CN109523558A - A kind of portrait dividing method and system - Google Patents
A kind of portrait dividing method and system Download PDFInfo
- Publication number
- CN109523558A CN109523558A CN201811204098.7A CN201811204098A CN109523558A CN 109523558 A CN109523558 A CN 109523558A CN 201811204098 A CN201811204098 A CN 201811204098A CN 109523558 A CN109523558 A CN 109523558A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- portrait
- lightweight
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
Abstract
The embodiment of the invention provides a kind of portrait dividing method and systems, this method comprises: based on the lightweight convolutional neural networks after training, portrait image is split, the lightweight convolutional neural networks are to add depthwise convolutional layer in convolutional neural networks to construct.Portrait dividing method provided in an embodiment of the present invention and system are reduced the parameter amount of convolutional neural networks, improve the efficiency for being carried out portrait segmentation based on convolutional neural networks by constructing and training lightweight convolutional neural networks.
Description
Technical field
The present embodiments relate to computer vision field more particularly to a kind of portrait dividing method and systems.
Background technique
Portrait segmentation is one of basic project of computer vision field, is all received widely in academia and industry
Pay attention to.With the rapid development of mobile Internet, in the application such as video conference and live streaming chat, traditional portrait dividing method
The prominent questions such as that there are processing speeds is slow, accuracy rate is low, algorithm complexity is high.
Recently as the development of deep learning, the portrait dividing method based on convolutional neural networks achieves great hair
Exhibition.In order to improve portrait segmentation accuracy rate, it will usually projected depth convolutional neural networks carry out portrait segmentation, can be partitioned into compared with
Good cut-off rule.However, depth convolutional neural networks while improving segmentation accuracy rate, but have ignored and want to segmentation rate
It asks, along with the compression processing of depth convolutional neural networks model is comparatively laborious and training pattern size is larger, so that processing
The efficiency of portrait segmentation is lower.
Therefore, in order to improve the efficiency of portrait segmentation, a kind of portrait dividing method is needed now and system is above-mentioned to solve
Problem.
Summary of the invention
The embodiment of the present invention is to solve to carry out portrait segmentation based on convolutional neural networks in the prior art that there are efficiency is lower
Defect, provide a kind of portrait dividing method and system.
In a first aspect, the embodiment of the invention provides a kind of portrait dividing methods, comprising: based on the lightweight volume after training
Product neural network, is split portrait image, the lightweight convolutional neural networks are added in convolutional neural networks
Depthwise convolutional layer construction.
Second aspect, the embodiment of the invention provides a kind of portrait segmenting systems, comprising: segmentation module, for based on instruction
Lightweight convolutional neural networks after white silk, are split portrait image, and the lightweight convolutional neural networks are in convolution mind
Through adding depthwise convolutional layer construction in network.
The third aspect the embodiment of the invention provides a kind of portrait splitting equipment, including memory, processor and is stored in
On memory and the computer program that can run on a processor, the processor realize such as first aspect when executing described program
The portrait dividing method.
Fourth aspect, the embodiment of the invention provides a kind of non-transient computer readable storage medium, the non-transient meter
Calculation machine readable storage medium storing program for executing stores computer instruction, and the computer instruction executes the computer as described in relation to the first aspect
Portrait dividing method.
Portrait dividing method provided in an embodiment of the present invention and system, by constructing and training lightweight convolutional Neural net
Network improves the efficiency that portrait segmentation is carried out based on convolutional neural networks to reduce the parameter amount of convolutional neural networks.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is the flow diagram of portrait dividing method provided in an embodiment of the present invention;
Fig. 2 is lightweight convolutional neural networks frame model schematic diagram provided in an embodiment of the present invention;
Fig. 3 is the structural schematic diagram of portrait segmenting system provided in an embodiment of the present invention;
Fig. 4 is the structural schematic diagram of portrait splitting equipment provided in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Portrait image described in the embodiment of the present invention refer to include face or character physical image, the people in image
Even if face or character physical it is imperfect or only include part, also can be used as the portrait image in the embodiment of the present invention.Portrait point
Cut refer to by portrait image portrait and background separate, be divided into different regions, and with different labels, lines or
Person's color distinguishes.
Fig. 1 is the flow diagram of portrait dividing method provided in an embodiment of the present invention, as shown in Figure 1, the present invention is implemented
Example provides a kind of portrait dividing method, comprising:
Step S1 is split portrait image based on the lightweight convolutional neural networks after training, the lightweight volume
Product neural network is to add depthwise convolutional layer in convolutional neural networks to construct.
In step sl, convolutional neural networks, that is, currently used are for scenes routine techniques such as image classification identifications, but
While being the limit of accuracy of the current convolutional neural networks model due to approximation computation machine visual task, depth and size
It is being doubled and redoubled, can only using under limited platform, can not be transplanted in mobile terminal and embedded chip at all.Even if logical
Network transmission is crossed, but higher bandwidth occupancy also allows many users daunting.On the other hand, large-sized model is also to equipment
Power consumption and the speed of service bring huge challenge.In view of the above-mentioned problems, the embodiment of the invention provides light after a kind of training
Magnitude convolutional neural networks particularly add depthwise convolutional layer construction, depthwise in convolutional neural networks
Convolutional layer is the framework that equally can be realized spatial convoluted under the precondition for keeping channel separation in neural network.Such as:
Depthwise convolutional layer in lightweight convolutional neural networks after training is connected on after convolutional layer, it is assumed that have one 3 ×
The convolutional layer of 3 sizes, input channel 16, output channel 32.Specifically, the convolution kernel of 32 3 × 3 sizes can traverse 16
Each data in a channel, to generate 16 × 32=512 characteristic spectrum, and then corresponding by being superimposed each input channel
Characteristic spectrum after fusion obtain 1 characteristic spectrum, 32 required output channels finally can be obtained.For this example application
Depthwise convolution traverses the data in 16 channels with the convolution kernel of 13 × 3 size, has obtained 16 characteristic spectrums.It is merging
Before operation, this 16 characteristic spectrums then are traversed with the convolution kernel of 32 1 × 1 sizes, carry out addition fusion.This process makes
With 16 × 3 × 3+16 × 32 × 1 × 1=656 parameter, far fewer than 16 × 32 × 3 × 3=4608 parameter above.Its
In above depth multiplier (depth multiplier) be set as 1.
It is understood that lightweight convolutional neural networks provided in an embodiment of the present invention are compared with convolutional neural networks,
Maximum difference is embodied in the parameter amount of model, even if the depth of whole network deepens, parameter amount can also be reduced, and make the meter of model
Calculation amount is reduced simultaneously.
Such as: the convolution kernel size of regular volume lamination is the conv layer (group=1) of k × k, and depthwise convolutional layer
In a convolution kernel size be k × k depthwise convolutional layer (being in the nature conv layers) (group=input channel)
In addition the separable convolutional layer (being in the nature conv layers) (group=1) that a convolution kernel size is 1 × 1.Due to 95%
Calculation amount all concentrates in 1 × 1 separable convolutional layer, therefore can directly be carried out using matrix operation acceleration model fast
Speed calculates, to improve computational efficiency.
It, can also be with it is understood that lightweight convolutional neural networks provided in an embodiment of the present invention needs are trained
Referred to as learn.The present invention is not especially limit this for the sample set and training mode of specific training study, any energy
The training method for enough completing portrait segmentation may be applicable to the embodiment of the present invention.
By adding depthwise convolutional layer in convolutional neural networks to constitute lightweight convolutional neural networks, reduce
The parameter amount of model, to reduce the complexity of convolutional neural networks, while also accelerating the detection speed of model, mentions
The high efficiency that portrait segmentation is carried out based on convolutional neural networks.
On the basis of the above embodiments, the lightweight convolutional neural networks based on after training, to portrait image into
Row segmentation, comprising:
Lightweight convolutional neural networks after the training are converted into Caffe model;
It is Coreml model by the Caffe model conversion;
The portrait image is input in the Coreml model, to be split to the portrait image.
Traditional portrait dividing method is all often first by completing portrait segmentation, image in the fixed platforms such as computer
Acquisition, later period are split again, can not carry out real-time portrait segmentation to the portrait image of acquisition.And in practice, people
To the real-time image obtained or shoot, if necessary to carry out portrait segmentation, then after needing by inputting an image into computer
Just can be carried out so that the inconvenience that operates, along with conventional convolution neural network often relate to it is not only wide but also deep, it is difficult to
Mobile terminal deployment uses.
In view of the above-mentioned problems, the embodiment of the present invention provides a kind of by way of model conversion, the deployment instruction on mobile terminal
Lightweight convolutional neural networks after white silk carry out real-time portrait segmentation to realize on mobile terminal.
Specifically, real-time portrait segmentation is carried out in mobile terminal in order to realize, by the lightweight convolutional neural networks after training
Model is converted to transition model Caffe model first, then by Caffe model conversion is Coreml model, in the converted of model
Cheng Zhong keeps the characteristic value of model constant, prevents loss of data, to ensure that the model accuracy after converting remains unchanged.It needs
Bright, Coreml model is used to be deployed in the mobile terminal of IOS system.
By establishing lightweight convolutional neural networks, so that the model parameter amount of convolutional neural networks is reduced, in guarantor
While as segmentation accuracy rate, to improve portrait segmentation rate, then pass through the lightweight convolutional neural networks after training
A series of model conversion is carried out, the convolutional neural networks model that can operate with mobile terminal is finally obtained, realizes in mobile terminal
Carry out real-time portrait segmentation.
On the basis of the above embodiments, the lightweight convolutional neural networks after described based on training, to facial image
Before being split, further includes:
Training sample set is pre-processed;
Based on pretreated training sample set, the lightweight convolutional neural networks are trained.
In embodiments of the present invention, it is training lightweight convolutional neural networks, needs to acquire the training sample with mark portrait
This, therefore to obtain training sample set, the embodiment of the present invention uses from open source portrait segmentation data and internet and crawls portrait mark
The mode that note data combine is acquired.Wherein, open source portrait segmentation data pass through in MSCOCO, Supervisely, ATR
It is acquired with EG1800;And from the portrait labeled data that internet acquires by being acquired in webcast website or Flickr, and carry out
Plain portrait mark item by item;In embodiments of the present invention, by acquiring in MSCOCO, Supervisely, ATR and EG1800
Portrait image about 25000, the portrait image acquired from internet about 12000.Finally by collected portrait image
Format specification is unified, training sample set is formed, so that lightweight convolutional neural networks are trained.
The embodiment of the present invention selects the sample conduct for having marked portrait image by multi-platform acquisition portrait image
Training data avoids the miscellaneous work of artificial screening portrait image, further increases the accuracy rate of model after training.
It is on the basis of the above embodiments, described that training sample set is pre-processed, comprising:
The image pattern without portrait is concentrated to be labeled as negative sample collection the training sample, and by the training sample set
In the image pattern containing portrait be labeled as positive sample collection.
In embodiments of the present invention, random acquisition 5000 opens the image not comprising portrait as negative sample from MSCOCO,
20000 images comprising portrait are acquired from Supervisely, ATR and EG1800 as positive sample, by the training after classification
Sample set is input to lightweight convolution planned network and is trained, the lightweight convolutional neural networks after thus obtained training.
In addition, 12000 portrait images acquired from internet do not divide positive negative sample, it is secondary as lightweight convolutional neural networks
Trained training sample set.
By carrying out positive and negative sample set classification to training sample set, so that the mould for concentrating training to obtain in this sample data
Type generalization improves, and can detect to a greater variety of images, meanwhile, when model concentrates training to finish in this training sample
Afterwards, picture of the transportable study to other refinement fields, the more conducively convergence of model.
On the basis of the above embodiments, described to be based on pretreated training sample set, to the lightweight convolution mind
It is trained through network, comprising:
Based on Pytorch frame and pretreated training sample set, the lightweight convolutional neural networks are carried out
Training, obtains Pytorch model;
Using the Pytorch model as the lightweight convolutional neural networks after the training.
Fig. 2 is lightweight convolutional neural networks frame model schematic diagram provided in an embodiment of the present invention, as shown in Fig. 2, logical
It crosses Pytorch frame and establishes convolutional neural networks model, in which:
Conv: common convolutional layer is indicated;
Conv_dw: depthwise convolutional layer, i.e. convolutional layer in mobilenet-v1 are indicated;
Convolutional layer with P:2 indicates to carry out 2 times of down-sampling by stride=2;
Deconv: indicating up-sampling layer ConvTranspose2d, and the learning rate of this part is 0 when hands-on, using double
Linear difference initialization is identical with deconv layers of outputization method in FCN;
Residual: it indicates to improve the residual block of edge definition.
In embodiments of the present invention, training image size is 224 × 224, and output image size is also 224 × 224, is being surveyed
When examination, the size of test image be can change, and is at present 224 × 176, is if desired handled test image size, pass through
Resize function zooms in and out original image, if long or wide insufficient, is filled with 128 pixel values, finally by up-sampling layer
The mask size of output is handled, keeps mask size and original image in the same size.
In current Open Framework, Pytorch frame is owned by terms of these three higher in flexibility, ease for use and speed
Level.Since the flexibility of PyTorch frame is not using speed as cost, the training speed performance of PyTorch frame surpasses it
His frame, so that obtaining the lightweight convolutional neural networks after training in a relatively short period of time.
On the basis of the above embodiments, described to be based on pretreated training sample set, to the lightweight convolution mind
It is trained through network, further includes: the pretreated training sample set is input to the lightweight convolutional neural networks
In, and Fusion Features layer is added in the lightweight convolutional neural networks.
In embodiments of the present invention, it after training sample being input to lightweight convolutional neural networks, can refer to shown in Fig. 2,
Add the lightweight convolutional neural networks of 1 Fusion Features layer.Wherein, Fusion Features layer has merged in depthwise convolutional layer
Low layer, middle layer and high level characteristic information, low layer channel number be 128, middle layer channel number be 256, high-rise channel
Channel number is 128+256+512=896 after number is 512, concate, can connect one conv_dw layers later channel
Number is reduced to 128.
The present embodiment improves the essence of portrait segmentation by adding Fusion Features layer in lightweight convolutional neural networks
Degree, while balancing the accuracy rate and efficiency of model inspection.
On the basis of the above embodiments, the lightweight convolutional neural networks after described based on training, to portrait image
Before being split, the method also includes:
Multiple depthwise convolutional layers, multiple up-sampling layers and multiple down-samplings are added in the convolutional neural networks
Layer, to obtain the lightweight convolutional neural networks.
In embodiments of the present invention, 10 depthwise convolutional layers, 4 down-sampling layers are added in convolutional neural networks
With 4 up-sampling layers, can refer to shown in Fig. 2, depthwise convolutional layer be sequentially connected in series with after convolutional layer, and to image carry out
Convolution reduces the parameter amount of image, and channel amount increases, and is reduced the display area of image by down-sampling layer, makes image
Scale further reduces, and reduces the time required to training.When portrait image is input to trained lightweight convolutional Neural net
After being handled in network, makes the mask size of output in the same size with original image by up-sampling layer, finally obtain portrait point
The image cut.
By adding multiple depthwise convolutional layers in convolutional neural networks, reduces the parameter amount of model, accelerate
The detection speed of model, and further reduced by multiple down-sampling layers, the training required time is greatly lowered, as input people
After the model after image to training, portrait segmented image is reduced into former portrait image size by multiple up-sampling layers, is protected
The precision of portrait segmented image is demonstrate,proved.
Fig. 3 is the structural schematic diagram of portrait segmenting system provided in an embodiment of the present invention, as shown in figure 3, the present invention is implemented
Example provides a kind of portrait segmenting system, including segmentation module 31, wherein segmentation module 31 is used for based on the lightweight after training
Convolutional neural networks are split portrait image, and the lightweight convolutional neural networks are added in convolutional neural networks
Depthwise convolutional layer construction.
Portrait segmenting system provided in an embodiment of the present invention be for executing above-mentioned each method embodiment, detailed process and
Detailed content please refers to above-described embodiment, and details are not described herein again.
Fig. 4 is the structural schematic diagram of portrait splitting equipment provided in an embodiment of the present invention, as shown in figure 4, the portrait is divided
Equipment may include: processor (processor) 41,42, memory communication interface (Communications Interface)
(memory) 43 and communication bus 44, wherein processor 41, communication interface 42, memory 43 complete phase by communication bus 44
Communication between mutually.Communication interface 42 can be used for the letter between the lightweight convolutional neural networks after portrait splitting equipment and training
Breath transmission.Processor 41 can call the logical order in memory 43, to execute following method: based on the lightweight after training
Convolutional neural networks are split portrait image, and the lightweight convolutional neural networks are added in convolutional neural networks
Depthwise convolutional layer construction.
In addition, the logical order in above-mentioned memory 43 can be realized and as only by way of SFU software functional unit
Vertical product when selling or using, can store in a computer readable storage medium.Based on this understanding, this hair
Substantially the part of the part that contributes to existing technology or the technical solution can be with soft in other words for bright technical solution
The form of part product embodies, which is stored in a storage medium, including some instructions are to make
It obtains a computer equipment (can be personal computer, server or the network equipment etc.) and executes each embodiment of the present invention
The all or part of the steps of the method.And storage medium above-mentioned include: USB flash disk, mobile hard disk, read-only memory (ROM,
Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. it is various
It can store the medium of program code.
The embodiment of the present invention provides a kind of non-transient computer readable storage medium, the non-transient computer readable storage medium
The instruction of matter storage server, the computer instruction make computer execute portrait dividing method provided by above-described embodiment, such as
Include: based on training after lightweight convolutional neural networks, portrait image is split, the lightweight convolutional neural networks
It is to add depthwise convolutional layer in convolutional neural networks to construct.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features;
And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (10)
1. a kind of portrait dividing method characterized by comprising
Based on the lightweight convolutional neural networks after training, portrait image is split, the lightweight convolutional neural networks
It is to add depthwise convolutional layer in convolutional neural networks to construct.
2. the method according to claim 1, wherein the lightweight convolutional neural networks based on after training,
Portrait image is split, comprising:
Lightweight convolutional neural networks after the training are converted into Caffe model;
It is Coreml model by the Caffe model conversion;
The portrait image is input in the Coreml model, to be split to the portrait image.
3. the method according to claim 1, wherein it is described based on training after lightweight convolutional Neural net
Network, before being split to facial image, further includes:
Training sample set is pre-processed;
Based on pretreated training sample set, the lightweight convolutional neural networks are trained.
4. according to the method described in claim 3, it is characterized in that, described pre-process training sample set, comprising:
It concentrates the image pattern without portrait to be labeled as negative sample collection the training sample, and training sample concentration is contained
There is the image pattern of portrait labeled as positive sample collection.
5. according to the method described in claim 3, it is characterized in that, described be based on pretreated training sample set, to described
Lightweight convolutional neural networks are trained, comprising:
Based on Pytorch frame and pretreated training sample set, the lightweight convolutional neural networks are trained,
Obtain Pytorch model;
Using the Pytorch model as the lightweight convolutional neural networks after the training.
6. according to the method described in claim 3, it is characterized in that, described be based on pretreated training sample set, to described
Lightweight convolutional neural networks are trained, further includes:
The pretreated training sample set is input in the lightweight convolutional neural networks, and is rolled up in the lightweight
Fusion Features layer is added in product neural network.
7. method according to any one of claims 1 to 6, which is characterized in that it is described based on training after lightweight volume
Product neural network, before being split to portrait image, the method also includes:
Multiple depthwise convolutional layers, multiple up-sampling layers and multiple down-sampling layers are added in the convolutional neural networks, with
Obtain the lightweight convolutional neural networks.
8. a kind of portrait segmenting system characterized by comprising
Divide module, for being split to portrait image based on the lightweight convolutional neural networks after training, the lightweight
Convolutional neural networks are to add depthwise convolutional layer in convolutional neural networks to construct.
9. a kind of portrait splitting equipment, can run on a memory and on a processor including memory, processor and storage
Computer program, which is characterized in that the processor realizes the people as described in any one of claim 1 to 7 when executing described program
As dividing method.
10. a kind of non-transient computer readable storage medium, which is characterized in that the non-transient computer readable storage medium is deposited
Computer instruction is stored up, the computer instruction makes the computer execute the portrait segmentation side as described in any one of claim 1 to 7
Method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811204098.7A CN109523558A (en) | 2018-10-16 | 2018-10-16 | A kind of portrait dividing method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811204098.7A CN109523558A (en) | 2018-10-16 | 2018-10-16 | A kind of portrait dividing method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109523558A true CN109523558A (en) | 2019-03-26 |
Family
ID=65770886
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811204098.7A Pending CN109523558A (en) | 2018-10-16 | 2018-10-16 | A kind of portrait dividing method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109523558A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111091130A (en) * | 2019-12-13 | 2020-05-01 | 南京邮电大学 | Real-time image semantic segmentation method and system based on lightweight convolutional neural network |
CN111401247A (en) * | 2020-03-17 | 2020-07-10 | 杭州趣维科技有限公司 | Portrait segmentation method based on cascade convolution neural network |
CN112380895A (en) * | 2020-09-30 | 2021-02-19 | 深圳点猫科技有限公司 | Portrait segmentation method, device and equipment based on deep learning |
CN114115296A (en) * | 2022-01-21 | 2022-03-01 | 南京理工大学 | Intelligent inspection and early warning system and method for key area |
WO2022116163A1 (en) * | 2020-12-04 | 2022-06-09 | 深圳市优必选科技股份有限公司 | Portrait segmentation method, robot, and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107886133A (en) * | 2017-11-29 | 2018-04-06 | 南京市测绘勘察研究院股份有限公司 | A kind of underground piping defect inspection method based on deep learning |
CN108010031A (en) * | 2017-12-15 | 2018-05-08 | 厦门美图之家科技有限公司 | A kind of portrait dividing method and mobile terminal |
CN108009598A (en) * | 2017-12-27 | 2018-05-08 | 北京诸葛找房信息技术有限公司 | Floor plan recognition methods based on deep learning |
US20180137406A1 (en) * | 2016-11-15 | 2018-05-17 | Google Inc. | Efficient Convolutional Neural Networks and Techniques to Reduce Associated Computational Costs |
-
2018
- 2018-10-16 CN CN201811204098.7A patent/CN109523558A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180137406A1 (en) * | 2016-11-15 | 2018-05-17 | Google Inc. | Efficient Convolutional Neural Networks and Techniques to Reduce Associated Computational Costs |
CN107886133A (en) * | 2017-11-29 | 2018-04-06 | 南京市测绘勘察研究院股份有限公司 | A kind of underground piping defect inspection method based on deep learning |
CN108010031A (en) * | 2017-12-15 | 2018-05-08 | 厦门美图之家科技有限公司 | A kind of portrait dividing method and mobile terminal |
CN108009598A (en) * | 2017-12-27 | 2018-05-08 | 北京诸葛找房信息技术有限公司 | Floor plan recognition methods based on deep learning |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111091130A (en) * | 2019-12-13 | 2020-05-01 | 南京邮电大学 | Real-time image semantic segmentation method and system based on lightweight convolutional neural network |
CN111401247A (en) * | 2020-03-17 | 2020-07-10 | 杭州趣维科技有限公司 | Portrait segmentation method based on cascade convolution neural network |
CN111401247B (en) * | 2020-03-17 | 2023-07-28 | 杭州小影创新科技股份有限公司 | Portrait segmentation method based on cascade convolution neural network |
CN112380895A (en) * | 2020-09-30 | 2021-02-19 | 深圳点猫科技有限公司 | Portrait segmentation method, device and equipment based on deep learning |
WO2022116163A1 (en) * | 2020-12-04 | 2022-06-09 | 深圳市优必选科技股份有限公司 | Portrait segmentation method, robot, and storage medium |
CN114115296A (en) * | 2022-01-21 | 2022-03-01 | 南京理工大学 | Intelligent inspection and early warning system and method for key area |
CN114115296B (en) * | 2022-01-21 | 2022-05-20 | 南京理工大学 | Intelligent inspection and early warning system and method for key area |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10943145B2 (en) | Image processing methods and apparatus, and electronic devices | |
CN109508681B (en) | Method and device for generating human body key point detection model | |
CN109523558A (en) | A kind of portrait dividing method and system | |
US8692830B2 (en) | Automatic avatar creation | |
CN110379020B (en) | Laser point cloud coloring method and device based on generation countermeasure network | |
CN106372648A (en) | Multi-feature-fusion-convolutional-neural-network-based plankton image classification method | |
CN112651978A (en) | Sublingual microcirculation image segmentation method and device, electronic equipment and storage medium | |
CN110827312B (en) | Learning method based on cooperative visual attention neural network | |
CN111597920B (en) | Full convolution single-stage human body example segmentation method in natural scene | |
CN109472193A (en) | Method for detecting human face and device | |
CN111985281B (en) | Image generation model generation method and device and image generation method and device | |
CN115457531A (en) | Method and device for recognizing text | |
CN111476133B (en) | Unmanned driving-oriented foreground and background codec network target extraction method | |
CN110349167A (en) | A kind of image instance dividing method and device | |
CN113704531A (en) | Image processing method, image processing device, electronic equipment and computer readable storage medium | |
CN110532959B (en) | Real-time violent behavior detection system based on two-channel three-dimensional convolutional neural network | |
CN109636764A (en) | A kind of image style transfer method based on deep learning and conspicuousness detection | |
CN113269133A (en) | Unmanned aerial vehicle visual angle video semantic segmentation method based on deep learning | |
CN112668638A (en) | Image aesthetic quality evaluation and semantic recognition combined classification method and system | |
JP2023131117A (en) | Joint perception model training, joint perception method, device, and medium | |
CN116580184A (en) | YOLOv 7-based lightweight model | |
CN112668675B (en) | Image processing method and device, computer equipment and storage medium | |
CN113297956B (en) | Gesture recognition method and system based on vision | |
CN113066074A (en) | Visual saliency prediction method based on binocular parallax offset fusion | |
CN116342877A (en) | Semantic segmentation method based on improved ASPP and fusion module in complex scene |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190326 |
|
RJ01 | Rejection of invention patent application after publication |