CN113298092B - Neural network training method and device for extracting multi-level image contour information - Google Patents
Neural network training method and device for extracting multi-level image contour information Download PDFInfo
- Publication number
- CN113298092B CN113298092B CN202110589188.8A CN202110589188A CN113298092B CN 113298092 B CN113298092 B CN 113298092B CN 202110589188 A CN202110589188 A CN 202110589188A CN 113298092 B CN113298092 B CN 113298092B
- Authority
- CN
- China
- Prior art keywords
- image
- training
- feature extraction
- feature
- contour
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a neural network training method and a device for extracting multi-level image contour information, wherein the method comprises the following steps: determining a network architecture of a feature extraction network training model; the feature extraction network training model comprises a feature extraction network model and a single-channel feature convolution layer connected to the output of the feature extraction network model; inputting the contour training image set into a feature extraction network training model to train until the first loss function of the feature extraction network training model converges, so as to obtain the trained feature extraction network model. Therefore, the network model obtained through training can be used for accurately extracting multi-channel and multi-layer contour information of the input image, so that the efficiency and the accuracy of the image recognition task can be improved when the image recognition task is carried out according to the contour information later.
Description
Technical Field
The invention relates to the technical field of neural networks, in particular to a neural network training method and device for extracting multi-level image contour information.
Background
In the existing business model, images are more attractive than characters, and the display and popularization effects are more remarkable. Therefore, the image starts to take on the function of more propaganda of goods or services, and in this case, how to extract the contour information of the image is of great importance in order to recognize the features of the image.
The existing image contour information extraction algorithm is generally only used for extracting single-level contour information of an image, and does not consider the advantages of multi-level image contour information in the field of image recognition, so that the existing image contour information extraction algorithm has defects and needs to be solved.
Disclosure of Invention
The invention aims to solve the technical problem of providing a neural network training determining method and device for extracting multi-level image contour information, which can enable a network model obtained by training to be used for accurately extracting multi-channel multi-level contour information of an input image so as to improve the efficiency and accuracy of the image recognition task when the image recognition task is carried out according to the contour information.
In order to solve the technical problem, the first aspect of the present invention discloses a neural network training method for extracting multi-level image contour information, which comprises the following steps:
Determining a network architecture of a feature extraction network training model; the feature extraction network training model comprises a feature extraction network model and a single-channel feature convolution layer connected to the output of the feature extraction network model; the characteristic extraction network model is used for outputting multi-channel and multi-layer first image contour characteristic information of an input image; the single-channel feature convolution layer is used for processing the first image contour feature information output by the feature extraction network model into single-channel image contour feature information;
inputting the contour training image set into a feature extraction network training model to train until the first loss function of the feature extraction network training model converges, so as to obtain the trained feature extraction network model.
As an optional implementation manner, in the first aspect of the present invention, the image profile feature information is multi-channel and multi-level image profile feature information; the feature extraction network model comprises a plurality of feature extraction layers for respectively extracting outline features of different levels and a corresponding plurality of dimension unification layers for unifying dimensions; inputting the contour training image set to a feature extraction network training model for training until a first loss function of the feature extraction network training model converges to obtain a trained feature extraction network model, wherein the method comprises the following steps of:
Inputting the contour training image set into a plurality of feature extraction layers to output a plurality of image contour features with different sizes;
inputting the image contour features output by each feature extraction layer to the corresponding dimension unification layer to obtain a plurality of image contour features with the same dimension;
Fusing the plurality of image contour features with the same size to obtain the first image contour feature information;
Inputting the first image contour feature information into the single-channel feature convolution layer to obtain single-channel image contour feature information;
Repeating the steps, updating the model parameters of the feature extraction network model based on back propagation until the first loss function converges, so as to obtain the trained feature extraction network model.
As an alternative embodiment, the contour training image set includes a plurality of annotated contour training images; the first loss function is a cross entropy loss between the single-channel image contour feature information and a corresponding contour training image.
As an alternative embodiment, in the first aspect of the present invention, the method further includes:
inputting the training image set into the trained feature extraction network model to perform feature extraction so as to obtain second image contour feature information;
inputting the training image set into a first gender classification network model to obtain image gender characteristic information;
fusing the second image contour feature information with the image sex feature information to obtain image fusion feature information;
Inputting the image fusion characteristic information into a second gender classification network model for training until convergence to obtain a target neural network model; the target neural network model is used for classifying the gender of the input image.
In an optional implementation manner, in a first aspect of the present invention, the inputting the image fusion feature information into the second gender classification network model to train until convergence, to obtain a target neural network model, includes:
Inputting the image fusion characteristic information into a second gender classification network model for training;
Based on the back propagation, model parameters of the first gender classification network model and/or the second gender classification network model are continuously updated until the second loss function converges to obtain a target neural network model.
As an optional implementation manner, in the first aspect of the present invention, the second loss function is a softmax loss of the gender prediction information output by the second gender classification network model and the real gender label of the corresponding training image.
As an alternative embodiment, in the first aspect of the present invention, the method further includes:
The training image set is processed using a data enhancement algorithm to obtain a training image set comprising more training images.
The second aspect of the present invention discloses a neural network training device for extracting multi-level image contour information, the device comprising:
The network determining module is used for determining the network architecture of the feature extraction network training model; the feature extraction network training model comprises a feature extraction network model and a single-channel feature convolution layer connected to the output of the feature extraction network model; the characteristic extraction network model is used for outputting multi-channel and multi-layer first image contour characteristic information of an input image; the single-channel feature convolution layer is used for processing the first image contour feature information output by the feature extraction network model into single-channel image contour feature information;
The network training module is used for inputting the contour training image set into the feature extraction network training model to train until the first loss function of the feature extraction network training model converges, so as to obtain the trained feature extraction network model.
In a second aspect of the present invention, as an optional implementation manner, the image profile feature information is multi-channel and multi-level image profile feature information; the feature extraction network model comprises a plurality of feature extraction layers for respectively extracting outline features of different levels and a corresponding plurality of dimension unification layers for unifying dimensions; the network training module inputs the contour training image set to a feature extraction network training model to train until a first loss function of the feature extraction network training model converges, so as to obtain a specific mode of the trained feature extraction network model, and the specific mode comprises the following steps:
Inputting the contour training image set into a plurality of feature extraction layers to output a plurality of image contour features with different sizes;
inputting the image contour features output by each feature extraction layer to the corresponding dimension unification layer to obtain a plurality of image contour features with the same dimension;
Fusing the plurality of image contour features with the same size to obtain the first image contour feature information;
Inputting the first image contour feature information into the single-channel feature convolution layer to obtain single-channel image contour feature information;
Repeating the steps, updating the model parameters of the feature extraction network model based on back propagation until the first loss function converges, so as to obtain the trained feature extraction network model.
As an alternative embodiment, the contour training image set includes a plurality of annotated contour training images; the first loss function is a cross entropy loss between the single-channel image contour feature information and a corresponding contour training image.
As an alternative embodiment, in the second aspect of the present invention, the apparatus further includes:
the contour extraction module is used for inputting the training image set into the trained feature extraction network model to perform feature extraction so as to obtain second image contour feature information;
the gender extraction module is used for inputting the training image set into a first gender classification network model so as to obtain image gender characteristic information;
the fusion module is used for fusing the second image contour characteristic information with the image sex characteristic information to obtain image fusion characteristic information;
The training module is used for inputting the image fusion characteristic information into a second gender classification network model for training until convergence to obtain a target neural network model; the target neural network model is used for classifying the gender of the input image.
In a second aspect of the present invention, as an optional implementation manner, the training module inputs the image fusion feature information to a second gender classification network model to perform training until convergence, to obtain a specific mode of the target neural network model, including:
Inputting the image fusion characteristic information into a second gender classification network model for training;
Based on the back propagation, model parameters of the first gender classification network model and/or the second gender classification network model are continuously updated until the second loss function converges to obtain a target neural network model.
As an optional implementation manner, in the second aspect of the present invention, the second loss function is a softmax loss of the gender prediction information output by the second gender classification network model and the real gender label of the corresponding training image.
As an alternative embodiment, in the second aspect of the present invention, the apparatus further includes:
and the data enhancement module is used for processing the training image set by using a data enhancement algorithm so as to obtain a training image set comprising more training images.
The third aspect of the present invention discloses another neural network training device for extracting multi-level image contour information, the device comprising:
a memory storing executable program code;
a processor coupled to the memory;
The processor invokes the executable program code stored in the memory to execute part or all of the steps in the neural network training method for extracting multi-level image contour information disclosed in the first aspect of the embodiment of the invention.
The fourth aspect of the embodiment of the invention discloses a computer storage medium, which stores computer instructions for executing part or all of the steps in the neural network training method for extracting multi-level image contour information disclosed in the first aspect of the embodiment of the invention when the computer instructions are called.
Compared with the prior art, the embodiment of the invention has the following beneficial effects:
In the embodiment of the invention, a network architecture of a feature extraction network training model is determined; the feature extraction network training model comprises a feature extraction network model and a single-channel feature convolution layer connected to the output of the feature extraction network model; the characteristic extraction network model is used for outputting multi-channel and multi-layer first image contour characteristic information of an input image; the single-channel feature convolution layer is used for processing the first image contour feature information output by the feature extraction network model into single-channel image contour feature information; inputting the contour training image set into a feature extraction network training model to train until the first loss function of the feature extraction network training model converges, so as to obtain the trained feature extraction network model. Therefore, the invention can extract multi-channel and multi-layer image contour feature information of the input image and process the multi-channel and multi-layer image contour feature information into single-channel feature information for model training, so that the trained network model can be used for accurately extracting the multi-channel and multi-layer contour information of the input image, thereby improving the efficiency and the accuracy of the image recognition task when the image recognition task is carried out according to the contour information.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a neural network training method for extracting multi-level image contour information according to an embodiment of the present invention;
FIG. 2 is a flowchart of another neural network training method for multi-level image profile information extraction according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a neural network training device for extracting multi-level image contour information according to an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of another neural network training device for multi-level image profile information extraction according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a neural network training device for extracting multi-level image contour information according to an embodiment of the present invention.
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The terms first, second and the like in the description and in the claims and in the above-described figures are used for distinguishing between different objects and not necessarily for describing a sequential or chronological order. Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, apparatus, article, or article that comprises a list of steps or elements is not limited to only those listed but may optionally include other steps or elements not listed or inherent to such process, method, article, or article.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the invention. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.
The invention discloses a neural network training method and device for extracting multi-level image contour information, which can extract multi-channel multi-level image contour characteristic information of an input image and process the multi-channel multi-level image contour characteristic information into single-channel characteristic information for model training, so that a network model obtained through training can be used for accurately extracting the multi-channel multi-level contour information of the input image, and the efficiency and the accuracy of an image recognition task can be improved when the image recognition task is carried out according to the contour information. The following will describe in detail.
Example 1
Referring to fig. 1, fig. 1 is a flowchart of a neural network training method for extracting multi-level image contour information according to an embodiment of the present invention. The method described in fig. 1 is applied to a training device of a neural network model, where the training device may be a corresponding training terminal, training device or server, and the server may be a local server or a cloud server, which is not limited by the embodiment of the present invention. As shown in fig. 1, the neural network training method for multi-level image profile information extraction may include the following operations:
101. a network architecture of the feature extraction network training model is determined.
In the embodiment of the invention, the feature extraction network training model comprises a feature extraction network model and a single-channel feature convolution layer connected to the output of the feature extraction network model, wherein the single-channel feature convolution layer is used for processing multi-channel and multi-layer image contour feature information output by the feature extraction network model into single-channel image contour feature information.
102. Inputting the contour training image set into the feature extraction network training model to train until the first loss function of the feature extraction network training model converges, so as to obtain a trained feature extraction network model.
In the embodiment of the invention, the final training is to obtain the feature extraction network model, namely the feature extraction network training model without the single-channel feature convolution layer, because the feature extraction network model is finally needed in the scheme of the invention to extract the multi-channel and multi-layer contour features of the image, and the single-channel feature convolution layer is only used for fusing the multi-channel and multi-layer contour features during training to facilitate subsequent loss calculation and can be discarded after training is finished.
Therefore, the method described by the embodiment of the invention can extract multi-channel and multi-layer image contour feature information of the input image and process the multi-channel and multi-layer image contour feature information into single-channel feature information for model training, so that the trained network model can be used for accurately extracting the multi-channel and multi-layer contour information of the input image, and the efficiency and the accuracy of the image recognition task can be improved when the image recognition task is carried out according to the contour information.
In an alternative embodiment, the image profile feature information is multi-channel and multi-level image profile feature information, and the feature extraction network model includes a plurality of feature extraction layers for extracting profile features of different levels respectively and a corresponding plurality of dimension unification layers for unifying dimensions.
Therefore, by implementing the alternative implementation mode, the feature extraction network model can be used for extracting contour features of different layers of a plurality of channels of an image, and the sizes of a plurality of feature images are unified through a size unifying layer so as to obtain multi-layer image contour feature information, and the feature extraction network model can be used for more accurately characterizing the contour information of the image, so that the efficiency and the accuracy of the image recognition task can be improved when the image recognition task is carried out according to the contour information.
In another alternative embodiment, in step 102, inputting the contour training image set into the feature extraction network training model for training until the first loss function of the feature extraction network training model converges to obtain a trained feature extraction network model includes:
Inputting the contour training image set into a plurality of feature extraction layers to output a plurality of image contour features with different sizes;
Inputting the image contour features output by each feature extraction layer into a corresponding dimension unification layer to obtain a plurality of image contour features with the same dimension;
Fusing a plurality of image contour features with the same size to obtain first image contour feature information;
inputting the first image contour feature information into a single-channel feature convolution layer to obtain single-channel image contour feature information;
Repeating the steps, and updating the model parameters of the feature extraction network model based on the back propagation until the first loss function converges to obtain the trained feature extraction network model.
In the embodiment of the invention, the plurality of feature extraction layers are a plurality of sequentially cascaded convolution layers, wherein the output of each convolution layer is connected to a corresponding dimension unification layer. Optionally, the feature extraction layer is a convolution layer including a residual module, where the residual module is used to enhance the feature extraction and back propagation capability, and may be ResNet, denseNet or SENet, and finally each feature extraction layer outputs a feature map with a different size, so as to obtain a feature map with multiple sizes. The shallow feature map is used for capturing fine texture features of the clothing image, and the deep feature map is used for capturing outline features. In an alternative embodiment, the convolution layer may be a3×3 convolution layer and a residual module in cascade.
Optionally, the dimension unifying layer may be an interpolation module, and optionally, the interpolation module interpolates the dimension unification of the image profile feature output by the feature extraction layer to the same dimension by adopting an interpolation algorithm. Alternatively, the interpolation algorithm may be one or more of bilinear interpolation, nearest neighbor interpolation, and deconvolution layers, which is not limited by the present invention.
In an embodiment of the present invention, optionally, the contour training image set includes a plurality of labeled contour training images, where the labeled contour training images are training images that are labeled manually for contours in the images. Optionally, the first loss function is a cross entropy loss between the image contour feature information and the corresponding contour training image of the single channel.
Therefore, the implementation of the optional implementation mode can train the feature extraction network training model until the first loss function of the feature extraction network training model is converged, so that the feature extraction network model for extracting multi-channel and multi-layer profile feature information can be obtained, and the efficiency and the accuracy of the image recognition task can be improved when the image recognition task is carried out according to the profile information.
Example two
Referring to fig. 2, fig. 2 is a flowchart of another neural network training method for extracting multi-level image contour information according to an embodiment of the present invention. The method described in fig. 2 is applied to a training device of a neural network model, where the training device may be a corresponding training terminal, training device or server, and the server may be a local server or a cloud server, which is not limited by the embodiment of the present invention. As shown in fig. 2, the neural network training method for multi-level image profile information extraction may include the following operations:
201. a network architecture of the feature extraction network training model is determined.
202. Inputting the contour training image set into the feature extraction network training model to train until the first loss function of the feature extraction network training model converges, so as to obtain a trained feature extraction network model.
Specific technical details and explanation of the technical terms of the steps 201 to 202 may refer to the description of the steps 101 to 102 in the implementation of the step, and will not be repeated here.
203. And inputting the training image set into the trained feature extraction network model to perform feature extraction so as to obtain second image contour feature information.
In the embodiment of the invention, the training image set may include a plurality of training images that are manually marked, and the manual marking is used for achieving the purpose of gender classification, and the gender label of the training image is set manually, where the gender label may be one or more of male and female, and is not limited herein.
In the embodiment of the present invention, the second image contour feature is used to represent the outer edge contour information of the image, which may be one or more of the combination of the overall closed contour feature of the image, the edge contour feature of the image, and the texture feature of the image.
204. The training image set is input into a first gender classification network model to obtain image gender characteristic information.
In an embodiment of the present invention, the first gender classification network model is a gender classification network model based on the first gender classification network model, which is used for extracting high-level semantic features for representing gender information in the training images. Alternatively, the first gender classification network model may be a combination of one or more of the EFFICIENTNET, SHUFFLENET, RESNET, MOBILENET or other convolutional neural classification networks. Alternatively, the first gender-classifying network model may be a gender-classifying network model that is trained in advance using an image dataset, such as an ImageNet dataset, and in an alternative embodiment, the first gender-classifying network model receives RGB three-channel clothing images as input, and finally extracts high-dimensional gender-classifying features.
205. And fusing the second image contour feature information with the image sex feature information to obtain image fusion feature information.
In the embodiment of the invention, the image contour feature information and the image sex feature information are fused, so that the image contour feature information and the image sex feature information can be spliced in the dimension of each channel, and/or the image contour feature information is obtained by carrying out feature fusion through a classification feature fusion layer formed by a plurality of convolution layers. Alternatively, the classification feature fusion layer may be composed of a plurality of concatenated nxn convolutional layers, for example, two concatenated 3×3 convolutional layers.
206. And inputting the image fusion characteristic information into a second gender classification network model for training until convergence, and obtaining a target neural network model.
In the embodiment of the invention, the target neural network model is used for classifying the gender of the input image. Optionally, the architecture of the target neural network model includes the feature extraction network model, the first gender classification network model and the second gender classification network model, and the data processing flow is similar to the training steps, so that those skilled in the art are familiar with the training steps and the actual prediction steps of the neural network model, and the technical details are the same or similar, and are not repeated here. Alternatively, the second gender classification network model may be a fully connected layer for gender classification, and reference may be made to the technical details of the first gender classification network model.
Therefore, the method described by implementing the embodiment of the invention can simultaneously extract the outline characteristic information and the sex characteristic information of the training image through the two-way network model, and train the sex classification model by combining the characteristics of the two information fusion, so that the outline information of the image can be introduced into the sex classification network training of the image, the sex classification accuracy of the network model obtained by subsequent training is improved, and meanwhile, compared with the existing sex classification network model training method, the model complexity is greatly reduced, the convergence speed is faster, and the cost of manpower and material resources is lower.
In an alternative embodiment, the image profile feature information is multi-channel and multi-level image profile feature information, and the feature extraction network model includes a plurality of feature extraction layers for extracting profile features of different levels respectively and a corresponding plurality of dimension unification layers for unifying dimensions.
It can be seen that by implementing this alternative embodiment, the feature extraction network model may be used to extract contour features of different levels of multiple channels of an image, and unify the sizes of multiple feature maps through a unified size layer, so as to obtain multi-level image contour feature information, which may be used to more accurately characterize contour information of the image, and subsequently, when this contour information is introduced into training of the target neural network, accuracy of determining, by the target neural network, image sex information based on the contour of the image may be improved.
In another optional embodiment, in step 203, inputting the training image set into the feature extraction network model for feature extraction to obtain the second image contour feature information may include:
Inputting the training image set into a plurality of feature extraction layers to output a plurality of image contour features of different sizes;
Inputting the image contour features output by each feature extraction layer into a corresponding dimension unification layer to obtain a plurality of image contour features with the same dimension;
A plurality of image contour features of the same size are determined as second image contour feature information.
Therefore, the implementation of the alternative implementation mode can obtain multi-level image contour feature information, the multi-level image contour feature information can be used for more accurately representing the contour information of an image, and the accuracy of judging the sex information of the image based on the contour of the image by the target neural network can be improved when the contour information is subsequently introduced into the training of the target neural network.
In yet another alternative embodiment, in step 206, inputting the image fusion feature information into the second gender classification network model for training until convergence, obtaining the target neural network model includes:
inputting the image fusion characteristic information into a second gender classification network model for training;
Based on the back propagation, model parameters of the first gender classification network model and/or the second gender classification network model are continuously updated until the second loss function converges to obtain the target neural network model.
In the embodiment of the present invention, it is preferable that the model parameters of the first gender classification network model and the second gender classification network model be selectively updated, but the model parameters of the feature extraction network model are not updated, because the feature extraction network model in the present model is trained in another manner, and its function is only used for extracting the image contour features, and if training it affects the characterization capability of the image contour features extracted later.
Optionally, the gender prediction information finally output by the second gender classification network model may be a predicted gender label corresponding to the training image and a corresponding confidence level. Optionally, the second loss function is a softmax loss of the gender prediction information output by the second gender classification network model and the real gender label of the corresponding training image, and model parameters of the first gender classification network model and/or the second gender classification network model are continuously updated by calculating back propagation gradient information obtained by the loss, so that the second loss function converges.
Therefore, according to the alternative implementation mode, based on back propagation, the model parameters of the first gender classification network model and/or the second gender classification network model are continuously updated until the second loss function converges to obtain the trained neural network model, and the model parameters of the feature extraction network model can not be updated in training, so that the convergence rate in training is improved, the workload is reduced, and on the other hand, the network model can pay more attention to the accuracy of gender classification rather than the accuracy of contour extraction, so that a better gender classification prediction effect is achieved.
In yet another alternative embodiment, prior to step 203, the method further comprises:
the training image set is processed using a data enhancement algorithm to obtain a training image set comprising more training images.
In the embodiment of the invention, the data enhancement algorithm may be an offline data enhancement algorithm or an online data enhancement algorithm, which may be a data enhancement method for transforming information such as size, direction, color, resolution of a training image, for example, one or more of processing operations such as flipping, rotation, clipping, scaling, translation, affine transformation, adding noise, brightness enhancement, contrast enhancement, sharpening, etc.,
Therefore, the optional implementation manner can process the training image set by using the data enhancement algorithm to obtain the training image set comprising more training images, so that the body volume of the training image data is increased under the condition of reducing the workload, the degree of model training is further improved, and the prediction accuracy of the trained model is improved.
In another alternative embodiment, in the step, the processing the training image set using a data enhancement algorithm to obtain a training image set including more training images includes:
Color information of one or more training images in the training image set is transformed to obtain a training image set comprising more training images.
Alternatively, the manner of transforming the color information may include randomly exchanging color channels, and randomly changing one or a combination of two of the characteristic values of a specific channel, which is not limited by the present invention. Optionally, the degree of color information transformation should be smaller than a preset threshold, for example, the number of pictures of the color channels to be exchanged should be smaller than a number threshold, or the difference value of the feature value of a specific channel should be smaller than a difference threshold, so as to prevent the color data distribution affecting the whole clothing image from being affected, resulting in the degradation of model accuracy.
It can be seen that this alternative embodiment can reduce the likelihood that the final trained gender classification model directly depends on the color information to output gender categories, thereby improving generalization of the model.
It should be noted that the method described in the above or the following embodiments of the present invention may be specifically applied to the field of multi-level image contour information extraction of clothing images, where the area uses contour information to characterize gender or clothing materials, and the corresponding training images or contour training images in the present invention may be clothing merchandise images, however, alternatively, other merchandise images or service image areas may be applied to the method described in the present invention, which is not limited in particular, but only describes the above preferred situation.
Example III
Referring to fig. 3, fig. 3 is a schematic structural diagram of a neural network training device for extracting multi-level image contour information according to an embodiment of the invention. The apparatus described in fig. 3 may be applied to a corresponding training terminal, training device, or server, and the server may be a local server or a cloud server, which is not limited by the embodiment of the present invention. As shown in fig. 3, the apparatus may include:
The network determining module 301 is configured to determine a network architecture of the feature extraction network training model.
In the embodiment of the invention, the feature extraction network training model comprises a feature extraction network model and a single-channel feature convolution layer connected to the output of the feature extraction network model, wherein the single-channel feature convolution layer is used for processing multi-channel and multi-layer image contour feature information output by the feature extraction network model into single-channel image contour feature information.
The network training module 302 is configured to input the contour training image set to the feature extraction network training model for training until the first loss function of the feature extraction network training model converges, so as to obtain a trained feature extraction network model.
In the embodiment of the invention, the final training is to obtain the feature extraction network model, namely the feature extraction network training model without the single-channel feature convolution layer, because the feature extraction network model is finally needed in the scheme of the invention to extract the multi-channel and multi-layer contour features of the image, and the single-channel feature convolution layer is only used for fusing the multi-channel and multi-layer contour features during training to facilitate subsequent loss calculation and can be discarded after training is finished.
Therefore, the method described by the embodiment of the invention can extract multi-channel and multi-layer image contour feature information of the input image and process the multi-channel and multi-layer image contour feature information into single-channel feature information for model training, so that the trained network model can be used for accurately extracting the multi-channel and multi-layer contour information of the input image, and the efficiency and the accuracy of the image recognition task can be improved when the image recognition task is carried out according to the contour information.
As an alternative implementation manner, the image contour feature information is multi-channel and multi-level image contour feature information, and the corresponding feature extraction network model comprises a plurality of feature extraction layers for respectively extracting contour features of different levels and a corresponding plurality of dimension unification layers for unifying dimensions.
Therefore, by implementing the alternative implementation mode, the feature extraction network model can be used for extracting contour features of different layers of a plurality of channels of an image, and the sizes of a plurality of feature images are unified through a size unifying layer so as to obtain multi-layer image contour feature information, and the feature extraction network model can be used for more accurately characterizing the contour information of the image, so that the efficiency and the accuracy of the image recognition task can be improved when the image recognition task is carried out according to the contour information.
As an alternative implementation manner, the network training module 302 inputs the contour training image set to the feature extraction network training model to train until the first loss function of the feature extraction network training model converges, so as to obtain a specific mode of the trained feature extraction network model, which includes:
Inputting the contour training image set into a plurality of feature extraction layers to output a plurality of image contour features with different sizes;
Inputting the image contour features output by each feature extraction layer into a corresponding dimension unification layer to obtain a plurality of image contour features with the same dimension;
Fusing a plurality of image contour features with the same size to obtain first image contour feature information;
inputting the first image contour feature information into a single-channel feature convolution layer to obtain single-channel image contour feature information;
Repeating the steps, and updating the model parameters of the feature extraction network model based on the back propagation until the first loss function converges to obtain the trained feature extraction network model.
In the embodiment of the invention, the plurality of feature extraction layers are a plurality of sequentially cascaded convolution layers, wherein the output of each convolution layer is connected to a corresponding dimension unification layer. Optionally, the feature extraction layer is a convolution layer including a residual module, where the residual module is used to enhance the feature extraction and back propagation capability, and may be ResNet, denseNet or SENet, and finally each feature extraction layer outputs a feature map with a different size, so as to obtain a feature map with multiple sizes. The shallow feature map is used for capturing fine texture features of the clothing image, and the deep feature map is used for capturing outline features. In an alternative embodiment, the convolution layer may be a3×3 convolution layer and a residual module in cascade.
Optionally, the dimension unifying layer may be an interpolation module, and optionally, the interpolation module interpolates the dimension unification of the image profile feature output by the feature extraction layer to the same dimension by adopting an interpolation algorithm. Alternatively, the interpolation algorithm may be one or more of bilinear interpolation, nearest neighbor interpolation, and deconvolution layers, which is not limited by the present invention.
In an embodiment of the present invention, optionally, the contour training image set includes a plurality of labeled contour training images, where the labeled contour training images are training images that are labeled manually for contours in the images. Optionally, the first loss function is a cross entropy loss between the image contour feature information and the corresponding contour training image of the single channel.
Therefore, the implementation of the optional implementation mode can train the feature extraction network training model until the first loss function of the feature extraction network training model is converged, so that the feature extraction network model for extracting multi-channel and multi-layer profile feature information can be obtained, and the efficiency and the accuracy of the image recognition task can be improved when the image recognition task is carried out according to the profile information.
As an alternative embodiment, as shown in fig. 4, the apparatus further includes:
the contour extraction module 303 is configured to input the training image set to the trained feature extraction network model for feature extraction, so as to obtain second image contour feature information.
In the embodiment of the invention, the training image set may include a plurality of training images that are manually marked, and the manual marking is used for achieving the purpose of gender classification, and the gender label of the training image is set manually, where the gender label may be one or more of male and female, and is not limited herein.
In the embodiment of the present invention, the second image contour feature is used to represent the outer edge contour information of the image, which may be one or more of the combination of the overall closed contour feature of the image, the edge contour feature of the image, and the texture feature of the image.
The gender extraction module 304 is configured to input the training image set to the first gender classification network model to obtain the image gender characteristic information.
In an embodiment of the present invention, the first gender classification network model is a gender classification network model based on the first gender classification network model, which is used for extracting high-level semantic features for representing gender information in the training images. Alternatively, the first gender classification network model may be a combination of one or more of the EFFICIENTNET, SHUFFLENET, RESNET, MOBILENET or other convolutional neural classification networks. Alternatively, the first gender-classifying network model may be a gender-classifying network model that is trained in advance using an image dataset, such as an ImageNet dataset, and in an alternative embodiment, the first gender-classifying network model receives RGB three-channel clothing images as input, and finally extracts high-dimensional gender-classifying features.
And the fusion module 305 is configured to fuse the second image contour feature information with the image gender feature information to obtain image fusion feature information.
In the embodiment of the invention, the image contour feature information and the image sex feature information are fused, so that the image contour feature information and the image sex feature information can be spliced in the dimension of each channel, and/or the image contour feature information is obtained by carrying out feature fusion through a classification feature fusion layer formed by a plurality of convolution layers. Alternatively, the classification feature fusion layer may be composed of a plurality of concatenated nxn convolutional layers, for example, two concatenated 3×3 convolutional layers.
And the training module 306 is configured to input the image fusion feature information to the second gender classification network model for training until convergence, so as to obtain the target neural network model.
In the embodiment of the invention, the target neural network model is used for classifying the gender of the input image. Optionally, the architecture of the target neural network model includes the feature extraction network model, the first gender classification network model and the second gender classification network model, and the data processing flow is similar to the training steps, so that those skilled in the art are familiar with the training steps and the actual prediction steps of the neural network model, and the technical details are the same or similar, and are not repeated here. Alternatively, the second gender classification network model may be a fully connected layer for gender classification, and reference may be made to the technical details of the first gender classification network model.
Therefore, the implementation of the alternative implementation mode can simultaneously extract the outline characteristic information and the sex characteristic information of the training image through the two-way network model, and the sex classification model is trained by combining the characteristics of the two information fusion, so that the outline information of the image can be introduced into the sex classification network training of the image, the sex classification accuracy of the network model obtained by subsequent training is improved, and meanwhile, compared with the existing sex classification network model training method, the model complexity is greatly reduced, the convergence speed is higher, and the cost of manpower and material resources is lower.
As an alternative implementation manner, the image contour feature information is multi-channel and multi-level image contour feature information, and the corresponding feature extraction network model comprises a plurality of feature extraction layers for respectively extracting contour features of different levels and a corresponding plurality of dimension unification layers for unifying dimensions.
It can be seen that by implementing this alternative embodiment, the feature extraction network model may be used to extract contour features of different levels of multiple channels of an image, and unify the sizes of multiple feature maps through a unified size layer, so as to obtain multi-level image contour feature information, which may be used to more accurately characterize contour information of the image, and subsequently, when this contour information is introduced into training of the target neural network, accuracy of determining, by the target neural network, image sex information based on the contour of the image may be improved.
As an optional implementation manner, the specific manner in which the contour extraction module 303 inputs the training image set to the feature extraction network model to perform feature extraction to obtain the second image contour feature information may include:
Inputting the training image set into a plurality of feature extraction layers to output a plurality of image contour features of different sizes;
Inputting the image contour features output by each feature extraction layer into a corresponding dimension unification layer to obtain a plurality of image contour features with the same dimension;
A plurality of image contour features of the same size are determined as second image contour feature information.
Therefore, the implementation of the alternative implementation mode can obtain multi-level image contour feature information, the multi-level image contour feature information can be used for more accurately representing the contour information of an image, and the accuracy of judging the sex information of the image based on the contour of the image by the target neural network can be improved when the contour information is subsequently introduced into the training of the target neural network.
As an optional implementation manner, the training module 306 inputs the image fusion feature information into the second gender classification network model to perform training until convergence, to obtain a specific mode of the target neural network model, which includes:
inputting the image fusion characteristic information into a second gender classification network model for training;
Based on the back propagation, model parameters of the first gender classification network model and/or the second gender classification network model are continuously updated until the second loss function converges to obtain the target neural network model.
In the embodiment of the present invention, it is preferable that the model parameters of the first gender classification network model and the second gender classification network model be selectively updated, but the model parameters of the feature extraction network model are not updated, because the feature extraction network model in the present model is trained in another manner, and its function is only used for extracting the image contour features, and if training it affects the characterization capability of the image contour features extracted later.
Optionally, the gender prediction information finally output by the second gender classification network model may be a predicted gender label corresponding to the training image and a corresponding confidence level. Optionally, the second loss function is a softmax loss of the gender prediction information output by the second gender classification network model and the real gender label of the corresponding training image, and model parameters of the first gender classification network model and/or the second gender classification network model are continuously updated by calculating back propagation gradient information obtained by the loss, so that the second loss function converges.
Therefore, according to the alternative implementation mode, based on back propagation, the model parameters of the first gender classification network model and/or the second gender classification network model are continuously updated until the second loss function converges to obtain the trained neural network model, and the model parameters of the feature extraction network model can not be updated in training, so that the convergence rate in training is improved, the workload is reduced, and on the other hand, the network model can pay more attention to the accuracy of gender classification rather than the accuracy of contour extraction, so that a better gender classification prediction effect is achieved.
As an alternative embodiment, as shown in fig. 4, the apparatus further includes:
the data enhancement module 307 is configured to process the training image set using a data enhancement algorithm to obtain a training image set including more training images.
In the embodiment of the invention, the data enhancement algorithm may be an offline data enhancement algorithm or an online data enhancement algorithm, which may be a data enhancement method for transforming information such as size, direction, color, resolution of a training image, for example, one or more of processing operations such as flipping, rotation, clipping, scaling, translation, affine transformation, adding noise, brightness enhancement, contrast enhancement, sharpening, etc.,
Therefore, the optional implementation manner can process the training image set by using the data enhancement algorithm to obtain the training image set comprising more training images, so that the body volume of the training image data is increased under the condition of reducing the workload, the degree of model training is further improved, and the prediction accuracy of the trained model is improved.
As an alternative embodiment, the data enhancement module 307 processes the training image set using a data enhancement algorithm to obtain a specific manner of training image set that includes more training images, including:
Color information of one or more training images in the training image set is transformed to obtain a training image set comprising more training images.
Alternatively, the manner of transforming the color information may include randomly exchanging color channels, and randomly changing one or a combination of two of the characteristic values of a specific channel, which is not limited by the present invention. Optionally, the degree of color information transformation should be smaller than a preset threshold, for example, the number of pictures of the color channels to be exchanged should be smaller than a number threshold, or the difference value of the feature value of a specific channel should be smaller than a difference threshold, so as to prevent the color data distribution affecting the whole clothing image from being affected, resulting in the degradation of model accuracy.
It can be seen that this alternative embodiment can reduce the likelihood that the final trained gender classification model directly depends on the color information to output gender categories, thereby improving generalization of the model.
Example IV
Referring to fig. 5, fig. 5 is a schematic structural diagram of a neural network training device for extracting multi-level image contour information according to an embodiment of the invention. As shown in fig. 5, the apparatus may include:
a memory 401 storing executable program codes;
A processor 402 coupled with the memory 401;
The processor 402 invokes the executable program code stored in the memory 401 to perform some or all of the steps in the neural network training method for multi-level image profile information extraction disclosed in the first or second embodiment of the present invention.
Example five
The embodiment of the invention discloses a computer storage medium which stores computer instructions for executing part or all of the steps in the neural network training method for extracting multi-level image contour information disclosed in the first embodiment or the second embodiment of the invention when the computer instructions are called.
The apparatus embodiments described above are merely illustrative, wherein the modules illustrated as separate components may or may not be physically separate, and the components shown as modules may or may not be physical, i.e., may be located in one place, or may be distributed over a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above detailed description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course by means of hardware. Based on such understanding, the foregoing technical solutions may be embodied essentially or in part in the form of a software product that may be stored in a computer-readable storage medium including Read-Only Memory (ROM), random access Memory (Random Access Memory, RAM), programmable Read-Only Memory (Programmable Read-Only Memory, PROM), erasable programmable Read-Only Memory (Erasable Programmable Read Only Memory, EPROM), one-time programmable Read-Only Memory (OTPROM), electrically erasable programmable Read-Only Memory (EEPROM), compact disc Read-Only Memory (Compact Disc Read-Only Memory, CD-ROM) or other optical disc Memory, magnetic disc Memory, tape Memory, or any other medium that can be used for computer-readable carrying or storing data.
Finally, it should be noted that: the embodiment of the invention discloses a neural network training method and device for extracting multi-level image contour information, which are disclosed by the embodiment of the invention only for illustrating the technical scheme of the invention, but not limiting the technical scheme; although the invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art will understand that; the technical scheme recorded in the various embodiments can be modified or part of technical features in the technical scheme can be replaced equivalently; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions.
Claims (9)
1. A neural network training method for multi-level image contour information extraction, the method comprising:
Determining a network architecture of a feature extraction network training model; the feature extraction network training model comprises a feature extraction network model and a single-channel feature convolution layer connected to the output of the feature extraction network model; the characteristic extraction network model is used for outputting multi-channel and multi-layer first image contour characteristic information of an input image; the single-channel feature convolution layer is used for processing the first image contour feature information output by the feature extraction network model into single-channel image contour feature information;
Inputting the contour training image set into a feature extraction network training model to train until a first loss function of the feature extraction network training model converges to obtain a trained feature extraction network model;
The feature extraction network model comprises a plurality of feature extraction layers for respectively extracting outline features of different levels and a corresponding plurality of dimension unification layers for unifying dimensions; inputting the contour training image set to a feature extraction network training model for training until a first loss function of the feature extraction network training model converges to obtain a trained feature extraction network model, wherein the method comprises the following steps of:
Inputting the contour training image set into a plurality of feature extraction layers to output a plurality of image contour features with different sizes;
inputting the image contour features output by each feature extraction layer to the corresponding dimension unification layer to obtain a plurality of image contour features with the same dimension;
Fusing the plurality of image contour features with the same size to obtain the first image contour feature information;
Inputting the first image contour feature information into the single-channel feature convolution layer to obtain single-channel image contour feature information;
Repeating the steps, updating the model parameters of the feature extraction network model based on back propagation until the first loss function converges, so as to obtain the trained feature extraction network model.
2. The neural network training method for multi-level image profile information extraction of claim 1, wherein the profile training image set comprises a plurality of annotated profile training images; the first loss function is a cross entropy loss between the single-channel image contour feature information and a corresponding contour training image.
3. The neural network training method for multi-level image profile information extraction of claim 1, further comprising:
inputting the training image set into the trained feature extraction network model to perform feature extraction so as to obtain second image contour feature information;
inputting the training image set into a first gender classification network model to obtain image gender characteristic information;
fusing the second image contour feature information with the image sex feature information to obtain image fusion feature information;
Inputting the image fusion characteristic information into a second gender classification network model for training until convergence to obtain a target neural network model; the target neural network model is used for classifying the gender of the input image.
4. The neural network training method for extracting multi-level image contour information according to claim 3, wherein said inputting the image fusion feature information into the second gender classification network model for training until convergence, obtaining the target neural network model, comprises:
Inputting the image fusion characteristic information into a second gender classification network model for training;
Based on the back propagation, model parameters of the first gender classification network model and/or the second gender classification network model are continuously updated until the second loss function converges to obtain a target neural network model.
5. The neural network training method for multi-level image profile information extraction of claim 4, wherein the second loss function is a softmax loss of gender prediction information output by the second gender classification network model and the real gender label of the corresponding training image.
6. The neural network training method for multi-level image profile information extraction of claim 3, further comprising:
The training image set is processed using a data enhancement algorithm to obtain a training image set comprising more training images.
7. A neural network training device for multi-level image profile information extraction, the device comprising:
The network determining module is used for determining the network architecture of the feature extraction network training model; the feature extraction network training model comprises a feature extraction network model and a single-channel feature convolution layer connected to the output of the feature extraction network model; the characteristic extraction network model is used for outputting multi-channel and multi-layer first image contour characteristic information of an input image; the single-channel feature convolution layer is used for processing the first image contour feature information output by the feature extraction network model into single-channel image contour feature information;
The network training module is used for inputting the contour training image set into the feature extraction network training model to train until the first loss function of the feature extraction network training model converges so as to obtain the trained feature extraction network model;
The feature extraction network model comprises a plurality of feature extraction layers for respectively extracting outline features of different levels and a corresponding plurality of dimension unification layers for unifying dimensions; the network training module inputs the contour training image set to a feature extraction network training model to train until a first loss function of the feature extraction network training model converges, so as to obtain a specific mode of the trained feature extraction network model, and the specific mode comprises the following steps:
Inputting the contour training image set into a plurality of feature extraction layers to output a plurality of image contour features with different sizes;
inputting the image contour features output by each feature extraction layer to the corresponding dimension unification layer to obtain a plurality of image contour features with the same dimension;
Fusing the plurality of image contour features with the same size to obtain the first image contour feature information;
Inputting the first image contour feature information into the single-channel feature convolution layer to obtain single-channel image contour feature information;
Repeating the steps, updating the model parameters of the feature extraction network model based on back propagation until the first loss function converges, so as to obtain the trained feature extraction network model.
8. A neural network training device for multi-level image profile information extraction, the device comprising:
a memory storing executable program code;
a processor coupled to the memory;
the processor invokes the executable program code stored in the memory to perform the neural network training method for multi-level image profile information extraction as claimed in any one of claims 1-6.
9. A computer storage medium storing computer instructions which, when invoked, are operable to perform the neural network training method for multi-level image profile information extraction of any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110589188.8A CN113298092B (en) | 2021-05-28 | 2021-05-28 | Neural network training method and device for extracting multi-level image contour information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110589188.8A CN113298092B (en) | 2021-05-28 | 2021-05-28 | Neural network training method and device for extracting multi-level image contour information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113298092A CN113298092A (en) | 2021-08-24 |
| CN113298092B true CN113298092B (en) | 2024-06-28 |
Family
ID=77325857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110589188.8A Active CN113298092B (en) | 2021-05-28 | 2021-05-28 | Neural network training method and device for extracting multi-level image contour information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113298092B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114358243B (en) * | 2021-12-14 | 2025-04-22 | 有米科技股份有限公司 | General feature extraction network training method, device and general feature extraction network |
| CN115841414A (en) * | 2022-09-17 | 2023-03-24 | 海南大学 | Medical image robust zero watermarking method based on Residual-DenseNet |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109196518A (en) * | 2018-08-23 | 2019-01-11 | 合刃科技（深圳）有限公司 | A kind of gesture identification method and device based on high light spectrum image-forming |
| CN110929697A (en) * | 2019-12-17 | 2020-03-27 | 中国人民解放军海军航空大学 | Neural network target identification method and system based on residual error structure |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8275172B2 (en) * | 2009-10-30 | 2012-09-25 | Raytheon Applied Signal Technology, Inc. | Multipass data integration for automatic detection and classification of objects |
| CN111210435B (en) * | 2019-12-24 | 2022-10-18 | 重庆邮电大学 | Image semantic segmentation method based on local and global feature enhancement module |
| CN111369440B (en) * | 2020-03-03 | 2024-01-30 | 网易（杭州）网络有限公司 | Model training and image super-resolution processing method, device, terminal and storage medium |
| CN112001403B (en) * | 2020-08-11 | 2023-12-15 | 北京化工大学 | An image contour detection method and system |
| CN112101190B (en) * | 2020-09-11 | 2023-11-03 | 西安电子科技大学 | A remote sensing image classification method, storage medium and computing device |
| CN112364699B (en) * | 2020-10-14 | 2024-08-02 | 珠海欧比特宇航科技股份有限公司 | Remote sensing image segmentation method, device and medium based on weighted loss fusion network |
| CN112598045A (en) * | 2020-12-17 | 2021-04-02 | 中国工商银行股份有限公司 | Method for training neural network, image recognition method and image recognition device |
-
2021
- 2021-05-28 CN CN202110589188.8A patent/CN113298092B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109196518A (en) * | 2018-08-23 | 2019-01-11 | 合刃科技（深圳）有限公司 | A kind of gesture identification method and device based on high light spectrum image-forming |
| CN110929697A (en) * | 2019-12-17 | 2020-03-27 | 中国人民解放军海军航空大学 | Neural network target identification method and system based on residual error structure |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113298092A (en) | 2021-08-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11176381B2 (en) | Video object segmentation by reference-guided mask propagation | |
| CN110516541B (en) | Text positioning method and device, computer readable storage medium and computer equipment | |
| CN108229341B (en) | Classification method and apparatus, electronic equipment, computer storage medium | |
| EP4220555B1 (en) | Training method and apparatus for image segmentation model, image segmentation method and apparatus, and device | |
| CN116051593B (en) | Clothing image extraction method and device, equipment, medium and product thereof | |
| CN113435330A (en) | Micro-expression identification method, device, equipment and storage medium based on video | |
| CN111914654B (en) | Text layout analysis method, device, equipment and medium | |
| CN110675339A (en) | Image inpainting method and system based on edge inpainting and content inpainting | |
| CN113591719A (en) | Method and device for detecting text with any shape in natural scene and training method | |
| CN114037893A (en) | High-resolution remote sensing image building extraction method based on convolutional neural network | |
| CN116993987B (en) | A method and system for image semantic segmentation based on a lightweight neural network model | |
| US20250190760A1 (en) | Neural Architecture Search Method and Apparatus, and Storage Medium | |
| CN113298092B (en) | Neural network training method and device for extracting multi-level image contour information | |
| CN119625328B (en) | Continuous semantic segmentation method and system based on balanced multi-granularity fusion feature distillation | |
| CN114444565A (en) | Image tampering detection method, terminal device and storage medium | |
| CN115861605B (en) | An image data processing method, a computer device, and a readable storage medium. | |
| CN114820436A (en) | Tamper detection method, device, storage medium and electronic equipment | |
| CN120259785B (en) | Target recognition and model training method and device based on global perception graph convolution | |
| CN115861637A (en) | A salient object detection method based on dual-branch network | |
| CN116563304A (en) | Image processing method and device and training method and device of image processing model | |
| CN115115979A (en) | Recognition and replacement method of constituent elements in video and video recommendation method | |
| CN113298156B (en) | Neural network training method and device for classifying image sexes | |
| CN116798041B (en) | Image recognition method and device and electronic equipment | |
| CN119006837A (en) | Image editing trace recognition and model training method, computer storage medium and program product | |
| CN118674812A (en) | Image processing and model training method, device, equipment and 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |