CN108446334B - Image retrieval method based on content for unsupervised countermeasure training - Google Patents
Image retrieval method based on content for unsupervised countermeasure training Download PDFInfo
- Publication number
- CN108446334B CN108446334B CN201810154813.4A CN201810154813A CN108446334B CN 108446334 B CN108446334 B CN 108446334B CN 201810154813 A CN201810154813 A CN 201810154813A CN 108446334 B CN108446334 B CN 108446334B
- Authority
- CN
- China
- Prior art keywords
- model
- data set
- picture
- pictures
- training
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000012549 training Methods 0.000 title claims abstract description 39
- 230000008569 process Effects 0.000 claims abstract description 21
- 238000012360 testing method Methods 0.000 claims abstract description 16
- 238000010276 construction Methods 0.000 claims abstract description 6
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 230000004913 activation Effects 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 5
- 238000007477 logistic regression Methods 0.000 claims description 4
- 239000013598 vector Substances 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 101100481876 Danio rerio pbk gene Proteins 0.000 claims description 2
- 101100481878 Mus musculus Pbk gene Proteins 0.000 claims description 2
- 230000006870 function Effects 0.000 description 24
- 238000007667 floating Methods 0.000 description 10
- 210000002569 neuron Anatomy 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- 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
- G06N3/047—Probabilistic or stochastic 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
- G06N3/088—Non-supervised learning, e.g. competitive learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Library & Information Science (AREA)
- Probability & Statistics with Applications (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
A method of unsupervised confrontation training content-based image retrieval, the method comprising the steps of: step one, network construction, wherein the unsupervised countermeasure network framework is composed of a generation model and a discrimination model. The generation model and the discrimination model are both formed by three layers of fully-connected networks; step two, preprocessing a data set; step three, network training, the process is as follows: step 3.1: initializing a generation model and distinguishing model parameters by using random weight; step 3.2: training a generating model; step 3.3: training a discrimination model; and step four, testing the precision. The invention provides the image retrieval method based on the content for the unsupervised countermeasure training, which has better robustness, lower requirement on training data and no need of a large amount of labeled information.
Description
Technical Field
The invention relates to multimedia big data processing and analysis in the field of computer vision, in particular to an unsupervised countermeasure content-based picture retrieval method, and belongs to the field of image retrieval.
Background
With the development of network sharing technology, more and more pictures on the network can be shared and received in real time. Content-based image retrieval techniques occupy a significant part of the image processing process. With the rapid development of deep learning methods in recent years, the image retrieval technology performance based on contents is greatly improved thanks to the accurate expression of the depth features to the image contents. But such promotion is based on labeled training. Supervised training methods based on labeled training may not work well in situations where training data labels are not available or training data is low.
Disclosure of Invention
In order to overcome the defects of poor robustness, high requirement on training data and the need of a large amount of labeled information in the conventional picture retrieval technology, the invention provides the content-based image retrieval method for unsupervised countermeasure training, which has better robustness, lower requirement on the training data and no need of a large amount of labeled information.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a method of unsupervised confrontation training content-based image retrieval, the method comprising the steps of:
step one, network construction, the process is as follows:
step 1.1: the unsupervised countermeasure network framework is composed of a generative model and a discriminant model. The generation model and the discrimination model are both formed by three layers of fully-connected networks;
step 1.2, a Relu activation function is connected behind a first full connection layer of the generated model;
step 1.3: a tanh activation function is connected behind a second full connection layer of the generated model, and the output is controlled to be {0, 1 };
step 1.4: generating a third full-connection layer of the model and then connecting a distance measurement function;
step 1.5: the first full-connection layer of the discriminant model is followed by Relu activation function
Step 1.6: a tanh activation function is connected behind the second full connection layer of the discrimination model, and the output is controlled to be {0, 1 };
step 1.7: the third full-connection layer of the discrimination model is connected with a similarity score function;
step 1.8: the discrimination model feeds back the calculated similarity score to the generation model;
step two, preprocessing the data set, wherein the process is as follows:
step 2.1: dividing the data into a query data set Q, a test data set Q' and a data set D to be retrieved, wherein a part of pictures are randomly extracted from the data set to be retrieved to serve as a data set F for fine-tuning network parameters when picture features are extracted;
step 2.2: extracting picture characteristics by using a pre-trained VGG model on ImageNet, wherein a small amount of pictures are required to be used for fine-tuning network parameters before the VGG is used for extracting the picture characteristics;
step 2.3: inputting the pictures into an unsupervised countermeasure network in the form of feature vectors;
step three, network training, the process is as follows:
step 3.1: initializing a generation model and distinguishing model parameters by using random weight;
step 3.2: training the generative model, the process is as follows:
step 3.2.1, sending the picture characteristics of the query data set Q and the data set D to be retrieved, which are extracted by the VGG network, into a generation model;
step 3.2.2, generating a model to optimize the weight of the characteristics of the input query data set Q and the data set D to be retrieved;
step 3.2.3: generating a model for each image to be queried, calculating the cosine distance between the model and the image in the data set to be retrieved, converting the similarity into the probability of selecting the image by using a softmax function, and selecting K image features from the data set D to be retrieved as the output of a generator according to the probability;
step 3.2.4: maximizing the difference value between the similarity of the query picture and the selected K pictures and 1 by using a logic loss function;
step 3.3: training a discriminant model by the following process:
step 3.3.1, taking the characteristics of the K pictures returned by the generator as the input of the discriminator, and carrying out weight optimization on the query picture and the characteristics of the K pictures returned by the generator again;
step 3.3.2, recalculating the cosine distance between each inquiry picture and the returned K pictures, and giving a similarity score according to the distance;
step 3.3.3, the judger feeds the calculated similarity score back to the generator, the similarity score is used for selecting the picture to be retrieved next by the generator, and the difference value between the distance between the inquiry picture and the returned K pictures and 0 is reduced by using a logistic regression function;
step 3.4: minimizing loss function using stochastic gradient descent algorithm
Step four, testing the precision, and the process is as follows:
step 4.1: sending the preprocessed test data set Q' into an optimal generator model;
step 4.2: the generator selects the picture with the highest degree of similarity of topK sheets from the data set D to be retrieved according to the given inquiry picture
Step 4.3: comparing whether the tags of the inquired pictures are consistent with the tags of the K pictures returned by the generator or not, and calculating the average accuracy of all the inquired pictures according to the evaluation criteria in the information retrieval;
through the operation of the steps, the retrieval of the test picture can be realized.
The invention has the following beneficial effects: the invention provides an unsupervised confrontation training image retrieval method. Under the condition of inputting label-free data, the generating model and the judging model improve the performance of the generating model and the judging model through maximum minimum confrontation training, wherein the generating model can find out K pictures with the highest similarity with the query picture, and the judging model can judge whether the picture output by the generator is similar to the query picture to the maximum extent. The method solves the problem that a large amount of labeled information is needed in the training process in deep learning, and meanwhile, the generation of the confrontation network is successfully realized in the picture retrieval task based on the content.
Drawings
FIG. 1 is a diagram of an unsupervised countermeasure training picture retrieval network framework used in the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
Referring to fig. 1, a content-based image retrieval method for unsupervised countermeasure training includes four processes of construction of an unsupervised countermeasure training network, data set preprocessing, network training and picture retrieval testing.
The pictures in this embodiment are divided into 10 types, and 600 pictures are provided for each type. Randomly selecting 20 pictures in each type of pictures, and equally dividing the pictures into two parts: the query picture Q and the test picture Q', the remaining 580 pictures constitute a data set D to be retrieved. The picture retrieval network structure framework is shown in fig. 1, and the operation steps comprise four processes of network construction, data set preprocessing, network training and picture retrieval testing.
The unsupervised confrontation training image retrieval method comprises the following steps:
step one, network construction, the process is as follows:
step 1.1: the unsupervised countermeasure network framework is composed of a generative model and a discriminant model. The generation model and the discrimination model are both formed by three layers of fully-connected networks;
step 1.2, setting the number of first full-connection layer neurons of a generated model to be 48, setting the weight value to be W _1, defining the number to be a floating point type variable, setting the bias value to be b _1, defining the number to be the floating point type variable, and then connecting a Relu activation function;
step 1.3: setting the number of neurons of a second full connection layer of the generated model as 32, setting the weight as W _2, defining as a floating point type variable with bias as b _2, and then connecting with a tanh activation function to control the output as {0, 1 };
step 1.4: the number of neurons of the third full connection layer of the generated model is 10, the weight is W _3, the model is defined as a floating point type variable, no bias exists, and a distance measurement function is connected after the floating point type variable is defined;
step 1.5: the number of the first full-connection layer neuron of the discrimination model is set to be 48, the weight value is W _4, the first full-connection layer neuron is defined as a floating point type variable, the bias value is b _4, the first full-connection layer neuron is defined as a floating point type variable, and the first full-connection layer neuron is followed by a Relu activating function
Step 1.6: setting the number of neurons of a second full connection layer of the discrimination model as 32, setting a weight value as W _5, defining as a floating point type variable, setting a bias as b _5, defining as a floating point type variable, and then connecting with a tanh activation function to control the output as {0, 1 };
step 1.7: the number of neurons of the third full connection layer of the generated model is 10, the weight is W _6, the model is defined as a floating point type variable without bias, and then a similarity score function is connected;
step 1.8: the discrimination model feeds back the calculated similarity score to the generation model in the form of a generator loss function weight;
step two, preprocessing the data set, wherein the process is as follows:
step 2.1: the data is divided into a query data set Q, a test data set Q' and a data set D to be retrieved. Randomly extracting 5000 pictures from the data set D to be retrieved to serve as a data set F for fine-tuning network parameters when picture features are extracted;
step 2.2: finely adjusting a pre-trained VGG model on ImageNet by using a data set F, and setting the characteristic dimension of an output picture to be 48 dimensions;
step 2.3: extracting feature vectors corresponding to the query data set Q and the data set D to be retrieved by using a fine-tuned VGG network model, normalizing feature values to be between {0, 1} by using a sigmoid function, and storing the feature vectors into a TXT format file;
step three, network training, the process is as follows:
step 3.1: initializing parameters in a generating model and a judging model by using random weight; setting the generated model to iterate 10 times each time, judging that the model iterates 3 times to be complete network training for one time, and totally performing 5 times of complete training;
step 3.2: training a generating model;
step 3.2.1, setting the learning rate to be 0.0001 and K to be 500;
step 3.2.1, sending the query data set Q and the data set D to be retrieved in the TXT format into a network as the input of a generation model;
step 3.2.2, the generated model utilizes a three-layer full-connection network to carry out weight optimization on the characteristics of the input query data set Q and the data set D to be retrieved;
step 3.2.3: generating a model for each image to be queried, calculating the similarity between the model and all images in a data set to be retrieved, converting the similarity into the probability of selecting the images by using a softmax function, and selecting 500 image features with high similarity probability from the data set D to be retrieved according to the probability as the output of a generator;
step 3.2.4: and (5) maximizing the similarity between the query picture and the selected 500 pictures by using a logistic regression function, and iteratively optimizing the network weight of the generated model. Calculating the average accurate precision of all inquiry pictures according to the pictures output by the generator;
step 3.2.5: using a random gradient descent algorithm to minimize a loss function, iterating for 10 times, and storing a generated network model when the average accuracy of all query pictures is highest;
step 3.3: training a discrimination model;
step 3.3.1, setting the learning rate to be 0.0001;
step 3.3.1, taking the 500 picture features returned by the generator as the input of the discriminator, and carrying out weight optimization on the picture features again by using a three-layer full-connection discrimination model;
step 3.3.2, recalculating the distance between each inquiry picture and the returned 500 pictures, and giving a similarity score according to the distance;
step 3.3.3: minimizing the distance between the query picture and the returned K pictures by using a logistic regression function;
3.3.3, using an advanced gradient descent algorithm ADAM algorithm to minimize a loss function, iterating for 3 times, feeding back the similarity score calculated by the last discriminator to a generator, and directly acting on the optimization of the weight of the generator in a loss function weight form;
step 3.4: saving the optimal generator model as the output of the training;
step four, testing the precision, and the process is as follows:
step 4.1: sending the preprocessed test data set Q' into an optimal generator model;
step 4.2: aiming at a given query picture, a generator selects top500 pictures with highest similarity from a data set D to be retrieved;
step 4.3: comparing whether the tags of the inquired pictures are consistent with the tags of the K pictures returned by the generator or not, calculating the average accuracy of all the inquired pictures according to the evaluation criteria in the information retrieval and outputting test results;
through the operation of the steps, the unsupervised confrontation retrieval of the picture can be realized.
The above detailed description is intended to illustrate the objects, aspects and advantages of the present invention, and it should be understood that the above detailed description is only exemplary of the present invention, and is not intended to limit the scope of the present invention, and any modifications, equivalents, improvements, etc. made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (1)
1. A method for content-based image retrieval for unsupervised countermeasure training, the method comprising the steps of:
step one, network construction, the process is as follows:
step 1.1: the unsupervised countermeasure network framework consists of a generation model and a discrimination model, wherein the generation model and the discrimination model are both formed by three layers of full-connection networks;
step 1.2, a Relu activation function is connected behind a first full connection layer of the generated model;
step 1.3: a tanh activation function is connected behind a second full connection layer of the generated model, and the output is controlled to be {0, 1 };
step 1.4: generating a cosine distance measurement function after the third full-connection layer of the model is generated;
step 1.5: the first full-connection layer of the discriminant model is followed by Relu activation function
Step 1.6: a tanh activation function is connected behind the second full connection layer of the discrimination model, and the output is controlled to be {0, 1 };
step 1.7: the third full-connection layer of the discrimination model is connected with a similarity score function;
step 1.8: the discrimination model feeds back the calculated similarity score to the generation model;
step two, preprocessing the data set, wherein the process is as follows:
step 2.1: dividing the data into a query data set Q, a test data set Q' and a data set D to be retrieved, wherein a part of pictures are randomly extracted from the data set to be retrieved to serve as a data set F for fine-tuning network parameters when picture features are extracted;
step 2.2: extracting picture features by using a pre-trained VGG model on ImageNet, wherein a small amount of picture data sets F mentioned in the step 2.1 are required to be used for fine-tuning network parameters before the VGG is used for extracting the picture features;
step 2.3: inputting the pictures into an unsupervised countermeasure network in the form of feature vectors;
step three, network training, the process is as follows:
step 3.1: initializing a generation model and distinguishing model parameters by using random weight;
step 3.2: the process of training the generative model is as follows:
step 3.2.1, sending the picture characteristics of the query data set Q and the data set D to be retrieved, which are extracted by the VGG network, into a generation model;
step 3.2.2, generating a model to optimize the weight of the characteristics of the input query data set Q and the data set D to be retrieved;
step 3.2.3: generating a model for each image to be queried, calculating the cosine distance between the image to be queried and the image in the data set to be retrieved, converting the cosine distance into the probability of selecting the image by using a softmax function, and selecting K image features from the data set D to be retrieved as the output of a generator according to the probability;
step 3.2.4: maximizing the difference value between the similarity of the query picture and the selected K pictures and 1 by using a logic loss function;
step 3.3: the process of training the discriminant model is as follows:
step 3.3.1, taking the characteristics of the K pictures returned by the generator as the input of the discriminator, and carrying out weight optimization on the query picture and the characteristics of the K pictures returned by the generator again;
step 3.3.2, recalculating the cosine distance between each inquiry picture and the returned K pictures, and giving a similarity score according to the distance;
step 3.3.3, the judger feeds the calculated similarity score back to the generator, the similarity score is used for selecting the picture to be retrieved next by the generator, and the difference value between the distance between the inquiry picture and the returned K pictures and 0 is reduced by using a logistic regression function;
step 3.4: minimizing a loss function by using a random gradient descent algorithm;
step four, testing the precision, and the process is as follows:
step 4.1: sending the preprocessed test data set Q' into an optimal generator model;
step 4.2: the generator selects a picture with highest topK similarity from the data set D to be retrieved aiming at a given query picture;
step 4.3: comparing whether the tags of the inquired pictures are consistent with the tags of the K pictures returned by the generator or not, and calculating the average accuracy of all the inquired pictures according to the evaluation criteria in the information retrieval;
through the operation of the steps, the retrieval of the test picture can be realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810154813.4A CN108446334B (en) | 2018-02-23 | 2018-02-23 | Image retrieval method based on content for unsupervised countermeasure training |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810154813.4A CN108446334B (en) | 2018-02-23 | 2018-02-23 | Image retrieval method based on content for unsupervised countermeasure training |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108446334A CN108446334A (en) | 2018-08-24 |
| CN108446334B true CN108446334B (en) | 2021-08-03 |
Family
ID=63192735
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810154813.4A Active CN108446334B (en) | 2018-02-23 | 2018-02-23 | Image retrieval method based on content for unsupervised countermeasure training |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108446334B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543674B (en) * | 2018-10-19 | 2023-04-07 | 天津大学 | Image copy detection method based on generation countermeasure network |
| CN109635273B (en) * | 2018-10-25 | 2023-04-25 | 平安科技（深圳）有限公司 | Text keyword extraction method, device, equipment and storage medium |
| CN109785399B (en) * | 2018-11-19 | 2021-01-19 | 北京航空航天大学 | Synthetic lesion image generation method, device, equipment and readable storage medium |
| CN110287357B (en) * | 2019-05-31 | 2021-05-18 | 浙江工业大学 | Image description generation method for generating countermeasure network based on condition |
| CN112712094B (en) * | 2019-10-24 | 2024-08-02 | 北京四维图新科技股份有限公司 | Model training method, device, equipment and storage medium |
| CN113269256B (en) * | 2021-05-26 | 2024-08-27 | 广州密码营地信息科技有限公司 | Construction method and application of MiSrc-GAN medical image model |
| CN113887504B (en) * | 2021-10-22 | 2023-03-24 | 大连理工大学 | Strong-generalization remote sensing image target identification method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951919A (en) * | 2017-03-02 | 2017-07-14 | 浙江工业大学 | A kind of flow monitoring implementation method based on confrontation generation network |
| CN107563428A (en) * | 2017-08-25 | 2018-01-09 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on generation confrontation network |
| US10275473B2 (en) * | 2017-04-27 | 2019-04-30 | Sk Telecom Co., Ltd. | Method for learning cross-domain relations based on generative adversarial networks |
| CN107464210B (en) * | 2017-07-06 | 2020-02-21 | 浙江工业大学 | Image style migration method based on generating type countermeasure network |
-
2018
- 2018-02-23 CN CN201810154813.4A patent/CN108446334B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951919A (en) * | 2017-03-02 | 2017-07-14 | 浙江工业大学 | A kind of flow monitoring implementation method based on confrontation generation network |
| US10275473B2 (en) * | 2017-04-27 | 2019-04-30 | Sk Telecom Co., Ltd. | Method for learning cross-domain relations based on generative adversarial networks |
| CN107464210B (en) * | 2017-07-06 | 2020-02-21 | 浙江工业大学 | Image style migration method based on generating type countermeasure network |
| CN107563428A (en) * | 2017-08-25 | 2018-01-09 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on generation confrontation network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108446334A (en) | 2018-08-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108446334B (en) | Image retrieval method based on content for unsupervised countermeasure training | |
| US11195051B2 (en) | Method for person re-identification based on deep model with multi-loss fusion training strategy | |
| US20230108692A1 (en) | Semi-Supervised Person Re-Identification Using Multi-View Clustering | |
| CN110162593B (en) | Search result processing and similarity model training method and device | |
| US20220147836A1 (en) | Method and device for text-enhanced knowledge graph joint representation learning | |
| CN108416384B (en) | Image label labeling method, system, equipment and readable storage medium | |
| CN111353076B (en) | Method for training cross-modal retrieval model, cross-modal retrieval method and related device | |
| CN107657008B (en) | Cross-media training and retrieval method based on deep discrimination ranking learning | |
| Huang et al. | Cost-effective vehicle type recognition in surveillance images with deep active learning and web data | |
| CN113806746B (en) | Malicious code detection method based on improved CNN (CNN) network | |
| CN110222218B (en) | Image retrieval method based on multi-scale NetVLAD and depth hash | |
| CN108399185B (en) | Multi-label image binary vector generation method and image semantic similarity query method | |
| CN112819023A (en) | Sample set acquisition method and device, computer equipment and storage medium | |
| CN109635140B (en) | Image retrieval method based on deep learning and density peak clustering | |
| CN106503661B (en) | Face gender identification method based on fireworks deepness belief network | |
| CN114298122B (en) | Data classification method, apparatus, device, storage medium and computer program product | |
| CN112434628B (en) | Small sample image classification method based on active learning and collaborative representation | |
| CN114358188A (en) | Feature extraction model processing method, feature extraction model processing device, sample retrieval method, sample retrieval device and computer equipment | |
| CN109800768B (en) | Hash feature representation learning method of semi-supervised GAN | |
| CN110111365B (en) | Training method and device based on deep learning and target tracking method and device | |
| CN113806580B (en) | Cross-modal hash retrieval method based on hierarchical semantic structure | |
| CN112036511B (en) | Image retrieval method based on attention mechanism graph convolution neural network | |
| CN113537304A (en) | Cross-modal semantic clustering method based on bidirectional CNN | |
| CN112115806A (en) | Remote sensing image scene accurate classification method based on Dual-ResNet small sample learning | |
| CN110704665A (en) | Image feature expression method and system based on visual attention mechanism |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| OL01 | Intention to license declared | ||
| OL01 | Intention to license declared |