WO2022012178A1 - 用于生成目标函数的方法、装置、电子设备和计算机可读介质 - Google Patents

用于生成目标函数的方法、装置、电子设备和计算机可读介质 Download PDF

Info

Publication number
WO2022012178A1
WO2022012178A1 PCT/CN2021/096144 CN2021096144W WO2022012178A1 WO 2022012178 A1 WO2022012178 A1 WO 2022012178A1 CN 2021096144 W CN2021096144 W CN 2021096144W WO 2022012178 A1 WO2022012178 A1 WO 2022012178A1
Authority
WO
WIPO (PCT)
Prior art keywords
vector
hash code
value
objective function
target
Prior art date
Application number
PCT/CN2021/096144
Other languages
English (en)
French (fr)
Inventor
何轶
王长虎
Original Assignee
北京字节跳动网络技术有限公司
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 北京字节跳动网络技术有限公司 filed Critical 北京字节跳动网络技术有限公司
Priority to US18/005,081 priority Critical patent/US20230281956A1/en
Publication of WO2022012178A1 publication Critical patent/WO2022012178A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/761—Proximity, similarity or dissimilarity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/32—Normalisation of the pattern dimensions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/72—Data preparation, e.g. statistical preprocessing of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods

Definitions

  • Embodiments of the present disclosure relate to the field of computer technology, and in particular, to a method, an apparatus, an electronic device, and a computer-readable medium for generating an objective function.
  • some embodiments of the present disclosure provide a method for generating an objective function, the method comprising: normalizing a vector corresponding to each pixel in a target feature map set to generate a target vector, obtaining target vector set; generate a hash code corresponding to each vector in the above target vector set, and obtain a hash code set; determine the prior probability of occurrence of each hash code in the above hash code set; Entropy generates the objective function.
  • some embodiments of the present disclosure provide an apparatus for generating an objective function, the apparatus comprising: a normalization operation unit configured to normalize a vector corresponding to each pixel in the target feature map set processing to generate a target vector to obtain a target vector set; a first generating unit, configured to generate a hash code corresponding to each vector in the above-mentioned target vector set, to obtain a hash code set; a first determining unit, configured to determine A priori probability of occurrence of each hash code in the above hash code set; the second generating unit is configured to generate an objective function based on the entropy of each prior probability.
  • some embodiments of the present disclosure provide an electronic device, the network device includes: one or more processors; a storage device for storing one or more programs; A plurality of processors execute such that one or more processors implement a method as described in any implementation of the first aspect.
  • an embodiment of the present application provides a computer-readable medium on which a computer program is stored, and when the computer program is executed by a processor, implements the method described in any implementation manner of the first aspect.
  • One of the foregoing embodiments of the present disclosure has the following beneficial effects: normalizing the vector corresponding to each pixel in the target feature map set to generate a target vector to obtain a target vector set;
  • the hash code corresponding to each vector in the set is obtained, and the hash code set is obtained;
  • the image features are represented by the hash code, which provides a basis for determining the objective function in the next step.
  • the above process obtains the prior probability of each hash code on the basis of hash coding.
  • the prior probability can represent the distribution of image features, and the objective function is determined based on each prior probability.
  • the above objective function can make the image distribution more accurate. Uniform, with better discrimination.
  • FIG. 1 is a schematic diagram of an application scenario of a method for generating an objective function according to some embodiments of the present disclosure
  • FIG. 2 is a flowchart of some embodiments of a method for generating an objective function according to the present disclosure
  • FIG. 3 is a schematic diagram of another application scenario of the method for generating an objective function according to some embodiments of the present disclosure
  • FIG. 4 is a flowchart of further embodiments of methods for generating an objective function according to the present disclosure
  • FIG. 5 is a schematic structural diagram of some embodiments of an apparatus for generating an objective function according to the present disclosure
  • FIG. 6 is a schematic structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure.
  • FIG. 1 is a schematic diagram 100 of an application scenario of a method for generating an objective function according to some embodiments of the present disclosure.
  • the computing device 101 normalizes the vector corresponding to each pixel in the target feature map set 102 to generate the target vector set 103 .
  • the target feature map in the above target feature map set 102 may be a feature map obtained by extracting features of any target image through a feature extraction network.
  • the target image may be an image showing a dog.
  • the feature map thus contains contour maps of the main features of the dog's eyes, nose, ears, mouth, tail, etc.
  • the vector corresponding to each pixel in the above feature map can be normalized to generate a target vector.
  • the target vector set 103 is composed of target vectors corresponding to each feature map in the above target feature map set 102 .
  • 103 in the above target vector set may include the first target vector [0.1, 0.7, 0.8], the second target vector [0.2, 0.5, 0.6], ..., the nth target vector [ 0.7, 0.3, 0.9].
  • the hash code corresponding to each target vector in the above target vector set 103 is determined, and a hash code set 104 is obtained.
  • the hash codes corresponding to the first target vector, the second target vector, and the n-th target vector in the above target vector set 103 are [0, 1, 1], [0, 1, 1], [ 1, 0, 1]).
  • a priori probability 105 of occurrence of each hash code in the set 104 of hash codes described above is determined.
  • the target vectors P 15 , P 25 , and P 35 corresponding to the hash code are 0.7 respectively. , 0.3, 0.9.
  • the number of hash codes in the hash code set is 100, so it can be obtained that the prior probability of the occurrence of the fifth hash code is 0.023).
  • the objective function 106 is generated based on the prior probability of occurrence of each hash code.
  • the objective function 106 can be obtained by multiplying the prior probability of occurrence of each hash code in the above hash code set by the logarithmic value of the prior probability, and then subtracting the inverse number.
  • the prior probability of the first hash code is 0.2
  • the prior probability of the second hash code is 0.7
  • the prior probability of the third hash code is 0.7.
  • the test probability is 0.1. It can be obtained that the entropy of the above hash code set is -0.36).
  • the execution body of the method for generating the target function can be various software, it can be the computing device 101, or it can also be a server, and the execution body of the above method can also include the above-mentioned computing device 101 and the above-mentioned server through the network. integrated equipment.
  • the computing device 101 may be various electronic devices with information processing capabilities, including but not limited to smart phones, tablet computers, e-book readers, laptop computers, desktop computers, and the like.
  • the execution body of the method for generating the objective function is software, it can be installed in the electronic devices listed above. It can be implemented, for example, as multiple software or software modules for providing distributed services, or as a single software or software module. There is no specific limitation here.
  • FIG. 1 the number of terminal devices in FIG. 1 is merely illustrative. There can be any number of terminal devices according to implementation needs.
  • the method for generating an objective function includes the following steps:
  • Step 201 normalize the vector corresponding to each pixel in the target feature map set to generate a target vector, and obtain a target vector set.
  • the execution body of the method for generating the objective function may perform normalization processing on the vector corresponding to each pixel in the target feature map to obtain the target vector set.
  • the vector value in the above target vector is a number between 0 and 1.
  • the normalization operation is mainly for data processing, which maps the data value to the range of 0 to 1, which makes it more convenient to process the data.
  • an object feature map can be composed of many pixels.
  • each pixel can be represented as a 3-dimensional vector.
  • the 3-dimensional vector of each pixel is normalized to obtain a 3-dimensional target vector, and the vector value in the above 3-dimensional target vector will become a number between 0 and 1.
  • Step 202 Generate a hash code corresponding to each vector in the above target vector set to obtain a hash code set.
  • the above-mentioned execution body may determine a hash code corresponding to each target vector in the above-mentioned target vector set to obtain a hash code set.
  • the above vector is represented as a hash code by the following formula to obtain a set of hash codes.
  • the formula can be specifically expressed as:
  • C represents the coded value of the hash code corresponding to each position
  • i represents the ith dimension of the hash code corresponding to the target vector
  • C i represents the coded value of the hash code corresponding to the ith dimension of the above hash code
  • P represents the vector value in the target vector.
  • P i represents the vector value corresponding to the i-th dimension vector in the target vector
  • the i-th dimension value in the above target vector represents the probability that the i-th bit of the corresponding hash code takes 1.
  • a certain pixel in the image is represented as a three-dimensional vector.
  • the three vector values in the above three-dimensional vector are [20, 30, 40].
  • the target vector P is obtained, and the vector value in the vector P is [0, 0.5, 1], that is, P 1 , P 2 , and P 3 are 0, 0.5, and 1, respectively.
  • the hash code C[0, 1, 1] can be obtained, where the above hash code
  • the encoded values C 1 , C 2 , and C 3 of hash coding in C are 0, 1, 1.
  • Step 203 Determine a priori probability of occurrence of each hash code in the above hash code set.
  • the execution body of the above method may determine the prior probability of occurrence of each hash code in the above set of hash codes in various ways.
  • the prior probability of occurrence of each hash code is obtained by the following steps: adding the preset first multiplication result and the preset second multiplication result to obtain the addition result.
  • a mean of the cumulative sums of the addition results is determined, the mean representing the a priori probability of occurrence of each hash code.
  • the preset first multiplication result and the preset second multiplication are obtained by the following steps: determining the encoded value of each dimension in each of the above hash codes and each of the above The vector value corresponding to the coded value of each dimension in the hash coding; determine the value of the vector value corresponding to the coded value of each dimension and the coded value of each of the above-mentioned dimensions respectively by subtracting 1; Multiply the encoded value of each dimension in the encoding and the corresponding vector value to obtain the above-mentioned preset first multiplication result; the above-mentioned code value of each dimension and the above-mentioned each dimension are respectively subtracted by 1 Multiply the values of the vector values corresponding to the encoded values of , to obtain the above-mentioned preset second multiplication result.
  • the above vector is represented as a hash code by the following formula to obtain a set of hash codes.
  • the formula can be specifically expressed as:
  • P(c) represents the prior probability of a certain hash code c
  • m represents the number of hash codes in the hash code set
  • i represents the ith dimension of the hash code corresponding to the target vector
  • j represents the hash code
  • p ij represents the vector value corresponding to the ith dimension hash code in the jth hash code, that is, the vector value corresponding to the ith dimension vector in the above target vector
  • C i represents the encoded value of the hash code corresponding to the i-th dimension of the hash code.
  • the fifth hash code in the above hash code set is [1, 0, 1], and the encoded values c 1 , c 2 , and c 3 of the hash code are 1, 0, 1.
  • the target vector corresponding to the hash code is [0.7, 0.3, 0.9], that is, p 15 , p 25 , and p 35 are 0.7, 0.3, and 0.9, respectively.
  • the number of hash codes in the hash code set is 100, so it can be obtained that the prior probability that the fifth hash code appears is 0.023.
  • the prior probability of occurrence of each hash code in the above hash code set can be determined in turn.
  • Step 204 generating an objective function based on the entropy of each prior probability.
  • the execution body of the above method may generate the objective function based on the entropy of the prior probability of occurrence of each hash code obtained in step 203 .
  • the entropy of the above-mentioned prior probability can represent the degree of distribution of each hash code in the entire hash code set. The larger the entropy value is, the more uniform the distribution of hash codes in the entire hash code set is, that is, the difference between the corresponding feature maps is relatively large.
  • the objective function is generated by the entropy of each of the above-mentioned prior probabilities.
  • the above-mentioned objective function is obtained by the following steps: determining the logarithm value of the prior probability of occurrence of each of the above-mentioned hash codes; Multiplying the logarithmic values of the prior probabilities of occurrence of each of the above hash codes to obtain a multiplication result; determining the cumulative sum of the above multiplication results; taking the negative number of the above cumulative sum to obtain the above objective function.
  • J represents the objective function
  • i represents the index value corresponding to the above-mentioned hash code
  • p represents the prior probability
  • c represents the hash code
  • p i(c ) represents the prior probability of the occurrence of the ith hash code.
  • the objective function can be obtained by multiplying the prior probability of occurrence of each hash code in the above hash code set by the logarithmic value of the prior probability, and then removing the inverse number.
  • the prior probability of the first hash code is 0.2
  • the prior probability of the second hash code is 0.7
  • the prior probability of the third hash code is 0.7.
  • the probability is 0.1. It can be obtained that the entropy of the above hash code set is -0.36.
  • One of the foregoing embodiments of the present disclosure has the following beneficial effects: normalizing the vector corresponding to each pixel in the target feature map set to generate a target vector to obtain a target vector set;
  • the hash code corresponding to each vector in the set is obtained, and the hash code set is obtained;
  • the image features are represented by the hash code, which provides a basis for determining the objective function in the next step.
  • the above process obtains the prior probability of each hash code on the basis of hash coding.
  • the prior probability can represent the distribution of image features, and the objective function is determined based on each prior probability.
  • the above objective function can make the image distribution more accurate. Uniform, with better discrimination.
  • FIG. 3 a schematic diagram 300 of another application scenario of the method for generating an objective function according to some embodiments of the present disclosure.
  • the computing device 301 normalizes the vector corresponding to each pixel in the target feature map set 302 to generate a target vector set 303 .
  • the target feature map in the above target feature map set 302 may be a feature map obtained by extracting features from any target image through a feature extraction network.
  • the above target vector set 303 may include a first target vector [0.1, 0.7, 0.8], which may represent a dog's eyes, and a second target vector [0.2, 0.5, 0.6], which may represents the dog's nose, the third target vector [0.1, 0.8, 0.8], which can represent the dog's mouth, and the fourth target vector [0, 0.9, 0.6], which can represent the dog's tail, by
  • determine each target vector set 303 Hash codes corresponding to the target vectors to obtain the hash code set 304 for example, the target vectors [0.1, 0.7, 0.8], [0.2, 0.5, 0.6] corresponding to the eyes, nose, mouth, and tail of the dog in the above example , [0.1, 0.8, 0.8], [0, 0.9, 0.6] are converted to corresponding hash codes [0, 1, 1
  • the hash codes that meet the predetermined conditions in the above-mentioned hash code set 304 are selected as the first feature group and the second feature group 305 (for example, the hash codes corresponding to the main features of the dog's eyes, nose, mouth, etc. that meet the conditions can be selected [ 0, 1, 1], [0, 1, 1], [0, 1, 1] and the hash codes [1, 0, 1], [0, 1] corresponding to the main features of the cat's eyes, nose, mouth, etc. , 1], [0, 1, 1] as the first feature group and the second feature group).
  • the process 400 of the method for generating an objective function includes the following steps:
  • Step 401 normalize the vector corresponding to each pixel in the target feature map set to generate a target vector, and obtain a target vector set.
  • Step 402 Generate a hash code corresponding to each vector in the above target vector set to obtain a hash code set.
  • Step 403 Determine the prior probability of occurrence of each hash code in the above hash code set.
  • Step 404 generating an objective function based on the entropy of each prior probability.
  • steps 401-404 for the specific implementation of steps 401-404 and the technical effects brought about, reference may be made to steps 201-204 in those embodiments corresponding to FIG. 2, and details are not repeated here.
  • Step 405 Based on the above objective function, the above feature extraction network is optimized to obtain a trained feature extraction network.
  • the above-mentioned objective function is optimized, so that the feature extraction network can better learn the parameters of the above-mentioned feature extraction network by minimizing the negative likelihood logarithm.
  • Step 406 extract the image features of the two images based on the feature extraction network after the above training, and then obtain the hash code;
  • the trained feature extraction network performs feature extraction on two images to obtain feature maps of the two images.
  • the vector corresponding to each pixel in the above feature map is normalized to generate a target vector, and a target vector set is obtained.
  • Step 407 Determine the similarity of the above two images based on the obtained hash code.
  • the similarity of the above two images can be obtained by performing cross-comparison processing on the hash codes corresponding to the two feature maps.
  • the execution body of the above method is optimized based on the objective function generated in step 404, and the feature extraction network can better train the above feature extraction network by minimizing the negative likelihood logarithm, so that the extracted The image feature distribution is more uniform and has better discrimination.
  • the process 400 of the method for generating an objective function in some embodiments corresponding to FIG. 4 embodies the steps of optimizing the objective function. Therefore, the above feature extraction network has better network generalization ability, and effectively compares the similarity of two images.
  • the present disclosure provides some embodiments of an apparatus for generating an objective function, these apparatus embodiments correspond to those method embodiments shown in FIG. 2 , the The device can be specifically applied to various electronic devices.
  • an apparatus 500 for generating an objective function in some embodiments includes: a normalization operation unit 501 , a first generating unit 502 , a first determining unit 503 and a second generating unit 504 .
  • the normalization operation unit 501 is configured to normalize the vector corresponding to each pixel in the target feature map set to generate a target vector to obtain a target vector set;
  • the first generation unit 502 is configured to generate the target Hash codes corresponding to each vector in the vector set, to obtain a hash code set;
  • the first determining unit 503 is configured to determine the prior probability of occurrence of each hash code in the above-mentioned hash code set;
  • the second generating unit 504 is configured by is configured to generate an objective function based on the entropy of each prior probability.
  • the above target feature map set is further configured to perform feature extraction on the target image set through a feature extraction network to obtain a target feature map set.
  • the above-mentioned apparatus 500 is further configured to add the preset first multiplication result and the preset second multiplication result to obtain the addition result.
  • a mean of the cumulative sums of the addition results is determined, the mean representing the a priori probability of occurrence of each hash code.
  • the above-mentioned apparatus 500 is further configured to determine a vector value corresponding to the encoded value of each dimension in each of the above-mentioned hash codes and the encoded value of each of the above-mentioned dimensions in each of the above-mentioned hash codes value; determine the numerical value of the vector value corresponding to the above-mentioned encoded value of each dimension and the above-mentioned encoded value of each dimension by 1 respectively; The vector values are multiplied to obtain the above-mentioned preset first multiplication result; the above-mentioned values of the vector values corresponding to the encoding values of each dimension and the encoding values of the above-mentioned dimensions are multiplied by 1 respectively to obtain The above preset second multiplication result.
  • the above-mentioned apparatus 500 is further configured to determine the logarithm value of the a priori probability of occurrence of each of the above-mentioned hash codes; Multiply the logarithmic values of the prior probabilities of the hash codes to obtain the multiplication result; determine the cumulative sum of the above multiplication results; take the negative number of the above cumulative sum to obtain the above objective function.
  • the foregoing apparatus 500 further includes an optimization unit configured to optimize the foregoing feature extraction network based on the foregoing objective function to obtain a trained feature extraction network.
  • the above-mentioned apparatus 500 further includes a second determination unit configured to extract hash codes of the two images based on the above-mentioned trained feature extraction network; based on the extracted hash codes, determine similarity of the above two images.
  • the units recorded in the apparatus 500 correspond to the respective steps in the method described with reference to FIG. 2 . Therefore, the operations, features and beneficial effects described above with respect to the method are also applicable to the apparatus 500 and the units included therein, and details are not described herein again.
  • FIG. 6 a schematic structural diagram of an electronic device (eg, the computing device in FIG. 1 ) 600 suitable for implementing some embodiments of the present disclosure is shown.
  • Electronic devices in some embodiments of the present disclosure may include, but are not limited to, such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablets), PMPs (portable multimedia players), vehicle-mounted terminals Mobile terminals such as in-vehicle navigation terminals, etc., and stationary terminals such as digital TVs, desktop computers, and the like.
  • the electronic device shown in FIG. 6 is only an example, and should not impose any limitation on the function and scope of use of the embodiments of the present disclosure.
  • an electronic device 600 may include a processing device (eg, a central processing unit, a graphics processor, etc.) 601 that may be loaded into random access according to a program stored in a read only memory (ROM) 602 or from a storage device 608 Various appropriate actions and processes are executed by the programs in the memory (RAM) 603 . In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored.
  • the processing device 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604.
  • An input/output (I/O) interface 605 is also connected to bus 604 .
  • I/O interface 605 input devices 606 including, for example, a touch screen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; including, for example, a liquid crystal display (LCD), speakers, vibration An output device 607 of a computer, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609.
  • Communication means 609 may allow electronic device 600 to communicate wirelessly or by wire with other devices to exchange data. While FIG. 6 shows electronic device 600 having various means, it should be understood that not all of the illustrated means are required to be implemented or provided. More or fewer devices may alternatively be implemented or provided. Each block shown in FIG. 6 may represent one device, or may represent multiple devices as required.
  • the processes described above with reference to the flowcharts may be implemented as computer software programs.
  • some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the method illustrated in the flowchart.
  • the computer program may be downloaded and installed from a network via communication device 609, or from storage device 608, or from ROM 602.
  • the processing device 601 the above-mentioned functions defined in the methods of some embodiments of the present disclosure are performed.
  • the computer-readable medium described in some embodiments of the present disclosure 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 can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or a combination of any of the above. More specific examples of computer readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable Programmable read only memory (EPROM or flash memory), fiber optics, portable compact disk read only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
  • a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
  • a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code therein. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • a computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device .
  • Program code embodied on a computer readable medium may be transmitted using any suitable medium including, but not limited to, electrical wire, optical fiber cable, RF (radio frequency), etc., or any suitable combination of the foregoing.
  • the client and server can use any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol) to communicate, and can communicate with digital data in any form or medium Communication (eg, a communication network) interconnects.
  • HTTP HyperText Transfer Protocol
  • Examples of communication networks include local area networks (“LAN”), wide area networks (“WAN”), the Internet (eg, the Internet), and peer-to-peer networks (eg, ad hoc peer-to-peer networks), as well as any currently known or future development network of.
  • the above-mentioned computer-readable medium may be included in the above-mentioned electronic device; or may exist alone without being assembled into the electronic device.
  • the above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed by the electronic device, the electronic device: normalizes the vector corresponding to each pixel in the target feature map set Process to generate a target vector to obtain a target vector set; generate a hash code corresponding to each vector in the above-mentioned target vector set to obtain a hash code set; determine the prior probability that each hash code in the above-mentioned hash code set occurs ; Generate the objective function based on the entropy of each prior probability.
  • Computer program code for carrying out operations of some embodiments of the present disclosure may be written in one or more programming languages, including object-oriented programming languages—such as Java, Smalltalk, C++, or a combination thereof , as well as conventional procedural programming languages - such as "C" or similar programming languages.
  • the program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server.
  • the remote computer may be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (eg, using an Internet service provider to via Internet connection).
  • LAN local area network
  • WAN wide area network
  • each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more logical functions for implementing the specified functions executable instructions.
  • the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.
  • each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations can be implemented in dedicated hardware-based systems that perform the specified functions or operations , or can be implemented in a combination of dedicated hardware and computer instructions.
  • the units described in some embodiments of the present disclosure may be implemented by means of software, and may also be implemented by means of hardware.
  • the described unit may also be provided in the processor, for example, it may be described as: a processor includes a normalization operation unit, a first generation unit, a first determination unit and a second generation unit. Among them, the names of these units do not constitute a limitation of the unit itself in some cases.
  • the normalization operation unit can also be described as "normalize the vector corresponding to each pixel in the target feature map set. Process to generate a target vector, resulting in a unit of the target vector set".
  • 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 Logical Devices (CPLDs) and more.
  • FPGAs Field Programmable Gate Arrays
  • ASICs Application Specific Integrated Circuits
  • ASSPs Application Specific Standard Products
  • SOCs Systems on Chips
  • CPLDs Complex Programmable Logical Devices
  • a method for generating an objective function including: normalizing a vector corresponding to each pixel in a target feature map set to generate a target vector to obtain a target vector set; generate a hash code corresponding to each vector in the above target vector set, and obtain a hash code set; determine the prior probability of each hash code in the above hash code set; based on the entropy of each prior probability Generate the objective function.
  • the preset first multiplication result and the preset second multiplication result are added to obtain an addition result.
  • a mean of the cumulative sums of the addition results is determined, the mean representing the a priori probability of occurrence of each hash code.
  • the log value of the prior probability of occurrence of each hash code is determined; the prior probability of occurrence of each hash code and the prior probability of occurrence of each hash code are determined. Multiply the logarithmic values of the probabilities to obtain the multiplication result; determine the cumulative sum of the above multiplication results; take the negative number of the above cumulative sum to obtain the above objective function.
  • the above target feature map set is obtained through the following steps, including: performing feature extraction on the target image set through a feature extraction network to obtain the target feature map set.
  • the above feature extraction network is optimized to obtain a trained feature extraction network.
  • the hash codes of the two images are extracted based on the above-mentioned trained feature extraction network; based on the extracted hash codes, the similarity of the two images is determined.
  • an apparatus for generating an objective function comprising: a normalization operation unit configured to normalize a vector corresponding to each pixel in the target feature map set processing to generate a target vector to obtain a target vector set; a first generating unit, configured to generate a hash code corresponding to each vector in the above-mentioned target vector set, to obtain a hash code set; a first determining unit, configured to determine A priori probability of occurrence of each hash code in the above hash code set; the second generating unit is configured to generate an objective function based on the entropy of each prior probability.
  • the above-mentioned apparatus is further configured to add the preset first multiplication result and the preset second multiplication result to obtain the addition result.
  • a mean of the cumulative sums of the addition results is determined, the mean representing the a priori probability of occurrence of each hash code.
  • the above-mentioned apparatus is further configured to determine the encoded value of each dimension in each of the above-mentioned hash codes and the vector value corresponding to the above-mentioned encoded value of each dimension in each of the above-mentioned hash codes Determine the numerical value of the vector value corresponding to the coded value of each dimension and the coded value corresponding to the above-mentioned each dimension by 1 respectively;
  • the coded value of each dimension and the vector corresponding to it in each of the above-mentioned hash codes Multiply the values to obtain the above-mentioned preset first multiplication result; multiply the above-mentioned numerical values of the vector values corresponding to the encoding value of each dimension and the encoding value of the above-mentioned each dimension by 1 to obtain the above-mentioned Preset the second multiplication result.
  • the above-mentioned apparatus is further configured to determine a logarithm value of the prior probability of occurrence of each of the above-mentioned hash codes; Multiply the logarithm of the prior probability of the Greek code to obtain the multiplication result; determine the cumulative sum of the above multiplication results; take the negative number of the above cumulative sum to obtain the above objective function.
  • the above target feature map set is further configured to perform feature extraction on the target image set through a feature extraction network to obtain a target feature map set.
  • the apparatus 500 further includes an optimization unit configured to optimize the feature extraction network based on the objective function to obtain a trained feature extraction network.
  • the above-mentioned apparatus 500 further includes a second determination unit configured to extract hash codes of the two images based on the above-mentioned trained feature extraction network; based on the extracted hash codes, determine similarity of the above two images.

Landscapes

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

Abstract

本公开的实施例公开了用于生成目标函数的方法、装置、设备和计算机可读介质。该方法的一具体实施方式包括：将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；将图像特征用哈希编码来表示，为下一步确定目标函数提供了基础。确定上述哈希编码集合中每个哈希编码出现的先验概率；基于各个先验概率的熵生成目标函数。上述过程在哈希编码的基础上得到每个哈希编码的先验概率，上述先验概率可以表示图像特征的分布情况，基于各个先验概率确定目标函数，通过上述目标函数可以使图像分布更加均匀，有更好的区分度。

Description

用于生成目标函数的方法、装置、电子设备和计算机可读介质
相关申请的交叉引用
本申请要求于2020年07月16日提交的、申请号为202010685574.2、发明名称为“用于生成目标函数的方法、装置、电子设备和计算机可读介质”的中国专利申请的优先权，该申请的全文通过引用结合在本申请中。
技术领域
本公开的实施例涉及计算机技术领域，具体涉及用于生成目标函数的方法、装置、电子设备和计算机可读介质。
背景技术
随着互联网的发展和以深度学习为核心的人工智能技术的普及，计算机视觉技术应用到人们生活的各个领域，现有图像特征提取网络提取到的图像特征存在图像分布不均匀，没有很好的区分度等问题。
发明内容
本公开的内容部分用于以简要的形式介绍构思，这些构思将在后面的具体实施方式部分被详细描述。本公开的内容部分并不旨在标识要求保护的技术方案的关键特征或必要特征，也不旨在用于限制所要求的保护的技术方案的范围。本公开的一些实施例提出了用于生成目标函数的方法、装置、设备和计算机可读介质，来解决以上背景技术部分提到的技术问题。
第一方面，本公开的一些实施例提供了一种用于生成目标函数的方法，该方法包括：将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；确定上述哈 希编码集合中每个哈希编码出现的先验概率；基于各个先验概率的熵生成目标函数。
第二方面，本公开的一些实施例提供了一种用于生成目标函数的装置，装置包括：归一操作单元，被配置成将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；第一生成单元，被配置成生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；第一确定单元，被配置成确定上述哈希编码集合中每个哈希编码出现的先验概率；第二生成单元，被配置成基于各个先验概率的熵生成目标函数。
第三方面，本公开的一些实施例提供了一种电子设备，该网络设备包括：一个或多个处理器；存储装置，用于存储一个或多个程序；当一个或多个程序被一个或多个处理器执行，使得一个或多个处理器实现如第一方面中任一实现方式描述的方法。
第四方面，本申请实施例提供了一种计算机可读介质，其上存储有计算机程序，该计算机程序被处理器执行时实现如第一方面中任一实现方式描述的方法。
本公开的上述各个实施例中的一个实施例具有如下有益效果：将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；将图像特征用哈希编码来表示，为下一步确定目标函数提供了基础。确定上述哈希编码集合中每个哈希编码出现的先验概率；基于各个先验概率的熵生成目标函数。上述过程在哈希编码的基础上得到每个哈希编码的先验概率，上述先验概率可以表示图像特征的分布情况，基于各个先验概率确定目标函数，通过上述目标函数可以使图像分布更加均匀，有更好的区分度。
附图说明
结合附图并参考以下具体实施方式，本公开各实施例的上述和其他特征、优点及方面将变得更加明显。贯穿附图中，相同或相似的附 图标记表示相同或相似的元素。应当理解附图是示意性的，原件和元素不一定按照比例绘制。
图1是根据本公开的一些实施例的用于生成目标函数的方法的一个应用场景的示意图；
图2是根据本公开的用于生成目标函数的方法的一些实施例的流程图；
图3是根据本公开的一些实施例的用于生成目标函数的方法的另一个应用场景的示意图；
图4是根据本公开的用于生成目标函数的方法的另一些实施例的流程图；
图5是根据本公开的用于生成目标函数的装置的一些实施例的结构示意图；
图6是适于用来实现本公开的一些实施例的电子设备的结构示意图。
具体实施方式
下面将参照附图更详细地描述本公开的实施例。虽然附图中显示了本公开的某些实施例，然而应当理解的是，本公开可以通过各种形式来实现，而且不应该被解释为限于这里阐述的实施例。相反，提供这些实施例是为了更加透彻和完整地理解本公开。应当理解的是，本公开的附图及实施例仅用于示例性作用，并非用于限制本公开的保护范围。
另外还需要说明的是，为了便于描述，附图中仅示出了与有关发明相关的部分。在不冲突的情况下，本公开中的实施例及实施例中的特征可以相互组合。
需要注意，本公开中提及的“第一”、“第二”等概念仅用于对不同的装置、模块或单元进行区分，并非用于限定这些装置、模块或单元所执行的功能的顺序或者相互依存关系。
需要注意，本公开中提及的“一个”、“多个”的修饰是示意性而非限制性的，本领域技术人员应当理解，除非在上下文另有明确指出， 否则应该理解为“一个或多个”。
本公开实施方式中的多个装置之间所交互的消息或者信息的名称仅用于说明性的目的，而并不是用于对这些消息或信息的范围进行限制。
下面将参考附图并结合实施例来详细说明本公开。
图1是根据本公开一些实施例的用于生成目标函数的方法的一个应用场景的示意图100。
如图1所示，计算设备101对目标特征图集合102中与每个像素对应的向量进行归一化处理以生成目标向量集合103。例如，上述目标特征图集合102中目标特征图可以是任意目标图像通过特征提取网络提取特征得到的特征图。目标图像可以是一张显示有狗的图像。从而特征图包含狗的眼睛、鼻子、耳朵、嘴巴、尾巴等主要特征的轮廓图。在此基础上，可以将上述特征图中每个像素对应的向量进行归一化处理以生成目标向量。上述目标特征图集合102中各个特征图对应的目标向量组成目标向量集合103。作为示例，上述目标向量集合中103可以包括第一个目标向量[0.1，0.7，0.8]，第二个目标向量[0.2，0.5，0.6]，......，第n个目标向量[0.7，0.3，0.9]。
确定与上述目标向量集合103中每个目标向量对应的哈希编码，得到哈希编码集合104。(例如，上述目标向量集合103中第一个目标向量，第二个目标向量，第n个目标向量对应的哈希编码分别为[0，1，1]，[0，1，1]，[1，0，1])。
确定上述哈希编码集合104中每个哈希编码出现的先验概率105。作为示例，确定上述哈希编码集合104中某个哈希编码每个维度的编码值和目标向量值，若某一维度的编码值为0，可以转换为1减去0，也就是1，若某一维度的目标向量值小于0.5，可以转换为1减去该目标向量值。然后将每个维度的编码值和目标向量值进行相乘再累加求和，就可以得到上述某个哈希编码出现的先验概率。(哈希编码集合104中第5个哈希编码的编码值C 1，C 2，C 3是1，0，1。该哈希编码对应的目标向量P 15，P 25，P 35分别是0.7，0.3，0.9。哈希编码集合中哈希编码的数目为100，由此可以得到第5个哈希编码出现的先验概率是0.023)。
基于各个哈希编码出现的先验概率生成目标函数106。作为示例，将上述哈希编码集合中每个哈希编码出现的先验概率乘以该先验概率的对数值累加求和，然后去相反数，就可以得到目标函数106。(例如，上述哈希编码集合中有3个哈希编码，第1个哈希编码的先验概率为0.2，第2个哈希编码的先验概率为0.7，第3个哈希编码的先验概率为0.1。可以得到上述哈希编码集合的熵为-0.36)。
可以理解的是，用于生成目标函数的方法的执行主体可以是各种软件，可以是计算设备101，或者也可以是服务器，上述方法的执行主体还可以包括上述计算设备101与上述服务器通过网络相集成所构成的设备。其中，计算设备101可以是具有信息处理能力的各种电子设备，包括但不限于智能手机、平板电脑、电子书阅读器、膝上型便携计算机和台式计算机等等。当用于生成目标函数的方法的执行主体为软件时，可以安装在上述所列举的电子设备中。其可以实现成例如用来提供分布式服务的多个软件或软件模块，也可以实现成单个软件或软件模块。在此不做具体限定。
应该理解，图1中的终端设备数目仅仅是示意性的。根据实现需要，可以具有任意数目的终端设备。
继续参考图2，示出了根据本公开的用于生成目标函数的方法的一些实施例的流程200。该用于生目标函数的方法，包括以下步骤：
步骤201，将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合。
在一些实施例中，用于生成目标函数的方法的执行主体(例如图1所示的计算设备101)可以对目标特征图中每个像素对应的向量进行归一化处理，得到目标向量集合。其中，上述目标向量中的向量值为0到1之间的数。归一化操作主要是为了数据处理，把数据值映射到0到1范围之内，进而更加方便处理数据。例如，一幅目标特征图可以是由许多像素组成。其中，每个像素又可以表示为一个3维向量。对每个像素点的3维向量进行归一化操作，得到3维目标向量，上述3维目标向量 中的向量值会变成0到1之间的数。
作为示例，将目标特征图中某个像素对应的多维向量进行归一化，将上述多维向量中每个维度的向量值减去多维向量中最小的向量值，然后除以多维向量里面最大的向量值和最小的向量值的差，得到目标向量。
步骤202，生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合。
在一些实施例中，基于步骤201中得到的目标向量集合，上述执行主体可以确定与上述目标向量集合中每个目标向量对应的哈希编码，得到哈希编码集合。作为示例，通过如下公式将上述向量表示为哈希编码，得到哈希编码集合。其中，公式可以具体表达为：
Figure PCTCN2021096144-appb-000001
其中，C表示每个位置对应的哈希编码的编码值，i表示目标向量对应的哈希编码的第i维，C i表示上述哈希编码第i维所对应的哈希编码的编码值，P表示目标向量中的向量值。P i表示目标向量中第i维向量所对应的向量值，上述目标向量中第i维的值表示的是所对应哈希编码第i位取1的概率。
作为示例，将图像中某一个像素表示为三维向量，上述三维向量中的三个向量值是[20，30，40]，经过归一化之后就得到目标向量P，向量P中的向量值是[0，0.5，1]，也就是P 1，P 2，P 3分别是0，0.5，1，根据上述公式就可以得到哈希编码C[0，1，1]，其中，上述哈希编码C中哈希编码的编码值C 1，C 2，C 3是0，1，1。
步骤203，确定上述哈希编码集合中每个哈希编码出现的先验概率。
在一些实施例中，上述方法的执行主体可以通过各种方式确定上述哈希编码集合中每个哈希编码出现的先验概率。
在一些实施例的可选的实现方式中，上述每个哈希编码出现的先验概率通过以下步骤得到：将预设第一相乘结果和预设第二相乘结果相加，得到相加结果。确定相加结果的累加之和的均值，所述均值表示每个哈希编码出现的先验概率。
在一些实施例的可选的实现方式中，上述预设第一相乘结果和预 设第二相乘通过以下步骤得到：确定上述每个哈希编码中每个维度的编码值和上述每个哈希编码中每个维度的编码值对应的向量取值；确定通过1分别减去上述每个维度的编码值和上述每个维度的编码值对应的向量取值的数值；将上述每个哈希编码中每个维度的编码值和与之对应的向量取值相乘，得到上述预设第一相乘结果；将上述通过1分别减去所述每个维度的编码值和上述每个维度的编码值对应的向量取值的数值相乘，得到上述预设第二相乘结果。
作为示例，通过如下公式将上述向量表示为哈希编码，得到哈希编码集合。其中，公式可以具体表达为：
Figure PCTCN2021096144-appb-000002
其中，P(c)表示某个哈希编码c出现的先验概率，m表示哈希编码集合中哈希编码的数目，i表示目标向量对应的哈希编码的第i维，j表示哈希编码集合中第j个哈希编码，p ij表示第j个哈希编码中第i维的哈希编码所对应的向量值，也就是上述目标向量中第i维向量所对应的向量值，C i表示上述哈希编码第i维所对应的哈希编码的编码值。
作为示例，上述哈希编码集合中第5个哈希编码为[1，0，1]，哈希编码的编码值c 1，c 2，c 3是1，0，1。该哈希编码对应的目标向量为[0.7，0.3，0.9]，也就是p 15，p 25，p 35分别是0.7，0.3，0.9。哈希编码集合中哈希编码的数目为100，由此可以得到第5个哈希编码出现的先验概率是0.023。依次可以确定上述哈希编码集合中每个哈希编码出现的先验概率。
步骤204，基于各个先验概率的熵生成目标函数。
在一些实施例中，上述方法的执行主体可以基于步骤203中得到的各个哈希编码出现的先验概率的熵生成目标函数。上述先验概率的熵可以表示整个哈希编码集合的每个哈希编码的分布程度。熵值越大，说明整个哈希编码集合中哈希编码分布越均匀，也就是对应的特征图差异比较大。通过上述各个先验概率的熵生成目标函数。
在一些实施例中的可选的实现方式中，上述目标函数通过以下步骤得到：确定上述每个哈希编码出现的先验概率的对数值；将上述每个哈希编码出现的先验概率和上述每个哈希编码出现的先验概率的对数值相乘，得到相乘结果；确定上述相乘结果的累加之和；取上述累 加之和的负数，得到上述目标函数。
作为示例，上述目标函数是通过以下公式得到。具体公式如下：
Figure PCTCN2021096144-appb-000003
其中，J表示目标函数，i＝1表示i是哈希编码对应的索引值为1，i表示上述哈希编码对应的索引值，p表示先验概率，c表示哈希编码，p i(c)表示第i个哈希编码出现的先验概率。
作为示例，将上述哈希编码集合中每个哈希编码出现的先验概率乘以该先验概率的对数值累加求和，然后去相反数，就可以得到目标函数。例如，上述哈希编码集合中有3个哈希编码，第1个哈希编码的先验概率为0.2，第2个哈希编码的先验概率为0.7，第3个哈希编码的先验概率为0.1。可以得到上述哈希编码集合的熵为-0.36。
本公开的上述各个实施例中的一个实施例具有如下有益效果：将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；将图像特征用哈希编码来表示，为下一步确定目标函数提供了基础。确定上述哈希编码集合中每个哈希编码出现的先验概率；基于各个先验概率的熵生成目标函数。上述过程在哈希编码的基础上得到每个哈希编码的先验概率，上述先验概率可以表示图像特征的分布情况，基于各个先验概率确定目标函数，通过上述目标函数可以使图像分布更加均匀，有更好的区分度。
进一步参考图3，根据本公开一些实施例的用于生成目标函数的方法的另一个应用场景的示意图300。
如图3所示，计算设备301对目标特征图集合302中与每个像素对应的向量进行归一化处理以生成目标向量集合303。例如，上述目标特征图集合302中目标特征图可以是任意目标图像通过特征提取网络提取特征得到的特征图。(例如，上述目标向量集合303中可以包括第一个目标向量[0.1，0.7，0.8]，该目标向量可以表示狗的眼睛，第二个目标向量[0.2，0.5，0.6]，该目标向量可以表示狗的鼻子，第三个目标向量[0.1，0.8，0.8]，该目标向量可以表示狗的嘴巴，第四个目标向量[0， 0.9，0.6]，该目标向量可以表示狗的尾巴，由此得到一个表示一副目标特征图主要信息的目标向量集合，上述目标特征图集合包括多副目标特征图，所以目标向量集合也表示多副目标特征图信息)，确定上述目标向量集合303中每个目标向量对应的哈希编码，得到哈希编码集合304(例如，将上述例子中狗的眼睛，鼻子，嘴巴，尾巴对应的目标向量[0.1，0.7，0.8]，[0.2，0.5，0.6]，[0.1，0.8，0.8]，[0，0.9，0.6]转换为对应的哈希编码[0，1，1]，[0，1，1]，[0，1，1]，[0，1，1])。选择上述哈希编码集合304中符合预定条件的哈希编码作为第一特征组和第二特征组305(例如，可以选择符合条件的狗的眼睛，鼻子，嘴巴等主要特征对应的哈希编码[0，1，1]，[0，1，1]，[0，1，1]和猫的眼睛，鼻子，嘴巴等主要特征对应的哈希编码[1，0，1]，[0，1，1]，[0，1，1]作为第一特征组和第二特征组)。确定第一特征组和第二特征组的交并比，进而确定第一图像和第二图像的相似度306。(例如，上述第一特征组和第二特征组的相似度为0.67)。
进一步参考图4，其示出了用于生成目标函数的方法的另一些实施例的流程400。该用于生成目标函数的方法的流程400，包括以下步骤：
步骤401，将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合。
步骤402，生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合。
步骤403，确定上述哈希编码集合中每个哈希编码出现的先验概率。
步骤404，基于各个先验概率的熵生成目标函数。
在一些实施例中，步骤401-404的具体实现及所带来的技术效果可以参考图2对应的那些实施例中的步骤201-204，在此不再赘述。
步骤405，基于上述目标函数，对上述特征提取网络进行优化，得到训练后的特征提取网络。
在一些实施例中，对上述目标函数进行优化，通过负似然对数最小化进而使得特征提取网络能够更好的学习上述特征提取网络的参 数。
步骤406，基于上述训练后的特征提取网络提取两幅图像的图像特征，进而得到哈希编码；
在一些实施例中，将训练好的特征提取网络对两幅图像进行特征提取，得到两幅图像的特征图。将上述特征图中每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合。生成与上述目标向量集合中每个向量对应的哈希编码。
步骤407，基于所得到的哈希编码，确定上述两幅图像的相似度。
在一些实施例中，将两幅特征图对应的哈希编码进行交并比处理，就可以得到上述两幅图像的相似度。
在一些实施例中，上述方法的执行主体基于步骤404生成的目标函数进行优化，通过负似然对数最小化进而使得特征提取网络能够更好的训练上述特征提取网络，从而能够使提取到的图像特征分布更加均匀，有更好的区分度。
从图4中可以看出，与图2对应的一些实施例的描述相比，图4对应的一些实施例中的生成目标函数的方法的流程400体现了目标函数进行优化的步骤。由此，使得上述特征提取网络有更好的网络泛化能力，有效的比较两幅图像的相似度。
进一步参考图5，作为对上述各图所示方法的实现，本公开提供了一种生成目标函数的装置的一些实施例，这些装置实施例与图2所示的那些方法实施例相对应，该装置具体可以应用于各种电子设备中。
如图5所示，一些实施例的生成目标函数的装置500包括：归一操作单元501、第一生成单元502、第一确定单元503和第二生成单元504。其中，归一操作单元501被配置成将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；第一生成单元502被配置成生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；第一确定单元503被配置成确定上述哈希编码集合中每个哈希编码出现的先验概率；第二生成单元504被配置成基于各个先验概率的熵生成目标函数。
在一些实施例的可选实现方式中，上述目标特征图集合进一步被配置成通过特征提取网络对目标图像集合进行特征提取，得到目标特征图集合。
在一些实施例的可选实现方式中，上述装置500进一步被配置成将预设第一相乘结果和预设第二相乘结果相加，得到相加结果。确定相加结果的累加之和的均值，所述均值表示每个哈希编码出现的先验概率。
在一些实施例的可选实现方式中，上述装置500进一步被配置成确定上述每个哈希编码中每个维度的编码值和上述每个哈希编码中每个维度的编码值对应的向量取值；确定通过1分别减去上述每个维度的编码值和上述每个维度的编码值对应的向量取值的数值；将上述每个哈希编码中每个维度的编码值和与之对应的向量取值相乘，得到上述预设第一相乘结果；将上述通过1分别减去所述每个维度的编码值和上述每个维度的编码值对应的向量取值的数值相乘，得到上述预设第二相乘结果。
在一些实施例的可选实现方式中，上述装置500进一步被配置成确定上述每个哈希编码出现的先验概率的对数值；将上述每个哈希编码出现的先验概率和上述每个哈希编码出现的先验概率的对数值相乘，得到相乘结果；确定上述相乘结果的累加之和；取上述累加之和的负数，得到上述目标函数。
在一些实施例的可选实现方式中，上述装置500还包括优化单元，被配置成基于上述目标函数，对上述特征提取网络进行优化，得到训练后的特征提取网络。
在一些实施例的可选实现方式中，上述装置500还包括第二确定单元，被配置成基于上述训练后的特征提取网络提取两幅图像的哈希编码；基于所提取的哈希编码，确定上述两幅图像的相似度。
可以理解的是，该装置500中记载的诸单元与参考图2描述的方法中的各个步骤相对应。由此，上文针对方法描述的操作、特征以及产生的有益效果同样适用于装置500及其中包含的单元，在此不再赘述。
下面参考图6，其示出了适于用来实现本公开的一些实施例的电子设备(例如图1中的计算设备)600的结构示意图。本公开的一些实施例中的电子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6示出的电子设备仅仅是一个示例，不应对本公开的实施例的功能和使用范围带来任何限制。
如图6所示，电子设备600可以包括处理装置(例如中央处理器、图形处理器等)601，其可以根据存储在只读存储器(ROM)602中的程序或者从存储装置608加载到随机访问存储器(RAM)603中的程序而执行各种适当的动作和处理。在RAM 603中，还存储有电子设备600操作所需的各种程序和数据。处理装置601、ROM 602以及RAM 603通过总线604彼此相连。输入/输出(I/O)接口605也连接至总线604。
通常，以下装置可以连接至I/O接口605：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606；包括例如液晶显示器(LCD)、扬声器、振动器等的输出装置607；包括例如磁带、硬盘等的存储装置608；以及通信装置609。通信装置609可以允许电子设备600与其他设备进行无线或有线通信以交换数据。虽然图6示出了具有各种装置的电子设备600，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。图6中示出的每个方框可以代表一个装置，也可以根据需要代表多个装置。
特别地，根据本公开的一些实施例，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，本公开的一些实施例包括一种计算机程序产品，其包括承载在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。在这样的一些实施例中，该计算机程序可以通过通信装置609从网络上被下载和 安装，或者从存储装置608被安装，或者从ROM 602被安装。在该计算机程序被处理装置601执行时，执行本公开的一些实施例的方法中限定的上述功能。
需要说明的是，本公开的一些实施例所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。在本公开的一些实施例中，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。而在本公开的一些实施例中，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(HyperText 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：：将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；确定上述哈希编码集合中每个哈希编码出现的先验概率；基于各个先验概率的熵生成目标函数。
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一些实施例的操作的计算机程序代码，所述程序设计语言包括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。
描述于本公开的一些实施例中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。所描述的单元也可以设置在处理器中，例如，可以描述为：一种处理器包括归一操作单元、第一生成单元、第一确定单元和第二生成单元。其中，这些单元的名称在某种情况下并不构成对该单元本身的限定，例如，归一操作单元还可以被描述为“将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合的单元”。
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。
根据本公开的一个或多个实施例，提供了一种用于生成目标函数的方法，包括：将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；确定上述哈希编码集合中每个哈希编码出现的先验概率；基于各个先验概率的熵生成目标函数。
根据本公开的一个或多个实施例，将预设第一相乘结果和预设第二相乘结果相加，得到相加结果。确定相加结果的累加之和的均值，所述均值表示每个哈希编码出现的先验概率。
根据本公开的一个或多个实施例，确定上述每个哈希编码中每个维度的编码值和上述每个哈希编码中每个维度的编码值对应的向量取值；确定通过1分别减去上述每个维度的编码值和上述每个维度的编码值对应的向量取值的数值；将上述每个哈希编码中每个维度的编码值和与之对应的向量取值相乘，得到上述预设第一相乘结果；将上述通过1分别减去所述每个维度的编码值和上述每个维度的编码值对应的向量取值的数值相乘，得到上述预设第二相乘结果。
根据本公开的一个或多个实施例，确定上述每个哈希编码出现的先验概率的对数值；将上述每个哈希编码出现的先验概率和上述每个哈希编码出现的先验概率的对数值相乘，得到相乘结果；确定上述相 乘结果的累加之和；取上述累加之和的负数，得到上述目标函数。
根据本公开的一个或多个实施例，上述目标特征图集合通过以下步骤得到，包括：通过特征提取网络对目标图像集合进行特征提取，得到目标特征图集合。
根据本公开的一个或多个实施例，基于上述目标函数，对上述特征提取网络进行优化，得到训练后的特征提取网络。
根据本公开的一个或多个实施例，基于上述训练后的特征提取网络提取两幅图像的哈希编码；基于所提取的哈希编码，确定上述两幅图像的相似度。
根据本公开的一个或多个实施例，提供了一种用于生成目标函数的装置，包括：归一操作单元，被配置成将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；第一生成单元，被配置成生成与上述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；第一确定单元，被配置成确定上述哈希编码集合中每个哈希编码出现的先验概率；第二生成单元，被配置成基于各个先验概率的熵生成目标函数。
根据本公开的一个或多个实施例，上述装置进一步被配置成将预设第一相乘结果和预设第二相乘结果相加，得到相加结果。确定相加结果的累加之和的均值，所述均值表示每个哈希编码出现的先验概率。
根据本公开的一个或多个实施例，上述装置进一步被配置成确定上述每个哈希编码中每个维度的编码值和上述每个哈希编码中每个维度的编码值对应的向量取值；确定通过1分别减去上述每个维度的编码值和上述每个维度的编码值对应的向量取值的数值；将上述每个哈希编码中每个维度的编码值和与之对应的向量取值相乘，得到上述预设第一相乘结果；将上述通过1分别减去所述每个维度的编码值和上述每个维度的编码值对应的向量取值的数值相乘，得到上述预设第二相乘结果。
根据本公开的一个或多个实施例，上述装置进一步被配置成确定上述每个哈希编码出现的先验概率的对数值；将上述每个哈希编码出现的先验概率和上述每个哈希编码出现的先验概率的对数值相乘，得 到相乘结果；确定上述相乘结果的累加之和；取上述累加之和的负数，得到上述目标函数。
根据本公开的一个或多个实施例，上述目标特征图集合进一步被配置成通过特征提取网络对目标图像集合进行特征提取，得到目标特征图集合。
根据本公开的一个或多个实施例，上述装置500还包括优化单元，被配置成基于上述目标函数，对上述特征提取网络进行优化，得到训练后的特征提取网络。
根据本公开的一个或多个实施例，上述装置500还包括第二确定单元，被配置成基于上述训练后的特征提取网络提取两幅图像的哈希编码；基于所提取的哈希编码，确定上述两幅图像的相似度。
以上描述仅为本公开的一些较佳实施例以及对所运用技术原理的说明。本领域技术人员应当理解，本公开的实施例中所涉及的发明范围，并不限于上述技术特征的特定组合而成的技术方案，同时也应涵盖在不脱离上述发明构思的情况下，由上述技术特征或其等同特征进行任意组合而形成的其它技术方案。例如上述特征与本公开的实施例中公开的(但不限于)具有类似功能的技术特征进行互相替换而形成的技术方案。

Claims (10)

  1. 一种用于生成目标函数的方法，包括：
    将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；
    生成与所述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；
    确定所述哈希编码集合中每个哈希编码出现的先验概率；
    基于各个先验概率的熵生成目标函数。
  2. 根据权利要求1所述的方法，其中，所述每个哈希编码出现的先验概率通过以下步骤得到：
    将预设第一相乘结果和预设第二相乘结果相加，得到相加结果；
    确定相加结果的累加之和的均值，所述均值表示每个哈希编码出现的先验概率。
  3. 根据权利要求2所述的方法，其中，所述预设第一相乘结果和预设第二相乘通过以下步骤得到：
    确定所述每个哈希编码中每个维度的编码值和所述每个哈希编码中每个维度的编码值对应的向量取值；
    确定通过1分别减去所述每个维度的编码值和所述每个维度的编码值对应的向量取值的数值；
    将所述每个哈希编码中每个维度的编码值和与之对应的向量取值相乘，得到所述预设第一相乘结果；
    将所述通过1分别减去所述每个维度的编码值和所述每个维度的编码值对应的向量取值的数值相乘，得到所述预设第二相乘结果。
  4. 根据权利要求1所述的方法，其中，所述目标函数通过以下步骤得到：
    确定所述每个哈希编码出现的先验概率的对数值；
    将所述每个哈希编码出现的先验概率和所述每个哈希编码出现的先验概率的对数值相乘，得到相乘结果；
    确定所述相乘结果的累加之和；
    取所述累加之和的负数，得到所述目标函数。
  5. 根据权利要求1所述的方法，其中，所述目标特征图集合通过以下步骤得到，包括：
    通过特征提取网络对目标图像集合进行特征提取，得到所述目标特征图集合。
  6. 根据权利要求2所述的方法，其中，所述方法还包括：
    基于所述目标函数，对所述特征提取网络进行优化，得到训练后的特征提取网络。
  7. 根据权利要求3所述的方法，其中，所述方法还包括：
    基于所述训练后的特征提取网络提取两幅图像的图像特征，进而得到哈希编码；
    基于所得到的哈希编码，确定所述两幅图像的相似度。
  8. 一种特征提取网络的训练装置，包括：
    归一操作单元，被配置成将目标特征图集合中与每个像素对应的向量进行归一化处理以生成目标向量，得到目标向量集合；
    第一生成单元，被配置成生成与所述目标向量集合中每个向量对应的哈希编码，得到哈希编码集合；
    第一确定单元，被配置成确定所述哈希编码集合中每个哈希编码出现的先验概率；
    第二生成单元，被配置成基于各个先验概率的熵生成目标函数。
  9. 一种电子设备，包括：
    一个或多个处理器；
    存储装置，其上存储有一个或多个程序，
    当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如权利要求1-7中任一所述的方法。
  10. 一种计算机可读介质，其上存储有计算机程序，其中，所述程序被处理器执行时实现如权利要求1-7中任一所述的方法。
PCT/CN2021/096144 2020-07-16 2021-05-26 用于生成目标函数的方法、装置、电子设备和计算机可读介质 WO2022012178A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/005,081 US20230281956A1 (en) 2020-07-16 2021-05-26 Method for generating objective function, apparatus, electronic device and computer readable medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010685574.2 2020-07-16
CN202010685574.2A CN111915689B (zh) 2020-07-16 2020-07-16 用于生成目标函数的方法、装置、电子设备和计算机可读介质

Publications (1)

Publication Number Publication Date
WO2022012178A1 true WO2022012178A1 (zh) 2022-01-20

Family

ID=73281001

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/096144 WO2022012178A1 (zh) 2020-07-16 2021-05-26 用于生成目标函数的方法、装置、电子设备和计算机可读介质

Country Status (3)

Country Link
US (1) US20230281956A1 (zh)
CN (1) CN111915689B (zh)
WO (1) WO2022012178A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111915689B (zh) * 2020-07-16 2023-11-24 抖音视界有限公司 用于生成目标函数的方法、装置、电子设备和计算机可读介质
CN113033580B (zh) * 2021-03-31 2024-02-02 北京有竹居网络技术有限公司 图像处理方法、装置、存储介质及电子设备

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103744973A (zh) * 2014-01-11 2014-04-23 西安电子科技大学 基于多特征哈希的视频拷贝检测方法
CN109284411A (zh) * 2017-07-19 2019-01-29 哈尔滨工业大学深圳研究生院 一种基于有监督超图离散化图像二值编码方法
CN109960738A (zh) * 2019-03-15 2019-07-02 西安电子科技大学 基于深度对抗哈希学习的大规模遥感影像内容检索方法
US20190207753A1 (en) * 2017-02-15 2019-07-04 Tongmyong University Industry-Academy Cooperation Foundation Method of hashing vector data based on multi-scale curvature for vector content authentication
CN110209851A (zh) * 2019-06-10 2019-09-06 北京字节跳动网络技术有限公司 模型训练方法、装置、电子设备及存储介质
CN110929080A (zh) * 2019-11-26 2020-03-27 西安电子科技大学 基于注意力和生成对抗网络的光学遥感图像检索方法
CN111915689A (zh) * 2020-07-16 2020-11-10 北京字节跳动网络技术有限公司 用于生成目标函数的方法、装置、电子设备和计算机可读介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7761466B1 (en) * 2007-07-30 2010-07-20 Hewlett-Packard Development Company, L.P. Hash-based image identification
CN104298791A (zh) * 2014-11-19 2015-01-21 中国石油大学(华东) 一种基于集成哈希编码的快速图像检索方法
CN109325140B (zh) * 2018-07-12 2021-07-13 北京奇虎科技有限公司 从图像中提取哈希码的方法、装置及图像检索方法、装置
US10846554B2 (en) * 2018-07-17 2020-11-24 Avigilon Corporation Hash-based appearance search

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103744973A (zh) * 2014-01-11 2014-04-23 西安电子科技大学 基于多特征哈希的视频拷贝检测方法
US20190207753A1 (en) * 2017-02-15 2019-07-04 Tongmyong University Industry-Academy Cooperation Foundation Method of hashing vector data based on multi-scale curvature for vector content authentication
CN109284411A (zh) * 2017-07-19 2019-01-29 哈尔滨工业大学深圳研究生院 一种基于有监督超图离散化图像二值编码方法
CN109960738A (zh) * 2019-03-15 2019-07-02 西安电子科技大学 基于深度对抗哈希学习的大规模遥感影像内容检索方法
CN110209851A (zh) * 2019-06-10 2019-09-06 北京字节跳动网络技术有限公司 模型训练方法、装置、电子设备及存储介质
CN110929080A (zh) * 2019-11-26 2020-03-27 西安电子科技大学 基于注意力和生成对抗网络的光学遥感图像检索方法
CN111915689A (zh) * 2020-07-16 2020-11-10 北京字节跳动网络技术有限公司 用于生成目标函数的方法、装置、电子设备和计算机可读介质

Also Published As

Publication number Publication date
CN111915689A (zh) 2020-11-10
CN111915689B (zh) 2023-11-24
US20230281956A1 (en) 2023-09-07

Similar Documents

Publication Publication Date Title
CN111476309B (zh) 图像处理方法、模型训练方法、装置、设备及可读介质
CN111915480B (zh) 生成特征提取网络的方法、装置、设备和计算机可读介质
WO2022227886A1 (zh) 超分修复网络模型生成方法、图像超分修复方法及装置
CN112668588B (zh) 车位信息生成方法、装置、设备和计算机可读介质
WO2022012178A1 (zh) 用于生成目标函数的方法、装置、电子设备和计算机可读介质
WO2020207174A1 (zh) 用于生成量化神经网络的方法和装置
CN112270200B (zh) 一种文本信息的翻译方法、装置、电子设备和存储介质
CN112766284B (zh) 图像识别方法和装置、存储介质和电子设备
CN113449070A (zh) 多模态数据检索方法、装置、介质及电子设备
WO2023179310A1 (zh) 图像修复方法、装置、设备、介质及产品
WO2022134984A1 (zh) 热词识别方法、装置、介质和电子设备
CN112418249A (zh) 掩膜图像生成方法、装置、电子设备和计算机可读介质
WO2022052889A1 (zh) 图像识别方法、装置、电子设备和计算机可读介质
CN111626044B (zh) 文本生成方法、装置、电子设备及计算机可读存储介质
CN111680754B (zh) 图像分类方法、装置、电子设备及计算机可读存储介质
CN111814807B (zh) 用于处理图像的方法、装置、电子设备和计算机可读介质
CN111639198A (zh) 媒体文件识别方法、装置、可读介质及电子设备
CN111797932B (zh) 图像分类方法、装置、设备和计算机可读介质
CN111582376B (zh) 神经网络的可视化方法、装置、电子设备和介质
CN114863025B (zh) 三维车道线生成方法、装置、电子设备和计算机可读介质
CN112215789B (zh) 图像去雾方法、装置、设备和计算机可读介质
CN111539524B (zh) 轻量级自注意力模块和神经网络构架的搜索方法
CN113808050B (zh) 3d点云的去噪方法、装置、设备及存储介质
CN111898658B (zh) 图像分类方法、装置和电子设备
CN116342721A (zh) 图像压缩方法、装置、设备及存储介质

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: 21843450

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21843450

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 03/05/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 21843450

Country of ref document: EP

Kind code of ref document: A1