CN106372630A - Face direction detection method based on deep learning - Google Patents

Face direction detection method based on deep learning Download PDF

Info

Publication number
CN106372630A
CN106372630A CN201611037721.5A CN201611037721A CN106372630A CN 106372630 A CN106372630 A CN 106372630A CN 201611037721 A CN201611037721 A CN 201611037721A CN 106372630 A CN106372630 A CN 106372630A
Authority
CN
China
Prior art keywords
face
picture
deep learning
direction detection
face direction
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
Application number
CN201611037721.5A
Other languages
Chinese (zh)
Inventor
金连文
李梦茹
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201611037721.5A priority Critical patent/CN106372630A/en
Publication of CN106372630A publication Critical patent/CN106372630A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/19—Recognition using electronic means
    • G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
    • G06V30/194—References adjustable by an adaptive method, e.g. learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Neurology (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Databases & Information Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Human Computer Interaction (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a face direction detection method based on deep learning, comprising the following steps: building a face direction detection database; building a convolution neural network; inputting a normalized face image used for deep learning to the convolution neural network for training to get a face direction detection model; and inputting an image needing face direction detection to the face direction detection model to get the direction of a face in the image. Compared with the prior art, the traditional manual face image extraction method is abandoned, and the image features are extracted automatically using the convolution neural network in deep learning; feature extraction and face direction detection are integrated, overall optimization is facilitated, and end-to-end face direction detection is truly realized; and the detection model can be further optimized by use of face big data, and the model can still be highly robust even under interference of light and background complexity due to increase in face data.

Description

A kind of face direction detection method based on deep learning
Technical field
The present invention relates to computer picture data processing and mode identification technology, more particularly, to one kind are based on depth The face direction detection method practised.
Background technology
With the development of computer picture Processing Technique, its utilization in the life of people also increasingly extensive.In recent years Come practical more and more extensive, such as Face datection, recognition of face, face U.S. of the technology about face aspect in computer vision The research of the problems such as change is also increasingly deeply and ripe.With the extensive application of the product of face aspect, some practical problems also with , such as in the problems such as recognition of face, Face datection, its research premise is all built upon the human face data of correct direction On basis, therefore its application system is all premised on the face picture inputting correct direction.And in real use, when (or when digital equipment does not have the function of providing picture directional information), people when the directional information of the picture in digital equipment is lost The correct direction of face picture just has no way of finding out about it.These pictures without directional information are input in Face datection or identifying system When, if picture anisotropy, system can do the judgement making mistake, and obtains the result of mistake.Therefore, examine in recognition of face, face In the systems such as survey, embedded face angle detecting module is extremely necessary, it can improve detection accuracy and the user of system Experience sense.
But solve the problems, such as that face angle detecting great majority are also merely resting on the feature of manual extraction picture at present, Then on the basis being learnt with the algorithm of traditional machine learning.In this traditional face direction detection method, no Only feature extraction is relatively complicated, and the extraction of feature depends on experience and the knowledge of researcher.Meanwhile, traditional face direction inspection Survey method speed in detection is slower, thus affecting the speed of whole application system, Consumer's Experience greatly reduces.Depth The method practised can abandon the work of loaded down with trivial details manual feature extraction, sign is extracted and direction prediction is integrated.Using depth Spend self study to extract the more structural feature that can represent face direction with level of image, it is possible to achieve to face side To more accurately predicting.
Content of the invention
For overcoming the deficiencies in the prior art, provide a kind of method of the face angle detecting based on deep learning.By depth Convolutional neural networks in degree study set up face angle detecting model, in conjunction with the training side of substantial amounts of training sample and stratification Method, so that model can learn to the feature that can represent face direction, predicts correct people to the face picture of any input Face direction, thus realizing face angle detecting end to end, the present invention proposes a kind of face angle detecting based on deep learning Method.
The technical scheme is that and be achieved in that:
A kind of face direction detection method based on deep learning, including step
S1: build for face angle detecting data base, pretreatment is carried out to the face picture for deep learning, then The described face picture being used for deep learning is carried out size normalization；
S2: build convolutional neural networks, described convolutional neural networks comprise five convolutional layers and three full articulamentums, described Five convolutional layers comprise 20,50,100,150,200 characteristic patterns successively respectively, and the convolution kernel size of five convolutional layers is successively Be respectively the convolution step-length of 3*3,2*2,2*2,2*2,2*2, five convolutional layer be all 1, be designed with after each convolutional layer under one Sample level, the core of each down-sampling layer is 2*2 size, and step-length is 2, and described five down-sampling layers all adopt average sampling side Method；First and second full articulamentum in described three full articulamentums comprises 1024 neurons, the 3rd full articulamentum Comprise 4 neurons；
S3: the face picture being used for deep learning described in after normalization is input in described convolutional neural networks and carries out Training, obtains face angle detecting model；
S4: the picture carrying out face angle detecting will be needed to be input to described face angle detecting model, obtain in picture Face direction.
Further, in described step s1, face angle detecting data base is in celebfaces attributes data The face angle detecting data base building on storehouse.
Further, described in step s1 be used for deep learning face picture in face direction be divided into front, to Right upset, downwardly turn over and to the left overturn four direction.
Further, obtaining the face direction in picture in step s4 is 0.、90.、180.Or 270..
Further, the face picture being used for deep learning in step s1 is normalized to the picture of 158*158 pixel.
Further, step s2 includes step
S21: the face picture being used for deep learning is carried out face inspection in celebfaces attributes data base Survey, detect foursquare human face region；
S22: outwards intercept the area of described one times of size of human face region centered on human face region, beyond human face region Place is with white filling；
S23: the picture being truncated to is adjusted to 156*156 pixel；
S24: by the picture rotation being truncated to be face, overturn to the right, downwardly turn over or to the left overturn four direction picture.
Especially further, in the training process of step s3, the picture for being trained can be become 138* by random cropping 138 sizes.
The beneficial effects of the present invention is, compared with prior art, the manual extraction face picture of abandoning tradition of the present invention Method, automatically extract picture feature using the convolutional neural networks in deep learning；By feature extraction and face angle detecting It is integrated, be conducive to global optimization it is achieved that real face angle detecting end to end；Combination with face big data makes Detection model can further be optimized, and the increase of human face data can make model disturb in illumination, background complexity etc. Under still there is stronger robustness.
Brief description
Fig. 1 is a kind of face direction detection method flow chart based on deep learning of the present invention.
Fig. 2 is the convolutional neural networks structure chart in the present invention.
Specific embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Site preparation description is it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.It is based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of not making creative work Embodiment, broadly falls into the scope of protection of the invention.
Refer to Fig. 1, a kind of face direction detection method based on deep learning of the present invention, including step
S1: build for face angle detecting data base, pretreatment is carried out to the face picture for deep learning, then The described face picture being used for deep learning is carried out size normalization；
S2: build convolutional neural networks, described convolutional neural networks comprise five convolutional layers and three full articulamentums, described Five convolutional layers comprise 20,50,100,150,200 characteristic patterns successively respectively, and the convolution kernel size of five convolutional layers is successively Be respectively the convolution step-length of 3*3,2*2,2*2,2*2,2*2, five convolutional layer be all 1, be designed with after each convolutional layer under one Sample level, the core of each down-sampling layer is 2*2 size, and step-length is 2, and described five down-sampling layers all adopt average sampling side Method；First and second full articulamentum in described three full articulamentums comprises 1024 neurons, the 3rd full articulamentum Comprise 4 neurons；
S3: the face picture being used for deep learning described in after normalization is input in described convolutional neural networks and carries out Training, obtains face angle detecting model；
S4: the picture carrying out face angle detecting will be needed to be input to described face angle detecting model, obtain in picture Face direction.
Technical key point is as follows:
The structure in face bearing data storehouse:
In the present invention, a crucial technology point is the structure in face bearing data storehouse, and face bearing data storehouse is to be based on Build in celebfaces Basis of Database.Celebfaces attributes data base comprises 200,000 containing face Picture.When building face angle detecting data base, first to carrying out face in celebfaces attributes data base Detection.Then on the basis of the square human face region detecting 1 times of outward expansion (if having beyond boundary member, with white Filling) and intercepted.Then the size reduction of the face picture being truncated to (or amplification) is 156*156 pixel；Then will Face picture rotates to be the picture of four direction respectively；
Convolutional neural networks:
The structural model of convolutional neural networks characterizes to model learning face direction has important impact.Through repeatedly real Test, the present invention targetedly proposes one and has larger picture input size, the volume of less convolution kernel size and deeper structure Long-pending neutral net, as shown in Figure 2, its detail parameters is as described below for its brief configuration:
The image input size of network is 156*156, can be become 138*138 size by random cropping in the training process.Net Network comprises 5 convolutional layers (convolutional layer) and 3 full articulamentums (fully-connected layer), each Convolutional layer comprises 20,50,100,150,200 characteristic patterns (feature mao) respectively；The convolution kernel of each convolutional layer (convolution kernel) size is 3*3 respectively, 2*2,2*2,2*2,2*2, and convolution step-length (stride) is all 1；Each A down-sampling layer (pooling layer) is closelyed follow, under each after followed by one each convolutional layer of down-sampling layer after convolutional layer The core of sample level is all 2x2 size, and step-length is that 2,5 down-sampling layers adopt the average method of sampling (average pooling)；Net The full articulamentum of network first and second comprises 1024 neurons, and the 3rd full articulamentum comprises 4 neurons, that is, export net The face direction of network prediction；
The input of prototype network be any face picture be rgb pixel value, carry without carrying out extra feature again Take it is achieved that face angle detecting end to end；Using face picture direction (ground truth) and network output fraction it Between softmax normalization probability as loss function.Defining predictive value x is the corresponding value predicting each class, lnFor one The integer value of label, its scope is ln∈ [0 ..., n-1], then the loss function of network be:
lo s s = - 1 n σ n = 1 n l o g ( p ^ n , l n )
WhereinProbability for the class of softmax output:
p ^ n , l n = e x n σ j = 0 n e x j
Network training process:
100,000 pictures that random choose from face bearing data storehouse goes out, as training data, training data are inputted It is trained in specific depth convolutional neural networks.(removing training data) random choose again from face bearing data storehouse 20000 pictures as test data, for the test of network model；Depth convolutional neural networks pass through iterate, train, surveying Examination, when accuracy rate is surely 99.1%, deconditioning, obtain final face angle detecting model.
The present invention can operate with following field:
(1) Face datection direction, can make face detection system input the picture in correct face direction, improve Face datection Accuracy；
(2) face beauty direction, can make face beauty system input the picture in correct face direction, so that face beauty System provides objective face beauty fraction；
(3) picture display system, makes system in correct display face picture.
The above is the preferred embodiment of the present invention it is noted that for those skilled in the art For, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications are also considered as Protection scope of the present invention.

Claims (7)

1. a kind of face direction detection method based on deep learning is it is characterised in that include step
S1: build for face angle detecting data base, pretreatment is carried out to the face picture for deep learning, then by institute State the face picture for deep learning and carry out size normalization；
S2: build convolutional neural networks, described convolutional neural networks comprise five convolutional layers and three full articulamentums, described five Convolutional layer comprises 20,50,100,150,200 characteristic patterns successively respectively, and the convolution kernel size of five convolutional layers is distinguished successively Be the convolution step-length of 3*3,2*2,2*2,2*2,2*2, five convolutional layer be all 1, be designed with a down-sampling after each convolutional layer Layer, the core of each down-sampling layer is 2*2 size, and step-length is 2, and described five down-sampling layers all adopt the average method of sampling；Institute State first and second full articulamentum in three full articulamentums and comprise 1024 neurons, the 3rd full articulamentum comprises 4 Individual neuron；
S3: the face picture being used for deep learning described in after normalization is input in described convolutional neural networks and is instructed Practice, obtain face angle detecting model；
S4: the picture carrying out face angle detecting will be needed to be input to described face angle detecting model, obtain the people in picture Face direction.
2. the face direction detection method based on deep learning as claimed in claim 1 is it is characterised in that in described step s1 Face angle detecting data base is the face angle detecting data base building on celebfaces attributes data base.
3. the face direction detection method based on deep learning as claimed in claim 1 is it is characterised in that described in step s1 Being divided into front, overturn to the right, downwardly turn over and upset four to the left for the face direction in the face picture of deep learning Direction.
4. the face direction detection method based on deep learning as claimed in claim 3 is it is characterised in that obtain in step s4 Face direction in picture is 0 °, 90 °, 180 ° or 270 °.
5. the face direction detection method based on deep learning as claimed in claim 1 is it is characterised in that be used in step s1 The face picture of deep learning is normalized to the picture of 158*158 pixel.
6. the face direction detection method based on deep learning as claimed in claim 5 is it is characterised in that step s2 includes walking Suddenly
S21: the face picture being used for deep learning is carried out Face datection in celebfaces attributes data base, Detect foursquare human face region；
S22: outwards intercept the area of described one times of size of human face region centered on human face region, beyond the place of human face region With white filling；
S23: the picture being truncated to is adjusted to 156*156 pixel；
S24: by the picture rotation being truncated to be face, overturn to the right, downwardly turn over or to the left overturn four direction picture.
7. the face direction detection method based on deep learning as claimed in claim 6 is it is characterised in that instruction in step s3 During white silk, the picture for being trained can be become 138*138 size by random cropping.
CN201611037721.5A 2016-11-23 2016-11-23 Face direction detection method based on deep learning Pending CN106372630A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611037721.5A CN106372630A (en) 2016-11-23 2016-11-23 Face direction detection method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611037721.5A CN106372630A (en) 2016-11-23 2016-11-23 Face direction detection method based on deep learning

Publications (1)

Publication Number Publication Date
CN106372630A true CN106372630A (en) 2017-02-01

Family

ID=57891119

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611037721.5A Pending CN106372630A (en) 2016-11-23 2016-11-23 Face direction detection method based on deep learning

Country Status (1)

Country Link
CN (1) CN106372630A (en)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106909909A (en) * 2017-03-08 2017-06-30 王华锋 A kind of Face datection and alignment schemes based on shared convolution feature
CN107301406A (en) * 2017-07-13 2017-10-27 珠海多智科技有限公司 Fast face angle recognition method based on deep learning
CN107545238A (en) * 2017-07-03 2018-01-05 西安邮电大学 Underground coal mine pedestrian detection method based on deep learning
CN108428247A (en) * 2018-02-27 2018-08-21 广州视源电子科技股份有限公司 Method and system for detecting direction of soldering tin point
CN108629291A (en) * 2018-04-13 2018-10-09 深圳市未来媒体技术研究院 A kind of face depth prediction approach of anti-grid effect
CN109344855A (en) * 2018-08-10 2019-02-15 华南理工大学 A kind of face beauty assessment method of the depth model returned based on sequence guidance
CN109740492A (en) * 2018-12-27 2019-05-10 郑州云海信息技术有限公司 A kind of identity identifying method and device
CN109902680A (en) * 2019-03-04 2019-06-18 四川长虹电器股份有限公司 The detection of picture rotation angle and bearing calibration based on convolutional neural networks

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101917548A (en) * 2010-08-11 2010-12-15 无锡中星微电子有限公司 Image pickup device and method for adaptively adjusting picture
CN104992167A (en) * 2015-07-28 2015-10-21 中国科学院自动化研究所 Convolution neural network based face detection method and apparatus
CN105701468A (en) * 2016-01-12 2016-06-22 华南理工大学 Face attractiveness evaluation method based on deep learning

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101917548A (en) * 2010-08-11 2010-12-15 无锡中星微电子有限公司 Image pickup device and method for adaptively adjusting picture
CN104992167A (en) * 2015-07-28 2015-10-21 中国科学院自动化研究所 Convolution neural network based face detection method and apparatus
CN105701468A (en) * 2016-01-12 2016-06-22 华南理工大学 Face attractiveness evaluation method based on deep learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
PHILIPP FISCHER 等: "Image Orientation Estimation with Convolutional Networks", 《PATTERN RECOGNITION》 *
SHEIKH FAISAL RASHID 等: "A Discriminative Learning Approach for Orientation Detection of Urdu Document Images", 《2009 IEEE 13TH INTERNATIONAL MULTITOPIC CONFERENCE 》 *

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106909909A (en) * 2017-03-08 2017-06-30 王华锋 A kind of Face datection and alignment schemes based on shared convolution feature
CN106909909B (en) * 2017-03-08 2021-02-02 王华锋 Face detection and alignment method based on shared convolution characteristics
CN107545238A (en) * 2017-07-03 2018-01-05 西安邮电大学 Underground coal mine pedestrian detection method based on deep learning
CN107301406A (en) * 2017-07-13 2017-10-27 珠海多智科技有限公司 Fast face angle recognition method based on deep learning
CN108428247A (en) * 2018-02-27 2018-08-21 广州视源电子科技股份有限公司 Method and system for detecting direction of soldering tin point
CN108629291A (en) * 2018-04-13 2018-10-09 深圳市未来媒体技术研究院 A kind of face depth prediction approach of anti-grid effect
CN108629291B (en) * 2018-04-13 2020-10-20 深圳市未来媒体技术研究院 Anti-grid effect human face depth prediction method
CN109344855A (en) * 2018-08-10 2019-02-15 华南理工大学 A kind of face beauty assessment method of the depth model returned based on sequence guidance
CN109344855B (en) * 2018-08-10 2021-09-24 华南理工大学 Depth model face beauty evaluation method based on sequencing guided regression
CN109740492A (en) * 2018-12-27 2019-05-10 郑州云海信息技术有限公司 A kind of identity identifying method and device
CN109902680A (en) * 2019-03-04 2019-06-18 四川长虹电器股份有限公司 The detection of picture rotation angle and bearing calibration based on convolutional neural networks

Similar Documents

Publication Publication Date Title
CN106372630A (en) Face direction detection method based on deep learning
CN104978580B (en) A kind of insulator recognition methods for unmanned plane inspection transmission line of electricity
CN111723732B (en) Optical remote sensing image change detection method, storage medium and computing equipment
CN105975931A (en) Convolutional neural network face recognition method based on multi-scale pooling
CN107392019A (en) A kind of training of malicious code family and detection method and device
CN104616664B (en) A kind of audio identification methods detected based on sonograph conspicuousness
CN110147743A (en) Real-time online pedestrian analysis and number system and method under a kind of complex scene
CN109948425A (en) A kind of perception of structure is from paying attention to and online example polymerize matched pedestrian's searching method and device
CN109344821A (en) Small target detecting method based on Fusion Features and deep learning
CN106683091A (en) Target classification and attitude detection method based on depth convolution neural network
CN111858882B (en) Text visual question-answering system and method based on concept interaction and associated semantics
CN110097029B (en) Identity authentication method based on high way network multi-view gait recognition
CN114092697B (en) Building facade semantic segmentation method with attention fused with global and local depth features
CN109785344A (en) The remote sensing image segmentation method of binary channel residual error network based on feature recalibration
CN106408030A (en) SAR image classification method based on middle lamella semantic attribute and convolution neural network
CN108647695A (en) Soft image conspicuousness detection method based on covariance convolutional neural networks
CN107463881A (en) A kind of character image searching method based on depth enhancing study
CN113343901A (en) Human behavior identification method based on multi-scale attention-driven graph convolutional network
CN106778526A (en) A kind of extensive efficient face identification method based on Hamming distance
CN107463954A (en) A kind of template matches recognition methods for obscuring different spectrogram picture
CN113487600B (en) Feature enhancement scale self-adaptive perception ship detection method
CN108985200A (en) A kind of In vivo detection algorithm of the non-formula based on terminal device
CN111222534B (en) Single-shot multi-frame detector optimization method based on bidirectional feature fusion and more balanced L1 loss
CN108021920A (en) A kind of method that image object collaboration is found
CN111709443B (en) Calligraphy character style classification method based on rotation invariant convolution neural network

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20170201