CN108985298B - Human body clothing segmentation method based on semantic consistency - Google Patents

Human body clothing segmentation method based on semantic consistency Download PDF

Info

Publication number
CN108985298B
CN108985298B CN201810631795.4A CN201810631795A CN108985298B CN 108985298 B CN108985298 B CN 108985298B CN 201810631795 A CN201810631795 A CN 201810631795A CN 108985298 B CN108985298 B CN 108985298B
Authority
CN
China
Prior art keywords
train
semantic
picture
image
clothes
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
Application number
CN201810631795.4A
Other languages
Chinese (zh)
Other versions
CN108985298A (en
Inventor
李玺
吉炜
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.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
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 Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN201810631795.4A priority Critical patent/CN108985298B/en
Publication of CN108985298A publication Critical patent/CN108985298A/en
Application granted granted Critical
Publication of CN108985298B publication Critical patent/CN108985298B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a human body clothing segmentation method based on semantic consistency, which is used for analyzing the semantic condition of a clothing region of each frame under the condition of giving a single-frame single-person clothing picture. The method specifically comprises the following steps: acquiring an image data set for training human body clothes segmentation, and defining an algorithm target; searching adjacent pictures of each single-frame image in the semantic space of the single-frame image in the data set and forming a picture pair; jointly modeling the neighborhood relationship of each set of picture pairs in the streaming space; establishing a prediction model for clothes segmentation; and analyzing semantic information of clothes in the picture by using the prediction model. The method is suitable for clothes segmentation analysis in a real image, and has better effect and robustness in the face of various complex conditions.

Description

Human body clothing segmentation method based on semantic consistency
Technical Field
The invention relates to the field of computer vision, in particular to a human body clothing segmentation method based on semantic consistency.
Background
The semantic segmentation of the clothes is used as a low-level visual technology and is often used as auxiliary information of some high-level visual tasks, such as clothes retrieval, clothes attribute analysis and the like. The aim of clothes segmentation is to give an image and predict the classification label of each pixel point in the image. The key factors of the clothes segmentation mainly include the characteristics of large interior appearance difference of the clothes types, non-rigidity of the clothes and extreme deformability of the clothes. The traditional method generally considers the clothes segmentation task as a semantic segmentation problem, and although some methods make a breakthrough in classification accuracy, the traditional method does not fully utilize the information of the existing data.
Due to the effectiveness of statistical modeling, current learning-based methods are increasingly applied to semantic segmentation tasks. The existing learning-based method mainly adopts an end-to-end deep learning framework, inputs an original three-channel color picture and outputs a predicted semantic segmentation picture. The deep learning can effectively solve the problem of feature representation, but when the problem that enough accurately labeled data sets are lacked in clothes segmentation is faced, the effect of the deep learning is limited due to the insufficient data quantity, and meanwhile, reasonable features cannot be extracted by common convolution due to the deformable characteristic of clothes.
Disclosure of Invention
Aiming at the problems and the requirement of power distribution network construction in practice, the invention provides a human body clothes segmentation method based on semantic consistency. The technical scheme adopted by the invention is as follows:
a human body clothing segmentation method based on semantic consistency comprises the following steps:
s1, acquiring an image data set for training human body clothes segmentation, and defining an algorithm target;
s2, searching adjacent pictures of each single-frame image in the semantic space of the single-frame image in the data set and forming a picture pair;
s3, carrying out joint modeling on adjacent relations of each group of picture pairs in a flow type space (namely a semantic space);
s4, establishing a prediction model for clothes division;
and S5, analyzing semantic information of clothes in the picture by using the prediction model.
Preferably, the image data set in S1 includes a single frame image ItrainAnd artificially labeled semantic segmentation graph Ptrain(ii) a The algorithm aims at predicting the semantic segmentation map of the clothes in the single-frame image
Figure BDA0001700195740000023
Preferably, the S2 includes the following substeps:
s21, for each single frame image ItrainExtracting human body posture characteristic C from pre-trained human body posture estimation model OpenposposeAnd image appearance characteristic CappearanceBy cascading CposeAnd CappearanceObtaining a single frame image ItrainCorresponding picture feature GI；
S22, calculating the similarity of the pictures in the image data set pairwise, wherein any two pictures ItrainAnd l'trainThe similarity α of (a) is calculated as follows:
Figure BDA0001700195740000021
wherein C isemptyIs a reaction with GIPicture features of one image with the same picture size and all values of 0; c'IIs for single frame image I'trainAccording to a single frame image ItrainPicture characteristics obtained by the same method; euclidean () represents the calculation of the Euclidean distance;
s23, calculating and comparing the similarity for each single frame image ItrainThe most similar image I' is searchedtrainTo obtain a picture pair (I)train，I＇train) And a corresponding similarity value alpha.
Preferably, the S3 includes the following substeps:
s31, using four layers of convolution operation and pooling operation to pair I in picture pairtrainAnd l'trainRespectively extracting the features to obtain SIAnd S'INamely:
SI＝fsingle(Itrain；θ)
S′I＝fsingle(I′train；θ)
wherein f issingle() A function is constructed for the four-layer convolution operation and the pooling operation, and theta is a convolution parameter;
S32for feature S obtained in S31IAnd S'IPerforming fusion to obtain fused features Sinteraction：
Sinteraction＝(1-α)*SI+α*S′I
S33, using triple-layer convolution operation on fused features SinteractionReconstructing semantic information of the picture to obtain an image I with the sizetrainOne-eighth sized semantic segmentation map
Figure BDA0001700195740000022
Simultaneous four-layer convolution and upsampling operations on a single image feature SIReconstructing semantic information to obtain a semantic information reconstruction image ItrainSemantic segmentation graph with consistent size
Figure BDA0001700195740000024
S34, the operations of S31-S33 are performed on all the picture pairs.
Preferably, the S4 includes the following substeps:
s41, establishing a deep convolution neural network, wherein the input of the neural network is a pair of pictures (I)train，I′train) Output as being relative to picture ItrainBy semantically segmenting the image
Figure BDA0001700195740000031
The structure of the neural network is represented as a map
Figure BDA0001700195740000032
Is formulated as:
Figure BDA0001700195740000033
wherein theta is1The convolution parameters used when the semantic segmentation result is predicted for the prediction model, and f () is a prediction function of the deep convolution neural network;
s42, the loss function of the neural network is as follows:
Figure BDA0001700195740000034
wherein P and PsmallRespectively represent
Figure BDA00017001957400000310
Corresponding true semantic segmentation image and
Figure BDA0001700195740000035
corresponding real semantic segmentation graph;
Figure BDA0001700195740000036
predictive semantic segmentation graph representing an original scale
Figure BDA0001700195740000037
And loss error of its true semantic picture;
Figure BDA0001700195740000038
predictive semantic segmentation graph representing small scales
Figure BDA0001700195740000039
And loss error of the real semantic picture, wherein the small scale is one eighth of the original scale; λ is a weight parameter;
s43, training the whole neural network under the loss function L by using an Adam optimization method and a back propagation algorithm until the neural network converges.
The method is based on the deep neural network, utilizes the adjacent relation of semantic information of similar pictures in a flow type space, adopts deformable convolution to model deformation characteristics of clothes, and can better adapt to clothes semantic segmentation under different scenes. Compared with the traditional clothes semantic segmentation method, the method has the following benefits:
firstly, the clothes semantic segmentation method defines three important problems in clothes semantic segmentation, namely the extreme deformable characteristic of clothes, the modeling of semantic consistent relation of similar pictures and the calculation accuracy. By seeking a solution of the two directions, the semantic segmentation of the clothes under the condition of insufficient data quantity can be effectively solved.
Secondly, the clothing semantic segmentation method is based on the deep convolutional neural network, a semantic consistency model is established, and meanwhile the method has the advantage of calculation accuracy. The deep convolutional neural network can better express visual features, in addition, the extraction of the visual features and the learning of the corresponding structural model are unified in the same frame, and the final effect of the method is improved.
Finally, the clothes semantic segmentation method of the invention provides that the semantic consistency relation of similar picture pairs is modeled by using a convolutional neural network, so as to predict clothes semantic segmentation, and the feature information of clothes is extracted by using variability convolution according to the characteristic of clothes variability. The method can effectively mine the semantic consistency relationship of the picture pairs with similar contents and keep the constraint on the semantic consistency structure in the semantic space.
The method can effectively improve the accuracy and efficiency of retrieval and analysis in the process of clothing retrieval and clothing attribute analysis, and has good application value. For example, in the application scene of clothing e-commerce retail, the method can quickly and accurately analyze the clothing regions and categories on the model, so that the clothing categories on the model can be quickly completed, and a basis is provided for e-commerce retail of the same type retrieval.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a graph showing the effect of the experiment according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
On the contrary, the invention is intended to cover alternatives, modifications, equivalents and alternatives which may be included within the spirit and scope of the invention as defined by the appended claims. Furthermore, in the following detailed description of the present invention, certain specific details are set forth in order to provide a better understanding of the present invention. It will be apparent to one skilled in the art that the present invention may be practiced without these specific details.
Referring to fig. 1, a human clothing segmentation method based on semantic consistency includes the following steps:
s1, acquiring an image data set for training human body clothes segmentation, and defining an algorithm target;
the image data set in this step includes a single frame image ItrainAnd artificially labeled semantic segmentation graph Ptrain(ii) a The algorithm aims at predicting the semantic segmentation map of the clothes in the single-frame image
Figure BDA0001700195740000041
S2, searching adjacent pictures of each single-frame image in the semantic space of the single-frame image in the data set and forming a picture pair;
the method comprises the following substeps:
s21, for each single frame image ItrainExtracting human body posture characteristic C from pre-trained human body posture estimation model OpenposposeAnd image appearance characteristic CappearanceBy concatenation (i.e. direct splicing) of CposeAnd CappearanceObtaining a single frame image ItrainCorresponding picture feature GI；
And S22, calculating the similarity of every two pictures in the image data set to obtain the similarity of any two pictures. Wherein, any two pictures ItrainAnd l'trainThe similarity α of (a) is calculated as follows:
Figure BDA0001700195740000051
wherein C isemptyIs a reaction with GIPicture features of one image with the same picture size and all values of 0; c'IIs for single frame image I'trainAccording to a single frame image ItrainPicture characteristics obtained by the same method; euclidean () represents the calculation of the Euclidean distance;
s23, calculating and comparing the similarity for each single frame image ItrainThe most similar image I' is searchedtrainTo obtain a picture pair (I)train，I＇train) And a corresponding similarity value alpha.
S3, carrying out joint modeling on adjacent relations of each group of picture pairs in a flow type space (namely a semantic space);
the method comprises the following substeps:
s31, using four layers of convolution operation and pooling operation to pair of pictures (I)train，I＇train) In (1)trainAnd ItrainRespectively extracting the features to obtain SIAnd S'INamely:
SI＝fsingle(Itrain；θ)
S′I＝fsingle(I′train；θ)
wherein f issingle() A function is constructed for the four-layer convolution operation and the pooling operation, and theta is a convolution parameter;
s32, characteristic S obtained in S31IAnd S'IPerforming fusion to obtain fused features Sinteraction：
Sinteraction＝(1-α)*SI+α*S′I
Wherein α is the similarity value of the group of pictures;
s33, using triple-layer convolution operation on fused features SinteractionReconstructing semantic information of the picture to obtain an image I with the sizetrainOne-eighth sized semantic segmentation map
Figure BDA0001700195740000052
Simultaneous four-layer convolution and upsampling operations on a single image feature SIReconstructing semantic information to obtain a semantic information reconstruction image ItrainSemantic segmentation graph with consistent size
Figure BDA0001700195740000056
S34, the operations of S31-S33 are performed on all the picture pairs.
S4, establishing a prediction model for clothes division;
the method comprises the following substeps:
s41, establishing a deep convolution neural network, wherein the input of the neural network is a pair of pictures (I)train，I′train) Output as being relative to picture ItrainBy semantically segmenting the image
Figure BDA0001700195740000053
The structure of the neural network is represented as a map
Figure BDA0001700195740000054
Is formulated as:
Figure BDA0001700195740000055
wherein theta is1The convolution parameters used when the semantic segmentation result is predicted for the prediction model, and f () is a prediction function of the deep convolution neural network;
s42, the loss function of the neural network is as follows:
Figure BDA0001700195740000061
wherein P represents
Figure BDA0001700195740000062
Corresponding true semantically segmented images, i.e. images ItrainTrue semantic segmentation graph, PsmallTo represent
Figure BDA0001700195740000063
Corresponding true semantic segmentation map, i.e. of size image ItrainOne eighth of a real semantic segmentation graph;
Figure BDA0001700195740000064
predictive semantic segmentation graph representing an original scale
Figure BDA0001700195740000065
And its true semantic picture (i.e., the labeled semantic segmentation map in S1);
Figure BDA0001700195740000066
predictive semantic segmentation graph representing small scales
Figure BDA0001700195740000067
And loss error of its true semantic picture (i.e., the labeled semantic segmentation map in S1), the small scale being one-eighth of the original scale, and
Figure BDA0001700195740000068
the dimensions remain the same; λ is a weight parameter, here the value is 0.125;
s43, training the whole neural network under the loss function L by using an Adam optimization method and a back propagation algorithm until the neural network converges.
And S5, analyzing semantic information of clothes in the picture by using the prediction model.
The above-described method is applied to specific examples so that those skilled in the art can better understand the effects of the present invention.
Examples
The implementation method of this embodiment is as described above, and specific steps are not elaborated, and the effect is shown only for case data. The invention is implemented on three data sets with truth labels, which are respectively as follows:
fashinostat v0.2 dataset: the data set contains 685 images and has 56 types of semantic labels.
Refined Fashionista dataset: the data set contains 685 images and has 25 types of semantic labels.
CFPD dataset: the data set contained 2682 images with 23 classes of semantic labels.
In this example, a picture is selected from each data set for experiment, the closest picture is obtained by calculating the similarity, then the features of the two pictures are respectively extracted, and the adjacent relation of the group of pictures in the streaming space is subjected to combined modeling, so as to obtain the final semantic segmentation graph, as shown in fig. 2. In the figure, the grountruth represents a real semantic segmentation graph, and the predicted semantic segmentation graph obtained by the method is basically consistent with the real semantic segmentation graph.
The detection precision of the detection result of the embodiment is shown in the following table, and the detection precision of various methods is mainly compared by using two indexes, namely average Acc and IoU, wherein the average Acc index refers to the accuracy of the classification result of each pixel point, and can better reflect the prediction result; IoU refers to the intersection ratio of the areas between the semantic region and the truth region. As shown in the table, the average Acc and IoU indexes of the method have obvious advantages compared with other traditional methods.
Figure BDA0001700195740000071
In the above embodiment, the clothing semantic segmentation method of the present invention first performs joint modeling on the adjacent relationship of each group of pictures with similar contents in the flow pattern space. On the basis, the original problem is converted into an end-to-end structure learning problem, and a clothing semantic segmentation model is established based on a deep neural network. And finally, predicting the clothes semantic information of a new frame by using the trained clothes semantic segmentation model.
Through the technical scheme, the embodiment of the invention develops a human body clothes segmentation method based on semantic consistency based on a deep learning technology. The method can utilize the adjacent relation of semantic information of similar pictures in a streaming space and adopt deformable convolution to model deformation characteristics of clothes, and can better adapt to clothes semantic segmentation under different scenes.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (4)

1. A human body clothing segmentation method based on semantic consistency is characterized by comprising the following steps:
s1, acquiring an image data set for training human body clothes segmentation, and defining an algorithm target;
s2, searching adjacent pictures of each single-frame image in the semantic space of the single-frame image in the data set and forming a picture pair;
s3, carrying out joint modeling on the adjacent relation of each group of picture pairs in the flow type space;
s4, establishing a prediction model for clothes division;
s5, analyzing semantic information of clothes in the picture by using the prediction model;
the S3 includes the following substeps:
s31, using four layers of convolution operation and pooling operation to pair I in picture pairtrainAnd l'trainRespectively extracting the features to obtain SIAnd S'INamely:
SI＝fsingle(Itrain；θ)
S′I＝fsingle(I′train；θ)
wherein f issingle() A function is constructed for the four-layer convolution operation and the pooling operation, and theta is a convolution parameter;
s32, characteristic S obtained in S31IAnd S'IPerforming fusion to obtain fused features Sinteraction：
Sinteraction＝(1-α)*SI+α*S′I
S33, using triple-layer convolution operation on fused features SinteractionReconstructing semantic information of the picture to obtain an image I with the sizetrainOne-eighth sized semantic segmentation map
Figure FDA0003317680900000011
Simultaneous use of four-layer convolution and up-cutSample operation on single image features SIReconstructing semantic information to obtain a semantic information reconstruction image ItrainSemantic segmentation graph with consistent size
Figure FDA0003317680900000012
S34, the operations of S31-S33 are performed on all the picture pairs.
2. The method for segmenting human body clothes based on semantic consistency as claimed in claim 1, wherein the image data set in S1 comprises a single frame image ItrainAnd artificially labeled semantic segmentation graph Ptrain(ii) a The algorithm aims at predicting the semantic segmentation map of the clothes in the single-frame image
Figure FDA0003317680900000013
3. The human clothing segmentation method based on semantic consistency as claimed in claim 1, wherein the S2 comprises the following sub-steps:
s21, for each single frame image ItrainExtracting human body posture characteristic C from pre-trained human body posture estimation model OpenposposeAnd image appearance characteristic CappearanceBy cascading CposeAnd CappearanceObtaining a single frame image ItrainCorresponding Picture feature CI；
S22, calculating the similarity of the pictures in the image data set pairwise, wherein any two pictures ItrainAnd l'trainThe similarity α of (a) is calculated as follows:
Figure FDA0003317680900000021
wherein C isemptyIs and CIPicture features of one image with the same picture size and all values of 0; c'IIs for single frame image I'trainAccording to a single frame image ItrainPicture characteristics obtained by the same method; euclidean () represents the calculation of the Euclidean distance;
s23, calculating and comparing the similarity for each single frame image ItrainRetrieving the most similar picture I'trainTo obtain a picture pair (I)train，I′train) And a corresponding similarity value alpha.
4. The human clothing segmentation method based on semantic consistency as claimed in claim 1, wherein the S4 comprises the following sub-steps:
s41, establishing a deep convolution neural network, wherein the input of the neural network is a pair of pictures (I)train，I′train) Output as being relative to picture ItrainBy semantically segmenting the image
Figure FDA0003317680900000022
The structure of the neural network is represented as a map
Figure FDA0003317680900000023
Is formulated as:
Figure FDA0003317680900000024
wherein theta is1The convolution parameters used when the semantic segmentation result is predicted for the prediction model, and f () is a prediction function of the deep convolution neural network;
s42, the loss function of the neural network is as follows:
Figure FDA0003317680900000025
wherein P and PsmallRespectively represent
Figure FDA0003317680900000026
Corresponding realitySemantically segmenting an image and
Figure FDA0003317680900000027
a corresponding real semantic segmentation map is generated,
Figure FDA0003317680900000028
predictive semantic segmentation graph representing an original scale
Figure FDA0003317680900000029
And loss error of its true semantic picture;
Figure FDA00033176809000000210
predictive semantic segmentation graph representing small scales
Figure FDA00033176809000000211
And loss error of the real semantic picture, wherein the small scale is one eighth of the original scale; λ is a weight parameter;
s43, training the whole neural network under the loss function L by using an Adam optimization method and a back propagation algorithm until the neural network converges.
CN201810631795.4A 2018-06-19 2018-06-19 Human body clothing segmentation method based on semantic consistency Active CN108985298B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810631795.4A CN108985298B (en) 2018-06-19 2018-06-19 Human body clothing segmentation method based on semantic consistency

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810631795.4A CN108985298B (en) 2018-06-19 2018-06-19 Human body clothing segmentation method based on semantic consistency

Publications (2)

Publication Number Publication Date
CN108985298A CN108985298A (en) 2018-12-11
CN108985298B true CN108985298B (en) 2022-02-18

Family

ID=64540714

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810631795.4A Active CN108985298B (en) 2018-06-19 2018-06-19 Human body clothing segmentation method based on semantic consistency

Country Status (1)

Country Link
CN (1) CN108985298B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109858539A (en) * 2019-01-24 2019-06-07 武汉精立电子技术有限公司 A kind of ROI region extracting method based on deep learning image, semantic parted pattern
CN110807462B (en) * 2019-09-11 2022-08-30 浙江大学 Training method insensitive to context of semantic segmentation model
CN111028249A (en) * 2019-12-23 2020-04-17 杭州知衣科技有限公司 Garment image segmentation method based on deep learning
CN114092591B (en) * 2022-01-20 2022-04-12 中国科学院自动化研究所 Image generation method, image generation device, electronic equipment and storage medium
CN116543147A (en) * 2023-03-10 2023-08-04 武汉库柏特科技有限公司 Carotid ultrasound image segmentation method, device, equipment and storage medium

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2002075685A3 (en) * 2001-03-15 2003-03-13 Koninkl Philips Electronics Nv Automatic system for monitoring persons entering and leaving a changing room
GB2403363A (en) * 2003-06-25 2004-12-29 Hewlett Packard Development Co Tags for automated image processing
CN1920820A (en) * 2006-09-14 2007-02-28 浙江大学 Image meaning automatic marking method based on marking significance sequence
US8173772B2 (en) * 2005-12-30 2012-05-08 Spiber Technologies Ab Spider silk proteins and methods for producing spider silk proteins
CN105261017A (en) * 2015-10-14 2016-01-20 长春工业大学 Method for extracting regions of interest of pedestrian by using image segmentation method on the basis of road restriction
CN106327469A (en) * 2015-06-29 2017-01-11 北京航空航天大学 Video object segmentation method based on semantic label guidance
CN107729804A (en) * 2017-08-31 2018-02-23 广东数相智能科技有限公司 A kind of people flow rate statistical method and device based on garment ornament

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2002075685A3 (en) * 2001-03-15 2003-03-13 Koninkl Philips Electronics Nv Automatic system for monitoring persons entering and leaving a changing room
GB2403363A (en) * 2003-06-25 2004-12-29 Hewlett Packard Development Co Tags for automated image processing
US8173772B2 (en) * 2005-12-30 2012-05-08 Spiber Technologies Ab Spider silk proteins and methods for producing spider silk proteins
CN1920820A (en) * 2006-09-14 2007-02-28 浙江大学 Image meaning automatic marking method based on marking significance sequence
CN106327469A (en) * 2015-06-29 2017-01-11 北京航空航天大学 Video object segmentation method based on semantic label guidance
CN105261017A (en) * 2015-10-14 2016-01-20 长春工业大学 Method for extracting regions of interest of pedestrian by using image segmentation method on the basis of road restriction
CN107729804A (en) * 2017-08-31 2018-02-23 广东数相智能科技有限公司 A kind of people flow rate statistical method and device based on garment ornament

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Clothes Co-Parsing Via Joint Image Segmentation and Labeling With Application to Clothing Retrieval;xiaodan Liang 等;《IEEE TRANSACTIONS ON MULTIMEDIA》;20160630;第18卷(第6期);1175-1186 *
Retrieving Similar Styles to Parse Clothing;Kota Yamaguchi 等;《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》;20150531;第37卷(第5期);1028-1040 *
Where to Buy It: Matching Street Clothing Photos in Online Shops;M. Hadi Kiapour 等;《2015 IEEE International Conference on Computer Vision》;20160218;1-9 *
基于深度学习的视觉SLAM综述;赵洋 等;《机器人》;20171115;第39卷(第6期);889-896 *

Also Published As

Publication number Publication date
CN108985298A (en) 2018-12-11

Similar Documents

Publication Publication Date Title
CN111291212B (en) Zero sample sketch image retrieval method and system based on graph convolution neural network
CN108985298B (en) Human body clothing segmentation method based on semantic consistency
CN111858954B (en) Task-oriented text-generated image network model
Yang et al. Visual sentiment prediction based on automatic discovery of affective regions
Wang et al. Towards unified depth and semantic prediction from a single image
Song et al. Deep depth super-resolution: Learning depth super-resolution using deep convolutional neural network
CN106547880B (en) Multi-dimensional geographic scene identification method fusing geographic area knowledge
JP6395158B2 (en) How to semantically label acquired images of a scene
Häne et al. Dense semantic 3d reconstruction
Zhang et al. Deep hierarchical guidance and regularization learning for end-to-end depth estimation
Fooladgar et al. A survey on indoor RGB-D semantic segmentation: from hand-crafted features to deep convolutional neural networks
Wen et al. CF-SIS: Semantic-instance segmentation of 3D point clouds by context fusion with self-attention
CN108564012B (en) Pedestrian analysis method based on human body feature distribution
Cai et al. A robust interclass and intraclass loss function for deep learning based tongue segmentation
Qian et al. Learning pairwise inter-plane relations for piecewise planar reconstruction
Song et al. Contextualized CNN for scene-aware depth estimation from single RGB image
Xu et al. RGB-T salient object detection via CNN feature and result saliency map fusion
Wang et al. Improving random walker segmentation using a nonlocal bipartite graph
Parente et al. Integration of convolutional and adversarial networks into building design: A review
Zhang et al. DHNet: Salient object detection with dynamic scale-aware learning and hard-sample refinement
Robert The Role of Deep Learning in Computer Vision
Li et al. [Retracted] Deep‐Learning‐Based 3D Reconstruction: A Review and Applications
CN110942463B (en) Video target segmentation method based on generation countermeasure network
Liu et al. Attention-embedding mesh saliency
Wang et al. Interactive image segmentation based on label pair diffusion

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