CN113033767A - Knowledge distillation-based data compression recovery method and system for neural network - Google Patents
Knowledge distillation-based data compression recovery method and system for neural network Download PDFInfo
- Publication number
- CN113033767A CN113033767A CN202110188891.8A CN202110188891A CN113033767A CN 113033767 A CN113033767 A CN 113033767A CN 202110188891 A CN202110188891 A CN 202110188891A CN 113033767 A CN113033767 A CN 113033767A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- knowledge
- knowledge distillation
- representing
- 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.)
- Pending
Links
- 238000013140 knowledge distillation Methods 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000013144 data compression Methods 0.000 title claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 35
- 238000011084 recovery Methods 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000004458 analytical method Methods 0.000 claims abstract description 14
- 239000002775 capsule Substances 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 5
- 125000004122 cyclic group Chemical group 0.000 claims description 5
- 238000007906 compression Methods 0.000 abstract description 8
- 230000006835 compression Effects 0.000 abstract description 8
- 230000006870 function Effects 0.000 description 7
- 238000004590 computer program Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L69/00—Network arrangements, protocols or services independent of the application payload and not provided for in the other groups of this subclass
- H04L69/04—Protocols for data compression, e.g. ROHC
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/85—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using pre-processing or post-processing specially adapted for video compression
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Computer Security & Cryptography (AREA)
- Computer Networks & Wireless Communication (AREA)
- Multimedia (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The disclosure relates to the technical field of neural networks, in particular to a data compression recovery method and system of a neural network based on knowledge distillation. The method comprises the following steps: inputting the raw data into a knowledge distillation network for training; inputting compressed data obtained by compressing the original data into a target network; analyzing and training the target network into which the compressed data is input through the knowledge distillation network; and outputting the trained target network data to obtain a recovered analysis result. The system comprises: the data compression module is used for generating compressed data; a knowledge distillation module for extracting data incoming from the data compression module into a high quality feature stream; and the reasoning module is used for deploying the recovered analysis result. The method and the system can reduce the loss of the image, the video or the audio data caused by compression, greatly improve the performance of the target network and obtain the analysis result with greatly improved accuracy.
Description
Technical Field
The present disclosure relates to the field of neural network technology, and more particularly, to a method and system for data compression and recovery of a neural network based on knowledge distillation.
Background
With the rapid development of artificial neural networks, various neural network models are more and more widely applied. Data adopted by many neural networks are lossy-compressed, and lossy compression causes loss which is difficult to recover for data signals. This loss not only affects human perception of data, but also reduces the performance of various neural networks.
There are many recovery techniques that are dedicated to recovering such losses from the signal level. Taking picture compression as an example, the removal of picture compression artifacts is a technology aiming at recovering the loss of the picture caused by lossy compression as much as possible so as to meet the requirements of human viewing and various visual analysis tasks. The existing methods aim to recover the signal values of the pictures, but we find that such methods of recovering signal values do not contribute significantly to the neural network.
Knowledge distillation is a model compression method that is widely used in the industry because it is simple and efficient, but lossy compression can degrade the performance of the target network.
The present application therefore proposes an improved method and system to at least partially solve the above technical problem.
Disclosure of Invention
To achieve the above technical object, the present disclosure provides a data compression recovery method for a neural network based on knowledge distillation, including:
inputting the raw data into a knowledge distillation network for training;
inputting compressed data obtained by compressing the original data into a target network;
analyzing and training the target network into which the compressed data is input through the knowledge distillation network;
and outputting the trained target network data to obtain a recovered analysis result.
Specifically, the knowledge distillation network may be a teacher network, the teacher network being a presentation knowledge teacher network or a signal knowledge teacher network.
Further, the method for representing the training of the knowledge teacher network on the raw data comprises the following steps:
θ*＝argminθltask(f(x；θ)，y)
wherein, theta*Teacher network of representation knowledge obtained by representation training,/taskRepresenting the associated loss function, x and y representing the raw and output data, respectively.
Further, the method for training the raw data by the signal knowledge teacher network comprises the following steps:
wherein, FsRepresenting the characteristic representation of the target network, x representing the original data, and l representing a function for measuring the difference of the two data.
Further, when the network representing knowledge teachers trains the target network into which the compressed data is input, the difference between the feature representation extracted from the compressed data by the target network and the high-quality feature representation extracted by the network representing knowledge teachers is restrained, and the specific method is as follows:
wherein M isRKTIs a high-quality feature representation representing the knowledge teacher network, M is a feature representation obtained by the target network, and c (x) represents compressed data.
Preferably, the target network is a convolutional neural network, a cyclic neural network and/or a capsule network.
Preferably, the original data is picture data, video data and/or audio data, and the compressed data is picture compressed data, video compressed data and/or audio compressed data.
The present disclosure provides a data compression recovery system for a neural network based on knowledge distillation, comprising:
the data compression module is used for acquiring original data, compressing the original data and generating compressed data;
a knowledge distillation module for extracting data incoming from the data compression module into a high quality feature stream;
and the reasoning module is used for deploying the recovered analysis result output from the knowledge distillation module.
Specifically, the knowledge distillation module comprises a knowledge distillation network and a target network, and the knowledge distillation network analyzes and trains the target network into which the compressed data is input.
Specifically, the knowledge distillation network is a representation knowledge teacher network or a signal knowledge teacher network.
The beneficial effect of this disclosure does:
the invention provides a data compression recovery method and a data compression recovery system of a neural network based on knowledge distillation, which can reduce the loss of picture, video or audio data caused by compression, greatly improve the performance of a target network and obtain an analysis result with greatly improved accuracy.
Drawings
Fig. 1 shows a schematic flow diagram of embodiment 1 of the present disclosure;
fig. 2 shows a schematic structural diagram of embodiment 2 of the present disclosure;
fig. 3 shows a process schematic of embodiment 3 of the present disclosure.
Detailed Description
Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood that the description is illustrative only and is not intended to limit the scope of the present disclosure. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present disclosure. It will be apparent to one skilled in the art that the present disclosure may be practiced without one or more of these details. In other instances, well-known features of the art have not been described in order to avoid obscuring the present disclosure.
It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the disclosure. As used herein, the singular is intended to include the plural unless the context clearly dictates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
Exemplary embodiments according to the present disclosure will now be described in more detail with reference to the accompanying drawings. These exemplary embodiments may, however, be embodied in many different forms and should not be construed as limited to only the embodiments set forth herein. The figures are not drawn to scale, wherein certain details may be exaggerated and omitted for clarity. The shapes of various regions, layers, and relative sizes and positional relationships therebetween shown in the drawings are merely exemplary, and deviations may occur in practice due to manufacturing tolerances or technical limitations, and a person skilled in the art may additionally design regions/layers having different shapes, sizes, relative positions, as actually required.
Example 1:
the present disclosure provides a data compression recovery method of a neural network based on knowledge distillation, as shown in fig. 1, including:
inputting the raw data into a knowledge distillation network for training;
inputting compressed data obtained by compressing original data into a target network;
analyzing and training the target network into which the compressed data is input through a knowledge distillation network;
and outputting the trained target network data to obtain a recovered analysis result.
In particular, the knowledge distillation network may be a teacher network, the teacher network being a presentation knowledge teacher network or a signal knowledge teacher network.
Further, the method for representing the training of the knowledge teacher network on the raw data comprises the following steps:
θ*＝argminθ ltask(f(x；θ)，V)
wherein, theta*Teacher network of representation knowledge obtained by representation training,/taskRepresenting the associated loss function, x and y representing the raw and output data, respectively.
Further, the method for training the original data by the signal knowledge teacher network comprises the following steps:
wherein, FsRepresenting the characteristic representation of the target network, x representing the original data, and l representing a function for measuring the difference of the two data.
Further, when the expressed knowledge teacher network performs analysis training on the target network into which the compressed data is input, the expressed knowledge teacher network restricts the difference between the feature representation extracted from the compressed data by the target network and the high-quality feature representation extracted by the expressed knowledge teacher network, and the specific method is as follows:
wherein M isRKTIs a high-quality feature representation representing the knowledge teacher network, M is a feature representation obtained by the target network, and c (x) represents compressed data.
Preferably, the target network is a convolutional neural network, a cyclic neural network and/or a capsule network.
Preferably, the original data is picture data, video data and/or audio data, and the compressed data is picture compressed data, video compressed data and/or audio compressed data.
Example 2:
the present disclosure provides a data compression recovery system of a neural network based on knowledge distillation, as shown in fig. 2, including:
the data compression module is used for acquiring original data, compressing the original data and generating compressed data;
a knowledge distillation module for extracting data incoming from the data compression module into a high quality feature stream;
and the reasoning module is used for deploying the recovered analysis result output from the knowledge distillation module. Specifically, the knowledge distillation module includes a knowledge distillation network and a target network, the knowledge distillation network training the target network into which the compressed data has been input.
Specifically, the knowledge distillation network may be a teacher network, the teacher network being a presentation knowledge teacher network or a signal knowledge teacher network.
Further, the raw data are input into a knowledge distillation network of a knowledge distillation module for training, compressed data obtained by compressing the raw data in a data compression module are input into a target network of the knowledge distillation module, the target network into which the compressed data are input is analyzed and trained through the knowledge distillation network in the knowledge distillation module, and the trained target network data are output to an inference module for deploying tasks.
Specifically, the method for representing the training of the knowledge teacher network on the raw data comprises the following steps:
θ*＝argminθltask(f(x；θ)，y)
wherein, theta*Teacher network of representation knowledge obtained by representation training,/taskRepresenting the associated loss function, x and y representing the raw and output data, respectively.
Further, the method for training the raw data by the signal knowledge teacher network comprises the following steps:
wherein, FsRepresenting the characteristic representation of the target network, x representing the original data, and l representing a function for measuring the difference of the two data.
Further, when the expressed knowledge teacher network analyzes and trains the target network into which the compressed data is input, the expressed knowledge teacher network restricts the difference between the feature representation extracted from the compressed data by the target network and the high-quality feature representation extracted by the expressed knowledge teacher network, and the specific method is as follows:
wherein M isRKTIs a high-quality feature representation representing the knowledge teacher network, M is a feature representation obtained by the target network, and c (x) represents compressed data.
Preferably, the target network is a convolutional neural network, a cyclic neural network and/or a capsule network.
Preferably, the original data is picture data, video data and/or audio data, and the compressed data is picture compressed data, video compressed data and/or audio compressed data.
Example 3:
the present disclosure provides a data compression and recovery system of a neural network based on knowledge distillation, which includes a data compression module, a knowledge distillation module and an inference module, wherein, as shown in fig. 3, the knowledge distillation module includes a representation knowledge teacher network and a target network, and the process of representing knowledge distillation is realized from the knowledge teacher network to the target network.
Specifically, raw data is input into a knowledge teacher representation network of a knowledge distillation module for training, compressed data obtained by compressing the raw data in a data compression module is input into a target network of the knowledge distillation module, the target network into which the compressed data is input is trained through the knowledge teacher representation network in the knowledge distillation module, and finally the trained target network data is output to an inference module for deploying tasks.
Specifically, the method for training the raw data by the representation knowledge teacher network comprises the following steps:
θ*＝argminθ ltask(f(x；θ)，y)
wherein, theta*Teacher network of representation knowledge obtained by representation training,/taskRepresenting the associated loss function, x and y representing the raw and output data, respectively.
Further, when the expressed knowledge teacher network analyzes and trains the target network into which the compressed data is input, the expressed knowledge teacher network restricts the difference between the feature representation extracted from the compressed data by the target network and the high-quality feature representation extracted by the expressed knowledge teacher network, and the specific method is as follows:
wherein M isRKTIs a high-quality feature representation representing the knowledge teacher network, M is a feature representation obtained by the target network, and c (x) represents compressed data.
Preferably, the target network is a convolutional neural network, a cyclic neural network and/or a capsule network.
Preferably, the original data is picture data, video data and/or audio data, and the compressed data is picture compressed data, video compressed data and/or audio compressed data.
Example 4:
the present disclosure provides a computer device comprising a memory and a processor, the memory having stored therein computer readable instructions which, when executed by the processor, cause the processor to perform the steps of a method for data compression recovery for a neural network based on knowledge distillation: inputting the raw data into a knowledge distillation network for training; inputting compressed data obtained by compressing the original data into a target network; analyzing and training the target network into which the compressed data is input through the knowledge distillation network; and outputting the trained target network data to obtain a recovered analysis result.
Example 5:
the present disclosure also provides a storage medium having stored thereon computer-readable instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of a method for data compression recovery for a neural network based on knowledge distillation: inputting the raw data into a knowledge distillation network for training; inputting compressed data obtained by compressing the original data into a target network; analyzing and training the target network into which the compressed data is input through the knowledge distillation network; and outputting the trained target network data to obtain a recovered analysis result.
The method and the system for data compression and recovery of the neural network based on knowledge distillation provided by the embodiment of the disclosure can be realized by relying on a computer program. The computer program may be integrated into the application or may run as a separate tool-like application. The data compression recovery device and medium of the neural network based on knowledge distillation in the disclosed embodiment include but are not limited to: personal computers, tablet computers, handheld devices, in-vehicle devices, wearable devices, computing devices or other processing devices connected to a wireless modem, and the like. The user terminals may be called different names in different networks, for example: user equipment, access terminal, subscriber unit, subscriber station, mobile station, remote terminal, mobile device, user terminal, wireless communication device, user agent or user equipment, cellular telephone, cordless telephone, Personal Digital Assistant (PDA), terminal equipment in a 5G network or future evolution network, and the like.
The embodiments of the present disclosure have been described above. However, this is for illustrative purposes only and is not intended to limit the scope of the present disclosure. The scope of the disclosure is defined by the appended claims and equivalents thereof. Various alternatives and modifications can be devised by those skilled in the art without departing from the scope of the present disclosure, and such alternatives and modifications are intended to be within the scope of the present disclosure.
Claims (10)
1. A data compression recovery method of a neural network based on knowledge distillation is characterized by comprising the following steps:
inputting the raw data into a knowledge distillation network for training;
inputting compressed data obtained by compressing the original data into a target network;
analyzing and training the target network into which the compressed data is input through the knowledge distillation network;
and outputting the trained target network data to obtain a recovered analysis result.
2. The method for data compression recovery based on the neural network of knowledge distillation as claimed in claim 1, wherein the knowledge distillation network is a representation knowledge teacher network or a signal knowledge teacher network.
3. The method for data compression and recovery of the neural network based on knowledge distillation as claimed in claim 2, wherein the method for representing the training of the knowledge teacher network to the raw data is as follows:
θ*＝argminθltask(f(x；θ)，y)
wherein, theta*Teacher network of representation knowledge obtained by representation training,/taskRepresenting the associated loss function, x and y representing the raw and output data, respectively.
4. The method for data compression and recovery of the neural network based on knowledge distillation as claimed in claim 2, wherein the method for training the raw data by the signal knowledge teacher network is as follows:
wherein, FsRepresenting the characteristic representation of the target network, x representing the original data, and l representing a function for measuring the difference of the two data.
5. The method for data compression and recovery of a neural network based on knowledge distillation as claimed in claim 2, wherein the representing knowledge teacher network is used for analyzing and training the target network into which the compressed data is input, and restricting the difference between the feature representation extracted from the compressed data by the target network and the high-quality feature representation extracted by the representing knowledge teacher network, and the method comprises the following steps:
wherein M isRKTIs a high-quality feature representation representing the knowledge teacher network, M is a feature representation obtained by the target network, and c (x) represents compressed data.
6. The method for data compression recovery of a neural network based on knowledge distillation of claim 1, wherein the target network is a convolutional neural network, a cyclic neural network and/or a capsule network.
7. The method for data compression recovery of a neural network based on knowledge distillation of claim 1, wherein the raw data is picture data, video data and/or audio data.
8. A data compression recovery system for a neural network based on knowledge distillation, comprising:
the data compression module is used for acquiring original data, compressing the original data and generating compressed data;
a knowledge distillation module for extracting data incoming from the data compression module into a high quality feature stream;
and the reasoning module is used for deploying the recovered analysis result output from the knowledge distillation module.
9. The knowledge distillation based data compression recovery system for a neural network of claim 8, wherein the knowledge distillation module comprises a knowledge distillation network and a target network, the knowledge distillation network performing analytical training on the target network into which the compressed data has been input.
10. The knowledge distillation based neural network data compression recovery system of claim 9, wherein the knowledge distillation network is a representation knowledge teacher network or a signal knowledge teacher network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110188891.8A CN113033767A (en) | 2021-02-19 | 2021-02-19 | Knowledge distillation-based data compression recovery method and system for neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110188891.8A CN113033767A (en) | 2021-02-19 | 2021-02-19 | Knowledge distillation-based data compression recovery method and system for neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113033767A true CN113033767A (en) | 2021-06-25 |
Family
ID=76461310
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110188891.8A Pending CN113033767A (en) | 2021-02-19 | 2021-02-19 | Knowledge distillation-based data compression recovery method and system for neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113033767A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116382979A (en) * | 2023-04-07 | 2023-07-04 | 海南翔文飞网络科技有限公司 | Data loss prevention disaster recovery method and server combined with expert system neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110880036A (en) * | 2019-11-20 | 2020-03-13 | 腾讯科技（深圳）有限公司 | Neural network compression method and device, computer equipment and storage medium |
| CN111160533A (en) * | 2019-12-31 | 2020-05-15 | 中山大学 | Neural network acceleration method based on cross-resolution knowledge distillation |
| CN112200062A (en) * | 2020-09-30 | 2021-01-08 | 广州云从人工智能技术有限公司 | Target detection method and device based on neural network, machine readable medium and equipment |
-
2021
- 2021-02-19 CN CN202110188891.8A patent/CN113033767A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110880036A (en) * | 2019-11-20 | 2020-03-13 | 腾讯科技（深圳）有限公司 | Neural network compression method and device, computer equipment and storage medium |
| CN111160533A (en) * | 2019-12-31 | 2020-05-15 | 中山大学 | Neural network acceleration method based on cross-resolution knowledge distillation |
| CN112200062A (en) * | 2020-09-30 | 2021-01-08 | 广州云从人工智能技术有限公司 | Target detection method and device based on neural network, machine readable medium and equipment |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116382979A (en) * | 2023-04-07 | 2023-07-04 | 海南翔文飞网络科技有限公司 | Data loss prevention disaster recovery method and server combined with expert system neural network |
| CN116382979B (en) * | 2023-04-07 | 2024-03-19 | 国网冀北电力有限公司计量中心 | Data loss prevention disaster recovery method and server combined with expert system neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111598776B (en) | Image processing method, image processing device, storage medium and electronic apparatus | |
| US11138903B2 (en) | Method, apparatus, device and system for sign language translation | |
| CN111368685B (en) | Method and device for identifying key points, readable medium and electronic equipment | |
| CN110490296A (en) | A kind of method and system constructing convolutional neural networks (CNN) model | |
| CN107909583B (en) | Image processing method and device and terminal | |
| CN110837842A (en) | Video quality evaluation method, model training method and model training device | |
| US11716438B2 (en) | Method for motion estimation, non-transitory computer-readable storage medium, and electronic device | |
| CN111414879A (en) | Face shielding degree identification method and device, electronic equipment and readable storage medium | |
| WO2020062191A1 (en) | Image processing method, apparatus and device | |
| EP4207195A1 (en) | Speech separation method, electronic device, chip and computer-readable storage medium | |
| CN111694978A (en) | Image similarity detection method and device, storage medium and electronic equipment | |
| EP4249869A1 (en) | Temperature measuring method and apparatus, device and system | |
| CN111343356A (en) | Image processing method, image processing apparatus, storage medium, and electronic device | |
| CN111429374A (en) | Method and device for eliminating moire in image | |
| CN107145855B (en) | Reference quality blurred image prediction method, terminal and storage medium | |
| CN113033767A (en) | Knowledge distillation-based data compression recovery method and system for neural network | |
| CN112785669B (en) | Virtual image synthesis method, device, equipment and storage medium | |
| US20230281956A1 (en) | Method for generating objective function, apparatus, electronic device and computer readable medium | |
| CN113658065A (en) | Image noise reduction method and device, computer readable medium and electronic equipment | |
| CN113902636A (en) | Image deblurring method and device, computer readable medium and electronic equipment | |
| CN111626035B (en) | Layout analysis method and electronic equipment | |
| CN117541770A (en) | Data enhancement method and device and electronic equipment | |
| CN114283493A (en) | Artificial intelligence-based identification system | |
| CN113240599A (en) | Image toning method and device, computer-readable storage medium and electronic equipment | |
| CN111382764B (en) | Neural network model building method and device for face recognition or gesture recognition and computer readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210625 |