于上段内容可知，在一种可能的实施方式下，上文关键点预测序列的获取过程具体可以为：在获取到图像序列之后，针对该图像序列进行关键点预测处理，得到该图像序列的关键点预测序列。在另一种可能的实施方式下，上文关键点预测序列的获取过程具体可以为：在获取到图像序列之后，利用关键点预测模型对该图像序列进行关键点预测处理，得到该图像序列的关键点预测序列。其中，该关键点预测模型用于针对该关键点预测模型的输入数据进行关键点预测处理；而且本公开不限定该关键点预测模型。
图像序列对应的关键点标签序列用于描述该图像序列中的关键点实际位置变化情况；而且本公开不限定该关键点标签序列，比如，当该图像序列为{第1个图像数据，第2个图像数据，……，第N个图像数据}这一序列时，该图像序列对应的关键点标签序列可以包括该第1个图像数据的关键点标签结果、该第2个图像数据的关键点标签结果、……(以此类推)、该第N个图像数据的关键点标签结果。其中，第n个图像数据的关键点标签结果用于描述该第n个图像数据中的关键点实际分布情况；而且本公开不限定该第n个图像数据的关键点标签结果，比如，该第n个图像数据的关键点标签结果可以包括多个关键点的标签位置，该关键点的标签位置用于描述该关键点在该第n个图像数据中实际所处位置。n为正整数，n≤N，N为正整数，N表示该图像序列中的图像数据个数。
又如，在一种可能的实施方式下，当上文图像序列用于描述至少一个关键 点时，该图像序列对应的关键点标签序列可以包括至少一个关键点的标签轨迹。其中，对于任一关键点来说，该关键点的标签轨迹用于描述该关键点在上文图像序列中的实际位置变化情况；而且本公开不限定该关键点的标签轨迹，比如，当该图像序列为{第1个图像数据，第2个图像数据，……，第N个图像数据}这一序列时，该关键点的标签轨迹可以包括该关键点在第1个图像数据下对应的标签位置(比如，图2所示的g0)、该关键点在第2个图像数据下对应的标签位置(比如，图2所示的g1)、该关键点在第3个图像数据下对应的标签位置(比如，图2所示的g2)、……(以此类推)、该关键点在第N个图像数据下对应的标签位置(比如，图2所示的g3)。该关键点在第n个图像数据下对应的标签位置用于描述该关键点在该第n个图像数据中实际所处位置。n为正整数，n≤N，N为正整数，N表示该图像序列中的图像数据个数。
另外，本公开不限定上文图像序列对应的关键点标签序列的确定方式，比如，其可以借助人工标注方式进行实施。
基于上文S1的相关内容可知，对于一个图像序列来说，获取该图像序列的关键点预测序列以及该图像序列对应的关键点标签序列，以使该关键点预测序列能够描述出该图像序列中的关键点预测位置变化情况(比如，一些关键点的预测轨迹等)，并使得该关键点标签序列用于描述上文图像序列中的关键点实际位置变化情况(比如，一些关键点的标签轨迹等)，以便后续能够基于该关键点预测序列与该关键点标签序列之间在某些方面(比如，轨迹长度等)所呈现的差异性，确定针对该图像序列进行关键点预测处理时所呈现的性能(比如，稳定性等)。
S2：对于任一关键点，确定该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据，该轨迹相似表征数据用于表示该关键点的预测轨迹与该关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度。
其中，第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据用于表示该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的相似程度，尤其能够表示出该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度。m为正整数，m≤上文“至少一个关键点”中的关键点个数(比如，上文图像 序列中所有关键点的个数等)。
另外，本公开不限定上文轨迹相似表征数据的确定过程，为了便于理解，下面结合一些情况进行说明。
情况1，经研究发现，对于两个轨迹来说，如果这两个轨迹越相似，则这两个轨迹的轨迹长度应该越接近，故在一些应用场景下，可以通过比较两个轨迹的轨迹长度的方式，来确定这两个轨迹之间的相似性。
基于上述情况1，本公开提供了上文轨迹相似表征数据的确定过程的一种可能的实施方式，为了便于理解，下面以上文“第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据”的确定过程为例进行说明。
作为示例，当上文图像序列包括第一图像(比如，第n个图像数据)和第二图像(比如，第n+1个图像数据)，该第一图像在该图像序列中对应的时序早于该第二图像在该图像序列中对应的时序，该第一图像在该图像序列中对应的时序与该第二图像在该图像序列中对应的时序相邻，上文第m个关键点的预测轨迹包括该第m个关键点在该第一图像下对应的预测位置和该第m个关键点在该第二图像下对应的预测位置，该第m个关键点的标签轨迹包括该第m个关键点在该第一图像下对应的标签位置和该m个关键点在该第二图像下对应的标签位置时，上文“第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据”的确定过程可以包括下文步骤11-步骤14。
步骤11：计算第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的距离，得到该第m个关键点在该第二图像与该第一图像之间的轨迹段预测长度。
其中，第一图像用于代表上文图像序列内任意一对位置相邻的图像数据中所处位置比较靠前的图像数据；第二图像用于代表这一对位置相邻的图像数据中所处位置比较靠后的图像数据。
基于上段内容可知，对于上文图像序列来说，当该图像序列为{第1个图像数据，第2个图像数据，……，第N个图像数据}这一序列时，如果上文第一图像为该第1个图像数据，则上文第二图像为该第2个图像数据；如果该第一图像为该第2个图像数据，则该第二图像为该第3个图像数据；……(以 此类推)；如果该第一图像为第N-1个图像数据，则该第二图像为该第N个图像数据。可见，在一种可能的实施方式中，如果该第一图像为该图像序列中存在的第n个图像数据，则该第二图像为该图像序列中存在的第n+1个图像数据。其中，n为正整数，n+1≤N。
上文“第m个关键点在第一图像下对应的预测位置”是指利用该第一图像针对该第m个关键点预测所得的位置信息，以使该“第m个关键点在第一图像下对应的预测位置”能够表示出该第m个关键点在该第一图像下预测所处位置。
上文“第m个关键点在第二图像下对应的预测位置”是指利用该第二图像针对该第m个关键点预测所得的位置信息，以使该“第m个关键点在第二图像下对应的预测位置”能够表示出该第m个关键点在该第二图像下预测所处位置。
上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的距离”用于描述该第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的轨迹长度；而且本公开实施例不限定该“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的距离”的计算过程，例如，其可以采用现有的或者未来出现的任意一种能够计算出两点之间距离的方法进行实施。
上文“第m个关键点在该第二图像与该第一图像之间的轨迹段预测长度”用于描述该第m个关键点在该第二图像与该第一图像之间预测所得的轨迹长度。
基于上文步骤11的相关内容可知，在一些应用场景下，如果上文图像序列为{第1个图像数据，第2个图像数据，……，第N个图像数据}这一序列，则上文第m个关键点的预测轨迹可以包括该第m个关键点在该第1个图像数据下对应的预测位置、该第m个关键点在该第2个图像数据下对应的预测位置、……(以此类推)、该第m个关键点在该第N个图像数据下对应的预测位置，并且可以将第m个关键点在第2个图像数据下对应的预测位置与该第m个关键点在第1个图像数据下对应的预测位置之间的距离，确定为该第m个关键点在该第2个图像数据与该第1个图像数据之间的轨迹段预测长度； 将第m个关键点在第3个图像数据下对应的预测位置与该第m个关键点在第2个图像数据下对应的预测位置之间的距离，确定为该第m个关键点在该第3个图像数据与该第2个图像数据之间的轨迹段预测长度；……(以此类推)；将第m个关键点在第N个图像数据下对应的预测位置与该第m个关键点在第N-1个图像数据下对应的预测位置之间的距离，确定为该第m个关键点在该第N个图像数据与该第N-1个图像数据之间的轨迹段预测长度，如此能够得到该第m个关键点在该图像序列内任意一对位置相邻的图像数据中所呈现的轨迹段的预测长度。
步骤12：计算第m个关键点在第二图像下对应的标签位置与该第m个关键点在第一图像下对应的标签位置之间的距离，得到该第m个关键点在该第二图像与该第一图像之间的轨迹段标签长度。
上文“第m个关键点在第一图像下对应的标签位置”是指预先在该第一图像中针对该第m个关键点所标注的位置信息，以使该“第m个关键点在第一图像下对应的标签位置”能够表示出该第m个关键点在该第一图像下实际所处位置。
上文“第m个关键点在第二图像下对应的标签位置”是指预先在该第二图像中针对该第m个关键点所标注的位置信息，以使该“第m个关键点在第二图像下对应的标签位置”能够表示出该第m个关键点在该第二图像下实际所处位置。
上文“第m个关键点在第二图像下对应的标签位置与该第m个关键点在第一图像下对应的标签位置之间的距离”用于描述该第m个关键点在第二图像下对应的标签位置与该第m个关键点在第一图像下对应的标签位置之间的轨迹长度；而且本公开实施例不限定该“第m个关键点在第二图像下对应的标签位置与该第m个关键点在第一图像下对应的标签位置之间的距离”的计算过程，例如，该计算过程的实施方式类似于上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的距离”的计算过程的实施方式，为了简要起见，在此不再赘述。
上文“第m个关键点在该第二图像与该第一图像之间的轨迹段标签长度”用于描述该第m个关键点在该第二图像与该第一图像之间的实际轨迹长度。
基于上文步骤12的相关内容可知，在一些应用场景下，如果上文图像序 列为{第1个图像数据，第2个图像数据，……，第N个图像数据}这一序列，则上文第m个关键点的标签轨迹可以包括该第m个关键点在该第1个图像数据下对应的标签位置、该第m个关键点在该第2个图像数据下对应的标签位置、……(以此类推)、该第m个关键点在该第N个图像数据下对应的标签位置，并且可以将第m个关键点在第2个图像数据下对应的标签位置与该第m个关键点在第1个图像数据下对应的标签位置之间的距离，确定为该第m个关键点在该第2个图像数据与该第1个图像数据之间的轨迹段标签长度；将第m个关键点在第3个图像数据下对应的标签位置与该第m个关键点在第2个图像数据下对应的标签位置之间的距离，确定为该第m个关键点在该第3个图像数据与该第2个图像数据之间的轨迹段标签长度；……(以此类推)；将第m个关键点在第N个图像数据下对应的标签位置与该第m个关键点在第N-1个图像数据下对应的标签位置之间的距离，确定为该第m个关键点在该第N个图像数据与该第N-1个图像数据之间的轨迹段标签长度，如此能够得到该第m个关键点在该图像序列内任意一对位置相邻的图像数据中所呈现的轨迹段的标签长度(也就是，实际长度)。
步骤13：根据具有对应关系的轨迹段预测长度与轨迹段标签长度之间的差值，确定该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的长度相似表征数据，以使该长度相似表征数据能够表示出该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间在轨迹长度上所呈现的相似程度。
上文“具有对应关系的轨迹段预测长度与轨迹段标签长度”分别用于描述同一个关键点在同一对位置相邻的图像数据中所呈现的轨迹段的预测长度和实际长度。例如，该“具有对应关系的轨迹段预测长度与轨迹段标签长度”可以包括：上文“第m个关键点在该第n+1个图像数据与该第n个图像数据之间的轨迹段预测长度”和上文“第m个关键点在该第n+1个图像数据与该第n个图像数据之间的轨迹段标签长度”；其中，n为正整数，n+1≤N。
另外，对于上文“具有对应关系的轨迹段预测长度与轨迹段标签长度”来说，该对应关系是预先确定的，比如，该对应关系具体可以包括：上文“第m个关键点在该第n+1个图像数据与该第n个图像数据之间的轨迹段预测长度”与上文“第m个关键点在该第n+1个图像数据与该第n个图像数据之间的轨迹段标签长度”之间的对应关系。其中，n为正整数，n+1≤N。
此外，本公开不限定上文步骤13的实施方式，例如，其可以采用下文公式(1)进行实施。
式中，E表示一个关键点的预测轨迹与该关键点的标签轨迹之间的长度相似表征数据；xi表示该关键点在第i+1个图像数据下对应的预测位置；xi-1表示该关键点在第i个图像数据下对应的预测位置；||xi-xi-1||2表示该关键点在该第i+1个图像数据与该第i个图像数据之间的轨迹段预测长度；gi表示该关键点在第i+1个图像数据下对应的标签位置；gi-1表示该关键点在第i个图像数据下对应的标签位置；||gi-gi-1||2表示该关键点在该第i+1个图像数据与该第i个图像数据之间的轨迹段预测长度；N表示上文图像序列中的图像数据个数。
步骤14：依据第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的长度相似表征数据，确定该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据。
需要说明的是，本公开不限定上文步骤14的实施方式，例如，其具体可以为：直接将第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的长度相似表征数据，确定为该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据至少能够表示出该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间在轨迹长度上所呈现的相似程度。
基于上文步骤11至步骤14的相关内容可知，对于上文图像序列中所描述的任意一个关键点(比如，上文第m个关键点)来说，如果该图像序列包括第一图像和第二图像，该关键点的预测轨迹包括该关键点在第一图像下对应的预测位置和该关键点在第二图像下对应的预测位置，而且该关键点的标签轨迹包括该关键点在第一图像下对应的标签位置和该关键点在第二图像下对应的标签位置，则该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据的确定过程，具体可以为：首先，计算该关键点在第二图像下对应的预测位置与该关键点在第一图像下对应的预测位置之间的距离，得到轨迹段预测长度，并且计该算关键点在第二图像下对应的标签位置与该关键点在第一图像下对应的标签位置之间的距离，得到轨迹段标签长度；然后，根据 该轨迹段预测长度与该轨迹段标签长度之间的差值，确定长度相似表征数据；最后，依据该长度相似表征数据，确定该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据至少能够表示出该关键点的预测轨迹与该关键点的标签轨迹之间在轨迹长度上所呈现的相似程度。
情况2，经研究发现，对于两个轨迹来说，如果这两个轨迹越相似，则这两个轨迹的轨迹方向的变化趋势应该越相似，故在一些应用场景下，可以通过比较两个轨迹的轨迹方向的方式，来确定这两个轨迹之间的相似性。
基于上述情况2，本公开提供了上文轨迹相似表征数据的确定过程的一种可能的实施方式，为了便于理解，下面以上文“第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据”的确定过程为例进行说明。
作为示例，当上文图像序列包括第一图像(比如，第n个图像数据)和第二图像(比如，第n+1个图像数据)，该第一图像在该图像序列中对应的时序早于该第二图像在该图像序列中对应的时序，该第一图像在该图像序列中对应的时序与该第二图像在该图像序列中对应的时序相邻，第m个关键点的预测轨迹包括该第m个关键点在该第一图像下对应的预测位置和该第m个关键点在该第二图像下对应的预测位置，该第m个关键点的标签轨迹包括该第m个关键点在该第一图像下对应的标签位置和该m个关键点在该第二图像下对应的标签位置时，上文“第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据”的确定过程可以包括下文步骤21-步骤24。
步骤21：依据第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的方向表征数据，确定该第m个关键点在该第二图像与该第一图像之间的轨迹段预测方向表征数据。
上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的方向表征数据”用于描述该第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的轨迹方向。
另外，本公开不限定上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的方向表征数据”的实 施方式，例如，其可以借助现有的或者未来出现的任意一种能够表示出两个位置之间方向的方式(比如，位置向量、单位向量、直线斜率、或者直线倾斜角等)进行实施。为了便于理解，下面的结合五个示例进行说明。
示例1，上文步骤21具体可以为：利用第m个关键点在第二图像下对应的预测位置减去该第m个关键点在第一图像下对应的预测位置，得到位置向量，并将该位置向量视为上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的方向表征数据”。
示例2，上文步骤21具体可以为：先利用第m个关键点在第二图像下对应的预测位置减去该第m个关键点在第一图像下对应的预测位置，得到位置向量；再利用该位置向量，确定单位向量，以使该单位向量能够描述出该位置向量所描述的方向；再将该单位向量视为上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的方向表征数据”。
示例3，上文步骤21具体可以为：针对第m个关键点在第二图像下对应的预测位置和该第m个关键点在第一图像下对应的预测位置进行直线拟合，得到直线斜率，并将该直线斜率视为上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的方向表征数据”。
示例4，上文步骤21具体可以为：先针对第m个关键点在第二图像下对应的预测位置和该第m个关键点在第一图像下对应的预测位置进行直线拟合，得到直线斜率；再将该直线斜率转化为直线倾斜角，并将该直线倾斜角视为上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的方向表征数据”。
示例5，上文步骤21具体可以为：依据第m个关键点在第二图像下对应的预测位置和该第m个关键点在第一图像下对应的预测位置，进行直线倾斜角计算处理，得到直线倾斜角，并将该直线倾斜角视为上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的方向表征数据”。
上文“第m个关键点在该第二图像与该第一图像之间的轨迹段预测方向表征数据”用于描述该第m个关键点在该第二图像与该第一图像之间预测所 得的轨迹方向。
基于上文步骤21的相关内容可知，在一些应用场景下，如果上文图像序列为{第1个图像数据，第2个图像数据，……，第N个图像数据}这一序列，则上文第m个关键点的预测轨迹可以包括该第m个关键点在该第1个图像数据下对应的预测位置、该第m个关键点在该第2个图像数据下对应的预测位置、……(以此类推)、该第m个关键点在该第N个图像数据下对应的预测位置，并且可以将第m个关键点在第2个图像数据下对应的预测位置与该第m个关键点在第1个图像数据下对应的预测位置之间的方向表征数据，确定为该第m个关键点在该第2个图像数据与该第1个图像数据之间的轨迹段预测方向；将第m个关键点在第3个图像数据下对应的预测位置与该第m个关键点在第2个图像数据下对应的预测位置之间的方向表征数据，确定为该第m个关键点在该第3个图像数据与该第2个图像数据之间的轨迹段预测方向；……(以此类推)；将第m个关键点在第N个图像数据下对应的预测位置与该第m个关键点在第N-1个图像数据下对应的预测位置之间的方向表征数据，确定为该第m个关键点在该第N个图像数据与该第N-1个图像数据之间的轨迹段预测方向，如此能够得到该第m个关键点在该图像序列内任意一对位置相邻的图像数据中所呈现的轨迹段的预测方向。
步骤22：依据第m个关键点在第二图像下对应的标签位置与该第m个关键点在第一图像下对应的标签位置之间的方向表征数据，确定该第m个关键点在该第二图像与该第一图像之间的轨迹段标签方向表征数据。
上文“第m个关键点在第二图像下对应的标签位置与该第m个关键点在第一图像下对应的标签位置之间的方向表征数据”用于描述该第m个关键点在第二图像下对应的标签位置与该第m个关键点在第一图像下对应的标签位置之间的轨迹方向。
另外，上文“第m个关键点在第二图像下对应的标签位置与该第m个关键点在第一图像下对应的标签位置之间的方向表征数据”的实施方式类似于上文“第m个关键点在第二图像下对应的预测位置与该第m个关键点在第一图像下对应的预测位置之间的方向表征数据”的实施方式，为了简要起见，在此不再赘述。
上文“第m个关键点在该第二图像与该第一图像之间的轨迹段标签方向 表征数据”用于描述该第m个关键点在该第二图像与该第一图像之间的实际轨迹方向。
基于上文步骤22的相关内容可知，在一些应用场景下，如果上文图像序列为{第1个图像数据，第2个图像数据，……，第N个图像数据}这一序列，则上文第m个关键点的标签轨迹可以包括该第m个关键点在该第1个图像数据下对应的标签位置、该第m个关键点在该第2个图像数据下对应的标签位置、……(以此类推)、该第m个关键点在该第N个图像数据下对应的标签位置，并且可以将第m个关键点在第2个图像数据下对应的标签位置与该第m个关键点在第1个图像数据下对应的标签位置之间的方向表征数据，确定为该第m个关键点在该第2个图像数据与该第1个图像数据之间的轨迹段标签方向；将第m个关键点在第3个图像数据下对应的标签位置与该第m个关键点在第2个图像数据下对应的标签位置之间的方向表征数据，确定为该第m个关键点在该第3个图像数据与该第2个图像数据之间的轨迹段标签方向；……(以此类推)；将第m个关键点在第N个图像数据下对应的标签位置与该第m个关键点在第N-1个图像数据下对应的标签位置之间的方向表征数据，确定为该第m个关键点在该第N个图像数据与该第N-1个图像数据之间的轨迹段标签方向，如此能够得到该第m个关键点在该图像序列内任意一对位置相邻的图像数据中所呈现的轨迹段的标签方向(也就是，实际方向)。
步骤23：根据上文轨迹段预测方向表征数据和上文轨迹段标签方向表征数据，确定第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的方向相似表征数据，以使该方向相似表征数据能够表示出该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间在轨迹方向上所呈现的相似程度。
需要说明的是，本公开不限定步骤23的实施方式，例如，其具体可以为：先依据第m个关键点在该第n个图像数据与该第n+1个图像数据之间的轨迹段预测方向、以及该第m个关键点在该第n个图像数据与该第n+1个图像数据之间的轨迹段标签方向，确定该第m个关键点在该第n个图像数据与该第n+1个图像数据之间的方向差表征数据，以使该方向差表征数据能够表示出该轨迹段预测方向相对于该轨迹段标签方向发生的方向偏离情况，n为正整数，n+1≤N；再计算该第m个关键点对应的所有方向差表征数据的平均值，作为该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的方向相似表 征数据。
还需要说明的是，本公开不限定上段中“方向差表征数据”的计算方式，比如，如果上文轨迹段预测方向与上文轨迹段标签方向均采用向量(比如，位置向量或者单位向量)进行表示，则该“方向差表征数据”可以借助内积计算公式进行确定。又如，如果上文轨迹段预测方向与上文轨迹段标签方向均采用直线斜率进行表示，则该“方向差表征数据”可以直接借助差值计算公式或者两个直线斜率之间夹角计算公式进行实施。还如，如果上文轨迹段预测方向与上文轨迹段标签方向均采用直线倾斜角进行表示，则该“方向差表征数据”可以直接借助差值计算公式进行实施。
步骤24：依据第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的方向相似表征数据，确定该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据。
需要说明的是，本公开不限定上文步骤24的实施方式，例如，其具体可以为：直接将第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的方向相似表征数据，确定为该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据至少能够表示出该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间在轨迹方向上所呈现的相似程度。
基于上文步骤21至步骤24的相关内容可知，对于上文图像序列中所描述的任意一个关键点(比如，上文第m个关键点)来说，如果该图像序列包括第一图像和第二图像，该关键点的预测轨迹包括该关键点在第一图像下对应的预测位置和该关键点在第二图像下对应的预测位置，而且该关键点的标签轨迹包括该关键点在第一图像下对应的标签位置和该关键点在第二图像下对应的标签位置，则该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据的确定过程，具体可以为：首先，依据该关键点在第二图像下对应的预测位置与该关键点在第一图像下对应的预测位置之间的方向表征数据，确定轨迹段预测方向表征数据，并且依据该关键点在第二图像下对应的标签位置与该关键点在第一图像下对应的标签位置之间的方向表征数据，确定轨迹段标签方向表征数据；再根据该轨迹段预测方向表征数据和该轨迹段标签方向表征数据，确定方向相似表征数据；最后，依据该方向相似表征数据，确 定该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据至少能够表示出该关键点的预测轨迹与该关键点的标签轨迹之间在轨迹方向上所呈现的相似程度。
情况3，在一些应用场景下，为了更好地提高轨迹相似性确定效果，可以通过比较两个轨迹的轨迹方向以及轨迹长度的方式，来确定这两个轨迹之间的相似性。
基于上述情况3，本公开提供了上文轨迹相似表征数据的确定过程的一种可能的实施方式，其具体可以包括下文步骤31-步骤32。
步骤31：依据第m个关键点的预测轨迹与该第m个关键点的标签轨迹，确定该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的长度相似表征数据、以及该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的方向相似表征数据。m为正整数，m≤上文“至少一个关键点”中的关键点个数。
步骤32：依据第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的长度相似表征数据、以及该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的方向相似表征数据，确定该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据能够表示出该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间在轨迹长度以及轨迹方向上所呈现的相似程度。
需要说明的是，本公开不限定步骤32的实施方式，例如，该步骤32具体可以为：将第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的长度相似表征数据、以及该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的方向相似表征数据进行集合，得到该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据包括该长度相似表征数据和该方向相似表征数据。
又如，上文步骤32具体可以为：将第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的长度相似表征数据、以及该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的方向相似表征数据进行加权求和，得到该第m个关键点的预测轨迹与该第m个关键点的标签轨迹之间的轨迹相似表征数据。
基于上文步骤31至步骤32的相关内容可知，对于上文图像序列中所描述的任意一个关键点(比如，上文第m个关键点)来说，可以先依据该关键点的预测轨迹与该关键点的标签轨迹，确定长度相似表征数据以及方向相似表征数据；再依据该长度相似表征数据以及该方向相似表征数据，确定该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据至少能够表示出该关键点的预测轨迹与该关键点的标签轨迹之间在轨迹长度以及轨迹方向上所呈现的相似程度。
基于上文“轨迹相似表征数据”的相关内容可知，在一种可能的实施方式下，对于上文图像序列中所描述的任意一个关键点(比如，上文第m个关键点)来说，该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据包括长度相似表征数据和/或方向相似表征数据；该长度相似表征数据用于表示该关键点的预测轨迹与该关键点的标签轨迹在轨迹长度上所呈现的相似程度；该方向相似表征数据用于表示该关键点的预测轨迹与该关键点的标签轨迹在轨迹方向上所呈现的相似程度。
基于上文S2的相关内容可知，对于上文图像序列中所描述的任意一个关键点来说，在获取到该关键点的预测轨迹以及该关键点的标签轨迹之后，可以计算该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据能够表示出该关键点的预测轨迹与该关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度，以便后续能够基于该轨迹相似表征数据，确定针对该图像序列进行关键点预测处理时所呈现的性能(比如，稳定性等)。
S3：依据轨迹相似表征数据，确定关键点预测序列对应的关键点预测性能。
其中，关键点预测序列对应的关键点预测性能用于描述在针对上文图像序列预测得到该关键点预测序列时所呈现的性能(比如，稳定性等)。
另外，本公开不限定上文S3的实施方式，例如，其具体可以为：直接将所有关键点对应的轨迹相似表征数据进行某种统计分析(比如，平均值、最大值、中值等)，得到上文关键点预测序列对应的关键点预测性能。
经研究发现，预测轨迹与标签轨迹之间的相似性可以被用于衡量关键点预测结果的稳定性。基于此，本公开还提供了上文S3的一种可能的实施方式， 其具体可以包括下文步骤41-步骤42。
步骤41：依据上文轨迹相似表征数据，确定上文关键点预测序列对应的预测稳定性表征数据，以使该预测稳定性表征数据用于表示在针对上文图像序列预测得到该关键点预测序列时所呈现的预测稳定性。
需要说明的是，本公开不限定上文步骤41的实施方式，例如，该步骤41具体可以为：将所有关键点对应的轨迹相似表征数据进行集合，得到上文关键点预测序列对应的预测稳定性表征数据，以使该预测稳定性表征数据包括所有关键点对应的轨迹相似表征数据。
又如，上文步骤41也可以为：直接将所有关键点对应的轨迹相似表征数据之间的平均值，确定为上文关键点预测序列对应的预测稳定性表征数据，以使该预测稳定性表征数据用于表示在针对上文图像序列预测得到该关键点预测序列时所呈现的稳定性。
步骤42：根据上文关键点预测序列对应的预测稳定性表征数据，确定该关键点预测序列对应的关键点预测性能。
需要说明的是，本公开不限定上文步骤42的实施方式，例如，当上文关键点预测序列对应的预测稳定性表征数据是一个数据值(比如，上文“所有关键点对应的轨迹相似表征数据之间的平均值”等)时，该步骤42具体可以为：直接将该预测稳定性表征数据，确定为该关键点预测序列对应的关键点预测性能。
又如，当上文关键点预测序列对应的预测稳定性表征数据是一个数据集合(比如，该预测稳定性表征数据包括所有关键点对应的轨迹相似表征数据)时，上文步骤42具体可以为：将该数据集合中所有元素之间的平均值，确定为关键点预测序列对应的预测稳定性表征数据。
经研究发现，在一些应用场景下，对于图像序列来说，当评估针对该图像序列的关键点预测效果时，不仅可以考虑稳定性，还可以考虑精确性。基于此，本公开还提供了上文关键点预测序列对应的关键点预测性能的确定过程(也就是，上文步骤42)的一种可能的实施方式，其具体可以包括下文步骤421-步骤422。
步骤421：根据上文图像序列的关键点预测序列和该图像序列对应的关键点标签序列，确定该关键点预测序列对应的预测精确性表征数据，以使该预测 精确性表征数据用于表示在针对该图像序列预测得到该关键点预测序列时所呈现的预测精确性。
需要说明的是，本公开不限定上文步骤421的实施方式，例如，其可以采用现有的或者未来出现的任意一种能够借助预测信息以及标准信息进行预测精确性计算处理的方法(比如，计算预测信息以及标准信息之间的相似度的方式)进行实施。为了便于理解，下面结合两个示例进行说明。
假设上文图像序列为{第1个图像数据，第2个图像数据，……，第N个图像数据}这一序列；该图像序列的关键点预测序列包括该第m个关键点在该第1个图像数据下对应的预测位置、该第m个关键点在该第2个图像数据下对应的预测位置、……(以此类推)、该第m个关键点在该第N个图像数据下对应的预测位置，m为正整数，m≤上文“至少一个关键点”中的关键点个数；该图像序列对应的关键点标签序列包括该第m个关键点在该第1个图像数据下对应的标签位置、该第m个关键点在该第2个图像数据下对应的标签位置、……(以此类推)、该第m个关键点在该第N个图像数据下对应的标签位置，m为正整数，m≤上文“至少一个关键点”中的关键点个数。
示例1，基于上述假设可知，上文步骤421具体可以为：首先，计算第m个关键点在该第n个图像数据下对应的预测位置与该第m个关键点在该第n个图像数据下对应的标签位置之间的距离，得到该第m个关键点在该第n个图像数据下对应的位置预测损失，n为正整数，n≤N，m为正整数，m≤上文“至少一个关键点”中的关键点个数；然后，所有关键点在所有图像数据下对应的位置预测损失进行平均值计算处理(或者，加和处理)，得到上文关键点预测序列对应的预测精确性表征数据。
示例2，基于上述假设可知，上文步骤421具体可以为：首先，计算第m个关键点在该第n个图像数据下对应的预测位置与该第m个关键点在该第n个图像数据下对应的标签位置之间的距离，得到该第m个关键点在该第n个图像数据下对应的位置预测损失，n为正整数，n≤N，m为正整数，m≤上文“至少一个关键点”中的关键点个数；然后，将所有关键点在所有图像数据下对应的位置预测损失进行集合，得到上文关键点预测序列对应的预测精确性表征数据。
需要说明的是，本公开不限定上文步骤421的执行时间，只需确保该步 骤421的执行时间晚于上文S1的执行时间，并且确保该步骤421的执行时间早于下文步骤422的执行时间即可。
步骤422：根据上文关键点预测序列对应的预测稳定性表征数据以及该关键点预测序列对应的预测精确性表征数据，确定该关键点预测序列对应的关键点预测性能。
需要说明的是，本公开不限定上文步骤422的实施方式，例如，当上文关键点预测序列对应的预测稳定性表征数据与该关键点预测序列对应的预测精确性表征数据均是一个数据值时，该步骤422具体可以为：直接将该关键点预测序列对应的预测稳定性表征数据与该关键点预测序列对应的预测精确性表征数据进行加权求和处理，得到该关键点预测序列对应的关键点预测性能。
又如，当上文关键点预测序列对应的预测稳定性表征数据与该关键点预测序列对应的预测精确性表征数据均是一个数据集合时，上文步骤422具体可以为：先分别计算两个数据集合中所有元素之间的平均值；再将两个平均值进行加权求和处理，得到该关键点预测序列对应的关键点预测性能。
还如，当上文关键点预测序列对应的预测稳定性表征数据与该关键点预测序列对应的预测精确性表征数据均是一个数据集合时，上文步骤422具体可以为：首先，将第m个关键点在该第n+1个图像数据下对应的位置预测损失与该第m个关键点在该第n个图像数据下对应的位置预测损失之间的平均值，确定为该第m个关键点在该第n个图像数据与该第n+1个图像数据之间的轨迹段预测信息(比如，轨迹段预测长度和/或轨迹段预测方向)对应的加权权重，n为正整数，n≤N，m为正整数，m≤上文“至少一个关键点”中的关键点个数；然后，按照该加权权重，对这些轨迹段预测信息进行加权求和处理，得到该关键点预测序列对应的关键点预测性能。
基于上文步骤421至步骤422的相关内容可知，在一些应用场景下，对于一个图像序列来说，在获取到该图像序列的关键点预测序列以及该图像序列对应的关键点标签序列之后，不仅需要依据该关键点预测序列以及该关键点标签序列，分析在针对该图像序列预测得到该关键点预测序列时所呈现的预测稳定性，还需要依据该关键点预测序列以及该关键点标签序列，分析在针对该图像序列预测得到该关键点预测序列时所呈现的预测精确性，以便后续能够借助这两项分析结果，确定该关键点预测序列对应的关键点预测性能，以 使该关键点预测性能能够更好地表示出针对该图像序列预测得到该关键点预测序列时所呈现的预测效果。
基于上文步骤41至步骤42的相关内容可知，在一些应用场景下，对于一个图像序列来说，在获取到该图像序列的关键点预测序列以及该图像序列对应的关键点标签序列之后，可以先依据该关键点预测序列以及该关键点标签序列，确定该图像序列所描述的各个关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据；再依据这些轨迹相似表征数据，计算出该关键点预测序列对应的预测稳定性表征数据，以使该预测稳定性表征数据用于表示在针对该图像序列预测得到该关键点预测序列时所呈现的预测稳定性；最后，依据该预测稳定性表征数据(以及该关键点预测序列对应的预测精确性表征数据)，确定出该关键点预测序列对应的关键点预测性能，以使该关键点预测性能能够表示出针对该图像序列预测得到该关键点预测序列时在至少一个评估指标(比如，稳定性+精确性)下所达到的状态，从而使得该关键点预测性能能够更好地表示出针对该图像序列预测得到该关键点预测序列时所呈现的预测效果。
基于上文S1至S3的相关内容可知，对于本公开实施例提供的数据处理方法来说，先获取图像序列的关键点预测序列以及该图像序列对应的关键点标签序列，以使该关键点预测序列包括至少一个关键点的预测轨迹，并使得该关键点标签序列包括至少一个关键点的标签轨迹；然后，对于任一关键点，确定该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据能够表示出该关键点的预测轨迹与该关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度，从而使得该轨迹相似表征数据能够更好地表示出该关键点的预测轨迹相对于该关键点的标签轨迹所发生的抖动程度；最后，依据所有关键点对应的轨迹相似表征数据，确定该关键点预测序列对应的关键点预测性能，以使该关键点预测性能至少能够更好地表示出每个关键点的预测轨迹相对于其对应的标签轨迹所发生的抖动程度，从而使得该关键点预测性能至少能够更好地表示针对该图像序列预测得到该关键点预测序列时所呈现的预测稳定性，进而使得该关键点预测性能能够更好地表示出在针对一个图像序列进行关键点预测处理时所呈现的预测性能。
另外，对于本公开提供的数据处理方法来说，本公开不限定该数据处理方法的应用场景，比如，其可以应用于在任意一个需要针对图像序列的关键点预测序列进行评估处理的应用场景(比如，模型训练等场景)。为了更好地理解，下面以一个应用场景为例进行说明。
作为示例，在一些应用场景(比如，模型训练场景)下，本公开提供的数据处理方法可以包括下文步骤51-步骤55。
步骤51：获取图像序列以及该图像序列对应的关键点标签序列；该图像序列用于描述至少一个关键点，该关键点标签序列包括该至少一个关键点的标签轨迹。
需要说明的是，本公开不限定步骤51中图像序列的获取方式，比如，其具体可以为：从训练数据集中随机抽取一个图像序列，作为当前轮训练过程所需使用的图像序列。其中，该训练数据集是指在针对一个机器学习模型进行训练时所需依据的数据集；而且该机器学习模型在训练过程中的学习目标为：对一个图像序列进行关键点预测处理。另外，本公开不限定该训练数据集的实施方式，比如，该训练数据集可以包括多个图像序列以及各个图像序列对应的关键点标签序列。
还需要说明的是，上文步骤51中“图像序列对应的关键点标签序列”的相关内容请参见上文S1，为了简要起见，在此不再赘述。另外，本公开也不限定该步骤51中“图像序列对应的关键点标签序列”的获取方式，比如，其具体可以为：在从训练数据集中随机抽取到一个图像序列之后，从该训练数据集中提取该图像序列对应的关键点标签序列，作为当前轮训练过程所需使用的关键点标签序列。
步骤52：利用关键点预测模型对上文图像序列进行关键点预测处理，得到该图像序列的关键点预测序列；该关键点预测序列包括上文至少一个关键点的预测轨迹。
其中，关键点预测模型用于针对该关键点预测模型的输入数据进行关键点预测处理；而且本公开不限定该关键点预测模型，比如，其可以采用现有的或者未来出现的任意一种需要进行关键点预测方面学习训练的机器学习模型进行实施。
另外，上文步骤52中“图像序列的关键点预测序列”的相关内容请参见 上文S1，为了简要起见，在此不再赘述。
步骤53：对于任一关键点，确定该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据，该轨迹相似表征数据用于表示该关键点的预测轨迹与该关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度。
需要说明的是，步骤53的相关内容请参见上文S2，为了简要起见，在此不再赘述。
步骤54：依据上文轨迹相似表征数据，确定关键点预测序列对应的关键点预测性能。
需要说明的是，步骤54的相关内容请参见上文S3，为了简要起见，在此不再赘述。
步骤55：依据上文关键点预测序列对应的关键点预测性能，对关键点预测模型进行更新，并返回继续执行上文步骤51及其后续步骤，直至达到预设停止条件。
需要说明的是，本公开不限定上文步骤55中更新过程的实施方式，例如，其具体可以为：先利用上文关键点预测序列对应的关键点预测性能，确定关键点预测模型的模型损失；再依据该模型损失对该关键点预测模型进行更新，以得到更新后的关键点预测模型，以使该更新后的关键点预测模型具有更好地关键点预测性能，以便后续能够基于该更新后的关键点预测模型，继续执行上文步骤51及其后续步骤，以实现针对该关键点预测模型的下一轮训练过程。
上文预设停止条件是指在针对关键点预测模型的训练过程结束时所需达到的条件；而且该预设停止条件可以预先依据应用场景设定。比如，该预设停止条件具体可以包括：该关键点预测模型的模型损失低于预先设定的第一阈值。又如，该预设停止条件具体可以包括：该关键点预测模型的模型损失的变化率低于预先设定的第二阈值。还如，该预设停止条件具体可以包括：该关键点预测模型的更新次数高于预先设定的第三阈值。
基于上段内容可知，对于针对关键点预测模型的当前轮训练过程来说，如果确定达到预设停止条件，则可以确定该关键点预测模型的预测性能比较好，故可以直接停止针对该关键点预测模型的训练过程即可；但是，如果确定仍未达到预设停止条件，则可以确定该关键点预测模型的预测性能仍需完善，故可 以利用在当前轮训练过程中所确定的关键点预测序列对应的关键点预测性能，对该关键点预测模型进行更新，以使该更新后的关键点预测模型具有更好地关键点预测性能，以便后续能够基于该更新后的关键点预测模型，继续执行上文步骤51及其后续步骤，以实现针对该关键点预测模型的下一轮训练过程。如此能够实现针对该关键点预测模型进行至少一轮训练的目的。
基于上文步骤51至步骤55的相关内容可知，对于针对关键点预测模型的每一轮训练过程来说，可以先利用利用关键点预测模型对图像序列进行关键点预测处理，得到该图像序列的关键点预测序列；再依据该图像序列的关键点预测序列以及该图像序列对应的关键点标签序列，确定该关键点预测序列对应的关键点预测性能；最后，依据该关键点预测性能，对关键点预测模型进行更新，以使该更新后的关键点预测模型具有更好地关键点预测性能。其中，因该关键点预测性能能够更好地表示出针对该图像序列预测得到该关键点预测序列时所呈现的预测效果，以使该关键点预测性能能够更好地表示出该关键点预测模型的预测性能，从而使得该关键点预测性能能够更好地指导该关键点预测模型的更新过程，进而使得基于该关键点预测性能训练所得的关键点预测模型具有更好地预测性能(比如，预测稳定性+预测精确性)，如此有利于提高针对该关键点预测模型的模型训练效果。
基于本公开实施例提供的数据处理方法，本公开实施例还提供了一种数据处理装置，下面结合图3进行解释和说明。其中，图3为本公开实施例提供的一种数据处理装置的结构示意图。需要说明的是，本公开实施例提供的数据处理装置的技术详情，请参照上文数据处理方法的相关内容。
如图3所示，本公开实施例提供的数据处理装置300，包括：
数据获取单元301，用于获取图像序列的关键点预测序列以及所述图像序列对应的关键点标签序列；所述关键点预测序列包括至少一个关键点的预测轨迹；所述关键点标签序列包括所述至少一个关键点的标签轨迹；
第一确定单元302，用于对于任一所述关键点，确定所述关键点的预测轨迹与所述关键点的标签轨迹之间的轨迹相似表征数据；所述轨迹相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度；
第二确定单元303，用于依据所述轨迹相似表征数据，确定所述关键点预 测序列对应的关键点预测性能。
在一种可能的实施方式下，所述图像序列包括第一图像和第二图像，所述第一图像在所述图像序列中对应的时序早于所述第二图像在所述图像序列中对应的时序，所述第一图像在所述图像序列中对应的时序与所述第二图像在所述图像序列中对应的时序相邻；
所述关键点的预测轨迹包括所述关键点在所述第一图像下对应的预测位置和所述关键点在所述第二图像下对应的预测位置；
所述关键点的标签轨迹包括所述关键点在所述第一图像下对应的标签位置和所述关键点在所述第二图像下对应的标签位置；
所述第一确定单元302，具体用于：计算所述关键点在所述第二图像下对应的预测位置与所述关键点在所述第一图像下对应的预测位置之间的距离，得到轨迹段预测长度；计算所述关键点在所述第二图像下对应的标签位置与所述关键点在所述第一图像下对应的标签位置之间的距离，得到轨迹段标签长度；根据所述轨迹段预测长度与所述轨迹段标签长度之间的差值，确定长度相似表征数据；依据所述长度相似表征数据，确定所述轨迹相似表征数据。
在一种可能的实施方式下，所述图像序列包括第一图像和第二图像，所述第一图像在所述图像序列中对应的时序早于所述第二图像在所述图像序列中对应的时序，所述第一图像在所述图像序列中对应的时序与所述第二图像在所述图像序列中对应的时序相邻；
所述关键点的预测轨迹包括所述关键点在所述第一图像下对应的预测位置和所述关键点在所述第二图像下对应的预测位置；
所述关键点的标签轨迹包括所述关键点在所述第一图像下对应的标签位置和所述关键点在所述第二图像下对应的标签位置；
所述第一确定单元302，具体用于：依据所述关键点在所述第二图像下对应的预测位置与所述关键点在所述第一图像下对应的预测位置之间的方向表征数据，确定轨迹段预测方向表征数据；依据所述关键点在所述第二图像下对应的标签位置与所述关键点在所述第一图像下对应的标签位置之间的方向表征数据，确定轨迹段标签方向表征数据；根据所述轨迹段预测方向表征数据和所述轨迹段标签方向表征数据，确定所述方向相似表征数据；依据所述方向相似表征数据，确定所述轨迹相似表征数据。
在一种可能的实施方式下，所述轨迹相似表征数据包括长度相似表征数据和/或方向相似表征数据；所述长度相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹在轨迹长度上所呈现的相似程度；所述方向相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹在轨迹方向上所呈现的相似程度。
在一种可能的实施方式下，所述第二确定单元303，包括：
第一确定子单元，用于依据所述轨迹相似表征数据，确定所述关键点预测序列对应的预测稳定性表征数据；
第二确定子单元，用于根据所述关键点预测序列对应的预测稳定性表征数据，确定所述关键点预测序列对应的关键点预测性能。
在一种可能的实施方式下，所述数据处理装置300还包括：
第三确定单元，用于根据所述关键点预测序列和所述关键点标签序列，确定所述关键点预测序列对应的预测精确性表征数据；
所述第二确定子单元，具体用于：根据所述关键点预测序列对应的预测稳定性表征数据以及所述关键点预测序列对应的预测精确性表征数据，确定所述关键点预测序列对应的关键点预测性能。
在一种可能的实施方式下，所述数据获取单元301，具体用于：利用关键点预测模型对所述图像序列进行关键点预测处理，得到所述图像序列的关键点预测序列；
所述数据处理装置300还包括：
模型更新单元，用于依据所述关键点预测序列对应的关键点预测性能，对所述关键点预测模型进行更新。
基于上述数据处理装置300的相关内容可知，对于本公开实施例提供的数据处理装置300来说，先获取图像序列的关键点预测序列以及该图像序列对应的关键点标签序列，以使该关键点预测序列包括至少一个关键点的预测轨迹，并使得该关键点标签序列包括至少一个关键点的标签轨迹；然后，对于任一关键点，确定该关键点的预测轨迹与该关键点的标签轨迹之间的轨迹相似表征数据，以使该轨迹相似表征数据能够表示出该关键点的预测轨迹与该关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度，从而使得该轨迹相似表征数据能够更好地表示出该关键点的预测轨迹相对于该关 键点的标签轨迹所发生的抖动程度；最后，依据所有关键点对应的轨迹相似表征数据，确定该关键点预测序列对应的关键点预测性能，以使该关键点预测性能至少能够更好地表示出每个关键点的预测轨迹相对于其对应的标签轨迹所发生的抖动程度，从而使得该关键点预测性能至少能够更好地表示针对该图像序列预测得到该关键点预测序列时所呈现的预测稳定性，进而使得该关键点预测性能能够更好地表示出在针对一个图像序列进行关键点预测处理时所呈现的预测性能。
另外，本公开实施例还提供了一种电子设备，所述设备包括处理器以及存储器：所述存储器，用于存储指令或计算机程序；所述处理器，用于执行所述存储器中的所述指令或计算机程序，以使得所述电子设备执行本公开实施例提供的数据处理方法的任一实施方式。
参见图4，其示出了适于用来实现本公开实施例的电子设备400的结构示意图。本公开实施例中的终端设备可以包括但不限于诸如移动电话、笔记本电脑、数字广播接收器、PDA(个人数字助理)、PAD(平板电脑)、PMP(便携式多媒体播放器)、车载终端(例如车载导航终端)等等的移动终端以及诸如数字TV、台式计算机等等的固定终端。图4示出的电子设备仅仅是一个示例，不应对本公开实施例的功能和使用范围带来任何限制。
如图4所示，电子设备400可以包括处理装置(例如中央处理器、图形处理器等)401，其可以根据存储在只读存储器(ROM)402中的程序或者从存储装置408加载到随机访问存储器(RAM)403中的程序而执行各种适当的动作和处理。在RAM403中，还存储有电子设备400操作所需的各种程序和数据。处理装置401、ROM 402以及RAM 403通过总线404彼此相连。输入/输出(I/O)接口405也连接至总线404。
通常，以下装置可以连接至I/O接口405：包括例如触摸屏、触摸板、键盘、鼠标、摄像头、麦克风、加速度计、陀螺仪等的输入装置406；包括例如液晶显示器(LCD)、扬声器、振动器等的输出装置407；包括例如磁带、硬盘等的存储装置408；以及通信装置409。通信装置409可以允许电子设备400与其他设备进行无线或有线通信以交换数据。虽然图4示出了具有各种装置的电子设备400，但是应理解的是，并不要求实施或具备所有示出的装置。可以替代地实施或具备更多或更少的装置。
特别地，根据本公开的实施例，上文参考流程图描述的过程可以被实现为计算机软件程序。例如，本公开的实施例包括一种计算机程序产品，其包括承载在非暂态计算机可读介质上的计算机程序，该计算机程序包含用于执行流程图所示的方法的程序代码。在这样的实施例中，该计算机程序可以通过通信装置409从网络上被下载和安装，或者从存储装置408被安装，或者从ROM402被安装。在该计算机程序被处理装置401执行时，执行本公开实施例的方法中限定的上述功能。
本公开实施例提供的电子设备与上述实施例提供的方法属于同一发明构思，未在本实施例中详尽描述的技术细节可参见上述实施例，并且本实施例与上述实施例具有相同的有益效果。
本公开实施例还提供了一种计算机可读介质，所述计算机可读介质中存储有指令或计算机程序，当所述指令或计算机程序在设备上运行时，使得所述设备执行本公开实施例提供的数据处理方法的任一实施方式。
需要说明的是，本公开上述的计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质或者是上述两者的任意组合。计算机可读存储介质例如可以是——但不限于——电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子可以包括但不限于：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机访问存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本公开中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。而在本公开中，计算机可读信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。计算机可读信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读信号介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。计算机可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于：电线、光缆、RF(射频)等等，或者上述的任意合适的组合。
在一些实施方式中，客户端、服务器可以利用诸如HTTP(Hyper Text Transfer Protocol，超文本传输协议)之类的任何当前已知或未来研发的网络协议进行通信，并且可以与任意形式或介质的数字数据通信(例如，通信网络)互连。通信网络的示例包括局域网(“LAN”)，广域网(“WAN”)，网际网(例如，互联网)以及端对端网络(例如，ad hoc端对端网络)，以及任何当前已知或未来研发的网络。
上述计算机可读介质可以是上述电子设备中所包含的；也可以是单独存在，而未装配入该电子设备中。
上述计算机可读介质承载有一个或者多个程序，当上述一个或者多个程序被该电子设备执行时，使得该电子设备可以执行上述方法。
可以以一种或多种程序设计语言或其组合来编写用于执行本公开的操作的计算机程序代码，上述程序设计语言包括但不限于面向对象的程序设计语言—诸如Java、Smalltalk、C++，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络——包括局域网(LAN)或广域网(WAN)—连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。
附图中的流程图和框图，图示了按照本公开各种实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段、或代码的一部分，该模块、程序段、或代码的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。也应当注意，在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个接连地表示的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或操作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
描述于本公开实施例中所涉及到的单元可以通过软件的方式实现，也可 以通过硬件的方式来实现。其中，单元/模块的名称在某种情况下并不构成对该单元本身的限定。
本文中以上描述的功能可以至少部分地由一个或多个硬件逻辑部件来执行。例如，非限制性地，可以使用的示范类型的硬件逻辑部件包括：现场可编程门阵列(FPGA)、专用集成电路(ASIC)、专用标准产品(ASSP)、片上系统(SOC)、复杂可编程逻辑设备(CPLD)等等。
在本公开的上下文中，机器可读介质可以是有形的介质，其可以包含或存储以供指令执行系统、装置或设备使用或与指令执行系统、装置或设备结合地使用的程序。机器可读介质可以是机器可读信号介质或机器可读储存介质。机器可读介质可以包括但不限于电子的、磁性的、光学的、电磁的、红外的、或半导体系统、装置或设备，或者上述内容的任何合适组合。机器可读存储介质的更具体示例会包括基于一个或多个线的电气连接、便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦除可编程只读存储器(EPROM或快闪存储器)、光纤、便捷式紧凑盘只读存储器(CD-ROM)、光学储存设备、磁储存设备、或上述内容的任何合适组合。
需要说明的是，本公开中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其他实施例的不同之处，各个实施例之间相同相似部分互相参见即可。对于实施例公开的系统或装置而言，由于其与实施例公开的方法相对应，所以描述的比较简单，相关之处参见方法部分说明即可。
应当理解，在本公开中，“至少一个(项)”是指一个或者多个，“多个”是指两个或两个以上。“和/或”，用于描述关联对象的关联关系，表示可以存在三种关系，例如，“A和/或B”可以表示：只存在A，只存在B以及同时存在A和B三种情况，其中A，B可以是单数或者复数。字符“/”一般表示前后关联对象是一种“或”的关系。“以下至少一项(个)”或其类似表达，是指这些项中的任意组合，包括单项(个)或复数项(个)的任意组合。例如，a，b或c中的至少一项(个)，可以表示：a，b，c，“a和b”，“a和c”，“b和c”，或“a和b和c”，其中a，b，c可以是单个，也可以是多个。
还需要说明的是，在本文中，诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，术语“包括”、 “包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
结合本文中所公开的实施例描述的方法或算法的步骤可以直接用硬件、处理器执行的软件模块，或者二者的结合来实施。软件模块可以置于随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
对所公开的实施例的上述说明，使本领域专业技术人员能够实现或使用本公开。对这些实施例的多种修改对本领域的专业技术人员来说将是显而易见的，本文中所定义的一般原理可以在不脱离本公开的精神或范围的情况下，在其它实施例中实现。因此，本公开将不会被限制于本文所示的这些实施例，而是要符合与本文所公开的原理和新颖特点相一致的最宽的范围。

Claims (10)

  1. 一种数据处理方法，包括：
    获取图像序列的关键点预测序列以及所述图像序列对应的关键点标签序列；所述关键点预测序列包括至少一个关键点的预测轨迹；所述关键点标签序列包括所述至少一个关键点的标签轨迹；
    对于任一所述关键点，确定所述关键点的预测轨迹与所述关键点的标签轨迹之间的轨迹相似表征数据；所述轨迹相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度；
    依据所述轨迹相似表征数据，确定所述关键点预测序列对应的关键点预测性能。
  2. 根据权利要求1所述的方法，其中，所述图像序列包括第一图像和第二图像，所述第一图像在所述图像序列中对应的时序早于所述第二图像在所述图像序列中对应的时序，所述第一图像在所述图像序列中对应的时序与所述第二图像在所述图像序列中对应的时序相邻；
    所述关键点的预测轨迹包括所述关键点在所述第一图像下对应的预测位置和所述关键点在所述第二图像下对应的预测位置；
    所述关键点的标签轨迹包括所述关键点在所述第一图像下对应的标签位置和所述关键点在所述第二图像下对应的标签位置；
    所述轨迹相似表征数据的确定过程，包括：
    计算所述关键点在所述第二图像下对应的预测位置与所述关键点在所述第一图像下对应的预测位置之间的距离，得到轨迹段预测长度；
    计算所述关键点在所述第二图像下对应的标签位置与所述关键点在所述第一图像下对应的标签位置之间的距离，得到轨迹段标签长度；
    根据所述轨迹段预测长度与所述轨迹段标签长度之间的差值，确定长度相似表征数据；
    依据所述长度相似表征数据，确定所述轨迹相似表征数据。
  3. 根据权利要求1所述的方法，其中，所述图像序列包括第一图像和第二图像，所述第一图像在所述图像序列中对应的时序早于所述第二图像在所 述图像序列中对应的时序，所述第一图像在所述图像序列中对应的时序与所述第二图像在所述图像序列中对应的时序相邻；
    所述关键点的预测轨迹包括所述关键点在所述第一图像下对应的预测位置和所述关键点在所述第二图像下对应的预测位置；
    所述关键点的标签轨迹包括所述关键点在所述第一图像下对应的标签位置和所述关键点在所述第二图像下对应的标签位置；
    所述轨迹相似表征数据的确定过程，包括：
    依据所述关键点在所述第二图像下对应的预测位置与所述关键点在所述第一图像下对应的预测位置之间的方向表征数据，确定轨迹段预测方向表征数据；
    依据所述关键点在所述第二图像下对应的标签位置与所述关键点在所述第一图像下对应的标签位置之间的方向表征数据，确定轨迹段标签方向表征数据；
    根据所述轨迹段预测方向表征数据和所述轨迹段标签方向表征数据，确定所述方向相似表征数据；
    依据所述方向相似表征数据，确定所述轨迹相似表征数据。
  4. 根据权利要求1所述的方法，其中，所述轨迹相似表征数据包括长度相似表征数据和/或方向相似表征数据；所述长度相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹在轨迹长度上所呈现的相似程度；所述方向相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹在轨迹方向上所呈现的相似程度。
  5. 根据权利要求1所述的方法，其中，所述依据所述轨迹相似表征数据，确定所述关键点预测序列对应的关键点预测性能，包括：
    依据所述轨迹相似表征数据，确定所述关键点预测序列对应的预测稳定性表征数据；
    根据所述关键点预测序列对应的预测稳定性表征数据，确定所述关键点预测序列对应的关键点预测性能。
  6. 根据权利要求5所述的方法，其中，所述根据所述关键点预测序列对应的预测稳定性表征数据，确定所述关键点预测序列对应的关键点预测性能之前，所述方法还包括：
    根据所述关键点预测序列和所述关键点标签序列，确定所述关键点预测序列对应的预测精确性表征数据；
    所述根据所述关键点预测序列对应的预测稳定性表征数据，确定所述关键点预测序列对应的关键点预测性能，包括：
    根据所述关键点预测序列对应的预测稳定性表征数据以及所述关键点预测序列对应的预测精确性表征数据，确定所述关键点预测序列对应的关键点预测性能。
  7. 根据权利要求1-6任一项所述的方法，其中，所述获取图像序列的关键点预测序列，包括：
    利用关键点预测模型对所述图像序列进行关键点预测处理，得到所述图像序列的关键点预测序列；
    所述确定所述关键点预测序列对应的关键点预测性能之后，所述方法还包括：
    依据所述关键点预测序列对应的关键点预测性能，对所述关键点预测模型进行更新。
  8. 一种数据处理装置，包括：
    数据获取单元，被配置为获取图像序列的关键点预测序列以及所述图像序列对应的关键点标签序列；所述关键点预测序列包括至少一个关键点的预测轨迹；所述关键点标签序列包括所述至少一个关键点的标签轨迹；
    第一确定单元，被配置为对于任一所述关键点，确定所述关键点的预测轨迹与所述关键点的标签轨迹之间的轨迹相似表征数据；所述轨迹相似表征数据用于表示所述关键点的预测轨迹与所述关键点的标签轨迹之间在轨迹长度和/或轨迹方向上所呈现的相似程度；
    第二确定单元，被配置为依据所述轨迹相似表征数据，确定所述关键点预测序列对应的关键点预测性能。
  9. 一种电子设备，包括：处理器和存储器；
    所述存储器，被配置为存储指令或计算机程序；
    所述处理器，被配置为执行所述存储器中的所述指令或计算机程序，以使得所述电子设备执行权利要求1-7任一项所述的方法。
  10. 一种计算机可读介质，其中，所述计算机可读介质中存储有指令或计 算机程序，当所述指令或计算机程序在设备上运行时，使得所述设备执行权利要求1-7任一项所述的方法。
PCT/CN2024/099890 2023-07-04 2024-06-18 一种数据处理方法、装置、电子设备、计算机可读介质 Ceased WO2025007739A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202310816266.2A CN119273710A (zh) 2023-07-04 2023-07-04 一种数据处理方法、装置、电子设备、计算机可读介质
CN202310816266.2 2023-07-04

Publications (1)

Publication Number Publication Date
WO2025007739A1 true WO2025007739A1 (zh) 2025-01-09

Family

ID=94120164

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/099890 Ceased WO2025007739A1 (zh) 2023-07-04 2024-06-18 一种数据处理方法、装置、电子设备、计算机可读介质

Country Status (2)

Country Link
CN (1) CN119273710A (zh)
WO (1) WO2025007739A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108038880A (zh) * 2017-12-20 2018-05-15 百度在线网络技术（北京）有限公司 用于处理图像的方法和装置
CN111178308A (zh) * 2019-12-31 2020-05-19 北京奇艺世纪科技有限公司 一种手势轨迹的识别方法和装置
CN113627334A (zh) * 2021-08-10 2021-11-09 浙江大华技术股份有限公司 对象的行为识别方法及装置
CN115170998A (zh) * 2022-05-12 2022-10-11 无锡范特智能科技有限公司 人体行为动作识别方法、装置、计算机设备和存储介质
CN115601841A (zh) * 2022-11-10 2023-01-13 北京航空航天大学（Cn） 一种联合外观纹理和运动骨架的人体异常行为检测方法
US20230150550A1 (en) * 2021-11-16 2023-05-18 Waymo Llc Pedestrian behavior prediction with 3d human keypoints

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108038880A (zh) * 2017-12-20 2018-05-15 百度在线网络技术（北京）有限公司 用于处理图像的方法和装置
CN111178308A (zh) * 2019-12-31 2020-05-19 北京奇艺世纪科技有限公司 一种手势轨迹的识别方法和装置
CN113627334A (zh) * 2021-08-10 2021-11-09 浙江大华技术股份有限公司 对象的行为识别方法及装置
US20230150550A1 (en) * 2021-11-16 2023-05-18 Waymo Llc Pedestrian behavior prediction with 3d human keypoints
CN115170998A (zh) * 2022-05-12 2022-10-11 无锡范特智能科技有限公司 人体行为动作识别方法、装置、计算机设备和存储介质
CN115601841A (zh) * 2022-11-10 2023-01-13 北京航空航天大学（Cn） 一种联合外观纹理和运动骨架的人体异常行为检测方法

Also Published As

Publication number Publication date
CN119273710A (zh) 2025-01-07

Similar Documents

Publication Publication Date Title
CN113470626B (zh) 一种语音识别模型的训练方法、装置及设备
CN114898154B (zh) 一种增量式目标检测方法、装置、设备及介质
CN112907628B (zh) 视频目标追踪方法、装置、存储介质及电子设备
CN112650790B (zh) 目标点云平面确定方法，装置，电子设备及存储介质
CN113240108B (zh) 模型训练方法、装置和电子设备
CN114443897B (zh) 一种视频推荐方法、装置、电子设备和存储介质
CN113033680B (zh) 视频分类方法、装置、可读介质及电子设备
CN113222050B (zh) 图像分类方法、装置、可读介质及电子设备
CN110633717A (zh) 一种目标检测模型的训练方法和装置
CN112418233B (zh) 图像处理方法、装置、可读介质及电子设备
CN111680754B (zh) 图像分类方法、装置、电子设备及计算机可读存储介质
CN112183388B (zh) 图像处理方法、装置、设备和介质
WO2024183685A1 (zh) 容器搬运方法和装置
CN112990176B (zh) 书写质量评价方法、装置和电子设备
CN113051400B (zh) 标注数据确定方法、装置、可读介质及电子设备
CN114201674A (zh) 模型训练方法、推广内容的处理方法及相关装置
WO2025007739A1 (zh) 一种数据处理方法、装置、电子设备、计算机可读介质
CN113240027A (zh) 图像分类方法、装置、可读介质及电子设备
WO2024240144A1 (zh) 计算数据宽度的方法、装置、设备及存储介质
CN114625876B (zh) 作者特征模型的生成方法、作者信息处理方法和装置
CN112270170B (zh) 一种隐式表述语句的分析方法、装置、介质和电子设备
WO2024007938A1 (zh) 一种多任务预测方法、装置、电子设备及存储介质
CN116011553A (zh) 一种数据处理方法、装置、设备和存储介质
CN115272760A (zh) 适用于森林火灾烟雾检测的小样本烟雾图像细分类方法
CN118038029A (zh) 网络架构确定方法、模型构建方法、装置、设备、介质

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: 24835222

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE