CN109784350A - In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network - Google Patents
In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network Download PDFInfo
- Publication number
- CN109784350A CN109784350A CN201811634796.0A CN201811634796A CN109784350A CN 109784350 A CN109784350 A CN 109784350A CN 201811634796 A CN201811634796 A CN 201811634796A CN 109784350 A CN109784350 A CN 109784350A
- Authority
- CN
- China
- Prior art keywords
- globalnet
- key point
- network
- refinenet
- dress ornament
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to the dress ornament key independent positioning methods of a kind of combination cavity convolution and cascade pyramid network, and include three parts: ResNet-101, GlobalNet and RefineNet carry out image characteristics extraction by ResNet-101;GlobalNet carries out simple crucial point location;RefineNet integrates the characteristic present from GlobalNet, identifies remaining difficult key point.
Description
Technical field
The present invention relates to fashion world, field of image processing, key point positioning field, deep learning fields, by cascade gold
Word tower network (Cascade Pyramid Network, CPN) is combined with empty convolution and is improved, and realizes dress ornament key point
Location tasks.
Background technique
In recent years, with the fast development of electric business platform and fashion industry, it is more next for the algorithm requirements of dress ornament analysis
It is more urgent.Dress ornament key point location can effectively promote the alignment of dress ornament position, accelerate thingness identification, divide image can automatically
Class ownership, has caused social extensive concern.Be applied to human body critical point detection algorithm at present and have been achieved with tremendous development, but with
In the mutual fusion process of fashion industry, since dress ornament is in classification, ratio and apparent variability, dress ornament key point location algorithm
Still suffer from significant challenge.For human body key point location, most methods are all the coordinates for directly returning out human joint points,
But the flexibility and regression model scalability due to human motion are poor, the effect of such method is not ideal.
With the development of depth learning technology, answered extensively on image classification, identification and critical point detection
With CPM (Convolutional Pose Machines, CPM) network of the propositions such as Wei in 2016 passes through ordered convolution side
Formula carries out the expression of spatial information and texture information, realizes the stronger key point location algorithm of robustness.The same year Alejandro
Deng proposition Hourglass (Stack Hourglass Networks) network, passes through and introduce the full convolutional neural networks of multimode
(Convolutional Neural Network, CNN) solves single crucial point location, and each CNN module captures different rulers
The feature of picture is spent, human body spatial relationship is found with this, infers the artis position of human body.Then, more people's critical point detections
Algorithm gradually appears, and effect is preferably top-down algorithm, i.e., first detects one, repositions everyone key point.
The G-RMI algorithm of the propositions such as Papandreou in 2017 then makes first with multiple people in FasterR-CNN detection figure
Key point is accurately positioned with depth residual error network (Deep Residual Networks, ResNet);The same year, He Kaiming was in R-
MASK R-CNN innovatory algorithm is proposed on the basis of CNN, Fast R-CNN and Faster R-CNN, in example segmentation, bounding box inspection
The effect for being better than single model is all obtained in the multiple tasks such as survey and human body key point location;Subsequent RMPE algorithm is to overcome by list
People's detection block difference and the problem of cause key point position error, utilize the single goal detection algorithm SSD of pyramid structure
(Single ShotDetector, SSD) detects single people, reuses the key point inspection that Hourglass network carries out single posture
It surveys.Chen etc. artificially solves more difficult critical point detection and proposes cascade pyramid structure network (Cascaded Pyramid
Network, CPN), first by multiple people in the target detection topology discovery figure of MASK R-CNN, pass through later
The cascade network of GlobalNet (Global Pyramid Network) and RefineNet (Refined Pyramid Network)
Network structure is realized to everyone more difficult critical point detection, and human body critical point detection challenge match champion in 2017 is won.CPN network
Simply greatly improve the accuracy of positioning with difficult key point by distinguishing, but network is not still well by image
Low-level details information is used for crucial point location, it is therefore desirable to be further improved.
Summary of the invention
The object of the present invention is to provide the clothes that image low-level details information can be preferably used for crucial point location by one kind
Adorn crucial independent positioning method: ICPN (Improved Cascaded Pyramid Network, ICPN).ICPN algorithm is for pass
The semantic information of different levels merges problem in key point location task, using empty convolution, is not reducing the impression of high-level characteristic figure
In the case where open country, the spatial resolution of characteristic pattern is improved, to obtain more image detail information features, is further promoted and is closed
Key point detection accuracy is improved the robustness of algorithm by a variety of data enhancement operations, and avoids sky by the cutting of corresponding feature
Hole convolution bring computation complexity becomes larger problem.Technical solution is as follows:
The dress ornament key independent positioning method of a kind of combination cavity convolution and cascade pyramid network, includes three parts:
ResNet-101, GlobalNet and RefineNet carry out image characteristics extraction by ResNet-101;GlobalNet is carried out
Simple key point location;RefineNet integrates the characteristic present from GlobalNet, identifies remaining difficult key point, packet
It includes:
1) ResNet-101 feature extraction network: the input picture for being N × N for a Zhang great little introduces shortcut and skips certain
The connection of a little layers, then converge with main diameter.
2) based on GlobalNet extraction different scale feature cascade pyramid structure module: Conv4-Conv5 with
Empty convolution replaces the convolution operation of script, does not reduce spatial resolution while increasing receptive field, generates space ruler respectively
Degree is 256 × N/4 × N/4, and 512 × N/8 × N/8,512 × N/8 × N/8,512 × N/8 × N/8 characteristic pattern, latter three groups special
Sign figure scale is the same, and the characteristic pattern Conv2 and Conv3 of bottom have relatively high spatial resolution, but semantic information is relatively low;
And high-rise characteristic pattern Conv4 and Conv5 includes more semantic informations and spatial resolution does not reduce.
3) three groups of features after empty convolution makes the fusion different scale feature cascade module based on GlobalNet: are introduced
Figure has been of the same size and can directly be added, and only merges after the last layer needs to carry out up-sampling operation.
4) characteristic present from GlobalNet is polymerize come location difficulty key point using RefineNet,
The characteristic pattern that Conv2 and Conv4 is generated only is remained in RefineNet.
5) data such as corresponding image rotation, thermodynamic chart Gaussian Blur network training and test: are carried out for training image
Enhancing operates to improve data volume and promote network robustness, and by test data set test result, it is fixed to export dress ornament key point
Error rate of the key point coordinate of position result figure and final result relative to true tag.
The present invention carries out the task of dress ornament key point location by the method that empty convolution is combined with CPN network, with one
A little classical methods compare, and advantage is mainly reflected in:
Novelty: artificial intelligence is introduced fashion world by the present invention, effectively improves the precision of dress ornament key point location,
There is great commercial application value under the scenes such as electric business, fashion collocation.The present invention passes through empty convolution and improves CPN network, overcomes
The problem of feature pyramid structure in former network can largely lose the low-level details information in further feature figure is closed in dress ornament
Preferable effect is achieved in key point location task.
Robustness: algorithm of the invention is applicable to a variety of key point location tasks, and the present invention is revolved by corresponding image
Turn, the data enhancement operations such as thermodynamic chart Gaussian Blur further increase the robustness of model.
Detailed description of the invention
The improved CPN network algorithm frame of Fig. 1
Fig. 2 ResNet schematic network structure
(a) (b) (c) of Fig. 3 is empty convolution principle schematic diagram
Fig. 4 feature pyramid structure schematic diagram, (a) primitive character pyramid (b) improve feature pyramid
Fig. 5 heterogeneous networks testing result comparison diagram, (a) true tag;(b)Horglass;(c)CPM;(d)CPN; (e)
ICPN
Specific embodiment
The present invention includes three parts: ResNet-101, GlobalNet and RefineNet altogether.Wherein pass through ResNet-
101 carry out image characteristics extraction;GlobalNet is by the improved pyramid structure progress Fusion Features of empty convolution and simply
Crucial point location;RefineNet integrates the characteristic present from GlobalNet, polymerize the feature of different dimensions, identifies remaining
Difficult key point, to avoid becoming larger because of empty convolution bring computation complexity, the present invention carries out corresponding feature in the part
It cuts, only remains the characteristic pattern that Conv2 and Conv4 is generated.
Fig. 1 is total algorithm frame of the invention.Dress ornament key point location of the empty convolution in conjunction with CPN includes following several
Step:
1) feature extraction network, the input picture for being N × N for a Zhang great little, the present invention pass through ResNet-101 first
Carry out feature extraction, as shown in Fig. 2, the general network that compares, ResNet introduces the connection that shortcut skips certain layers, then with main diameter
Converge, so that the error of bottom can solve the problems, such as that gradient disappears to upper layer transfers by shortcut, is not increasing additional parameter
Increase the training speed of network model while not improving computation complexity again, improve training effect.
2) GlobalNet extracts the cascade pyramid structure module of different scale feature, as shown in figure 4, C2-C5 generation respectively
In table residual error network Conv2-Conv5 generate characteristic pattern, be different from original pyramid structure characteristic pattern scale (Fig. 4 (a)) by
The characteristics of layer successively decreases, (Fig. 4 (b)) of the invention replace the convolution operation of script in Conv4-Conv5 with empty convolution (such as Fig. 3),
Spatial resolution is not reduced while increasing receptive field, generating space scale respectively is 256 × N/4 × N/4,512 × N/8
× N/8,512 × N/16 × N/16,512 × N/16 × N/16 characteristic pattern.The characteristic pattern C2 and C3 of bottom have relatively high sky
Between resolution ratio, but semantic information is relatively low;And high-rise characteristic pattern C4, C5 includes more semantic informations and spatial discrimination
Rate does not reduce.
3) GlobalNet merges pyramid structure different scale characteristic module, as shown in figure 4, original pyramid structure
In each layer characteristic pattern carry out being required to carry out up-sampling operation when top-down Fusion Features so that characteristic pattern scale is unanimously again
Addition fusion is carried out, this inevitably affects the quality of characteristic pattern, and the present invention is due to introducing three after empty convolution makes
Group characteristic pattern has been of the same size and can directly be added, and only merges after the last layer needs to carry out up-sampling operation.
4) RefineNet, which polymerize the characteristic present from GlobalNet, carrys out location difficulty key point, and the present invention exists
The characteristic pattern that Conv2 and Conv4 is generated only is remained in RefineNet, it is intended under the premise of not influencing key point locating effect
Feature redundancy is reduced, the increase due to introducing empty convolution and bring calculation amount is reduced.
5) Alibaba Tianchi contest 2018FashionAI dress ornament key point location data collection training network is used, for
Training image carries out the data enhancement operations such as corresponding image rotation, thermodynamic chart Gaussian Blur to improve data volume and promote network
Robustness.By test data set test result, the key point for exporting dress ornament key point positioning result figure and final result is sat
Mark the error rate relative to true tag.
The testing result comparison of heterogeneous networks is as shown in Figure 5.Crucial positioning result of the invention closest to true tag, and
Key point position error in remaining network is larger, the more difficult positioning of key point especially more similar with background, algorithms of different
As a result gap is obvious.Specific normalization error rate is as shown in table 1, the present invention respectively jacket, housing, one-piece dress, half body skirt,
Best result is obtained respectively in the dress ornament key point location of five kinds of classifications of trousers and entirety.
The normalization detection error rate of 1 heterogeneous networks of table
Claims (1)
1. the dress ornament key independent positioning method of a kind of combination cavity convolution and cascade pyramid network, includes three parts:
ResNet-101, GlobalNet and RefineNet carry out image characteristics extraction by ResNet-101;GlobalNet is carried out
Simple key point location;RefineNet integrates the characteristic present from GlobalNet, identifies remaining difficult key point.Packet
It includes:
1) ResNet-101 feature extraction network: the input picture for being N × N for a Zhang great little introduces shortcut and skips certain layers
Connection, then converge with main diameter;
2) the cascade pyramid structure module of the extraction different scale feature based on GlobalNet: in Conv4-Conv5 with cavity
Convolution replaces the convolution operation of script, does not reduce spatial resolution while increasing receptive field, generates space scale respectively and is
256 × N/4 × N/4,512 × N/8 × N/8,512 × N/8 × N/8,512 × N/8 × N/8 characteristic pattern, rear three groups of characteristic patterns
Scale is the same, and the characteristic pattern Conv2 and Conv3 of bottom have relatively high spatial resolution, but semantic information is relatively low;And it is high
The characteristic pattern Conv4 and Conv5 of layer include more semantic informations and spatial resolution does not reduce;
3) it the fusion different scale feature cascade module based on GlobalNet: introduces empty convolution and has made rear three groups of characteristic patterns
Being of the same size directly to be added, and only merge after the last layer needs to carry out up-sampling operation;
4) characteristic present from GlobalNet is polymerize come location difficulty key point using RefineNet,
The characteristic pattern that Conv2 and Conv4 is generated only is remained in RefineNet;
5) network training and test: carrying out the data such as corresponding image rotation, thermodynamic chart Gaussian Blur for training image enhances
It operates to improve data volume and promote network robustness, passes through test data set test result, export dress ornament key point location knot
Error rate of the key point coordinate of fruit figure and final result relative to true tag.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811634796.0A CN109784350A (en) | 2018-12-29 | 2018-12-29 | In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811634796.0A CN109784350A (en) | 2018-12-29 | 2018-12-29 | In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109784350A true CN109784350A (en) | 2019-05-21 |
Family
ID=66498853
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811634796.0A Pending CN109784350A (en) | 2018-12-29 | 2018-12-29 | In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109784350A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110175613A (en) * | 2019-06-03 | 2019-08-27 | 常熟理工学院 | Street view image semantic segmentation method based on Analysis On Multi-scale Features and codec models |
CN110348445A (en) * | 2019-06-06 | 2019-10-18 | 华中科技大学 | A kind of example dividing method merging empty convolution sum marginal information |
CN110443808A (en) * | 2019-07-04 | 2019-11-12 | 杭州深睿博联科技有限公司 | Medical image processing method and device, equipment, storage medium for the detection of brain middle line |
CN110610499A (en) * | 2019-08-29 | 2019-12-24 | 杭州光云科技股份有限公司 | Method for automatically cutting local detail picture in image |
CN110674815A (en) * | 2019-09-29 | 2020-01-10 | 四川长虹电器股份有限公司 | Invoice image distortion correction method based on deep learning key point detection |
CN110689061A (en) * | 2019-09-19 | 2020-01-14 | 深动科技(北京)有限公司 | Image processing method, device and system based on alignment feature pyramid network |
CN111339883A (en) * | 2020-02-19 | 2020-06-26 | 国网浙江省电力有限公司 | Method for identifying and detecting abnormal behaviors in transformer substation based on artificial intelligence in complex scene |
CN112132013A (en) * | 2020-09-22 | 2020-12-25 | 中国科学技术大学 | Vehicle key point detection method |
CN112418239A (en) * | 2019-08-21 | 2021-02-26 | 青岛海尔智能技术研发有限公司 | Method and device for positioning clothing key points and clothes folding machine |
CN112418046A (en) * | 2020-11-17 | 2021-02-26 | 武汉云极智能科技有限公司 | Fitness guidance method, storage medium and system based on cloud robot |
CN112529768A (en) * | 2020-12-04 | 2021-03-19 | 中山大学 | Garment editing and generating method based on generation countermeasure network |
CN113610070A (en) * | 2021-10-11 | 2021-11-05 | 中国地质环境监测院(自然资源部地质灾害技术指导中心) | Landslide disaster identification method based on multi-source data fusion |
CN114428877A (en) * | 2022-01-27 | 2022-05-03 | 西南石油大学 | Intelligent clothing matching method and system |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105469087A (en) * | 2015-07-13 | 2016-04-06 | 百度在线网络技术(北京)有限公司 | Method for identifying clothes image, and labeling method and device of clothes image |
CN107516316A (en) * | 2017-07-19 | 2017-12-26 | 郑州禅图智能科技有限公司 | It is a kind of that the method that focus mechanism is split to static human image is introduced in FCN |
CN107918780A (en) * | 2017-09-01 | 2018-04-17 | 中山大学 | A kind of clothes species and attributive classification method based on critical point detection |
CN108229445A (en) * | 2018-02-09 | 2018-06-29 | 深圳市唯特视科技有限公司 | A kind of more people's Attitude estimation methods based on cascade pyramid network |
CN108229496A (en) * | 2017-07-11 | 2018-06-29 | 北京市商汤科技开发有限公司 | The detection method and device of dress ornament key point, electronic equipment, storage medium and program |
CN108229497A (en) * | 2017-07-28 | 2018-06-29 | 北京市商汤科技开发有限公司 | Image processing method, device, storage medium, computer program and electronic equipment |
CN108229559A (en) * | 2017-12-29 | 2018-06-29 | 深圳市商汤科技有限公司 | Dress ornament detection method, device, electronic equipment, program and medium |
CN108229288A (en) * | 2017-06-23 | 2018-06-29 | 北京市商汤科技开发有限公司 | Neural metwork training and clothes method for detecting color, device, storage medium, electronic equipment |
CN108764133A (en) * | 2018-05-25 | 2018-11-06 | 北京旷视科技有限公司 | Image-recognizing method, apparatus and system |
CN108932517A (en) * | 2018-06-28 | 2018-12-04 | 中山大学 | A kind of multi-tag clothes analytic method based on fining network model |
-
2018
- 2018-12-29 CN CN201811634796.0A patent/CN109784350A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105469087A (en) * | 2015-07-13 | 2016-04-06 | 百度在线网络技术(北京)有限公司 | Method for identifying clothes image, and labeling method and device of clothes image |
CN108229288A (en) * | 2017-06-23 | 2018-06-29 | 北京市商汤科技开发有限公司 | Neural metwork training and clothes method for detecting color, device, storage medium, electronic equipment |
CN108229496A (en) * | 2017-07-11 | 2018-06-29 | 北京市商汤科技开发有限公司 | The detection method and device of dress ornament key point, electronic equipment, storage medium and program |
CN107516316A (en) * | 2017-07-19 | 2017-12-26 | 郑州禅图智能科技有限公司 | It is a kind of that the method that focus mechanism is split to static human image is introduced in FCN |
CN108229497A (en) * | 2017-07-28 | 2018-06-29 | 北京市商汤科技开发有限公司 | Image processing method, device, storage medium, computer program and electronic equipment |
CN107918780A (en) * | 2017-09-01 | 2018-04-17 | 中山大学 | A kind of clothes species and attributive classification method based on critical point detection |
CN108229559A (en) * | 2017-12-29 | 2018-06-29 | 深圳市商汤科技有限公司 | Dress ornament detection method, device, electronic equipment, program and medium |
CN108229445A (en) * | 2018-02-09 | 2018-06-29 | 深圳市唯特视科技有限公司 | A kind of more people's Attitude estimation methods based on cascade pyramid network |
CN108764133A (en) * | 2018-05-25 | 2018-11-06 | 北京旷视科技有限公司 | Image-recognizing method, apparatus and system |
CN108932517A (en) * | 2018-06-28 | 2018-12-04 | 中山大学 | A kind of multi-tag clothes analytic method based on fining network model |
Non-Patent Citations (6)
Title |
---|
FISHER YU等: ""Multi-scale context aggregation by dilated convolutions"", 《PROCEEDINGS OF 2016 INTERNATIONAL CONFERENCE ON LEARNING REPRESENTATIONS》 * |
HONGMEI SONG等: ""Pyramid Dilated Deeper ConvLSTM for Video Salient Object Detection"", 《ECCV 2018》 * |
KUO MEN等: ""Cascaded atrous convolution and spatial pyramid pooling for more accurate tumor target segmentation for rectal cancer radiotherapy"", 《PHYS MED BIOL.》 * |
YILUN CHEN 等: ""Cascaded pyramid network for multi-person pose estimation"", 《HTTPS://ARXIV.ORG/PDF/1711.07319V1.PDF》 * |
乔文凡等: ""联合膨胀卷积残差网络和金字塔池化表达的高分影像建筑物自动识别"", 《地理与地理信息科学》 * |
郑婷月等: ""基于全卷积神经网络的多尺度视网膜血管分割"", 《HTTP://KNS.CNKI.NET/KCMS/DETAIL/31.1252.O4.20181007.1416.026.HTML》 * |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110175613A (en) * | 2019-06-03 | 2019-08-27 | 常熟理工学院 | Street view image semantic segmentation method based on Analysis On Multi-scale Features and codec models |
CN110348445A (en) * | 2019-06-06 | 2019-10-18 | 华中科技大学 | A kind of example dividing method merging empty convolution sum marginal information |
CN110348445B (en) * | 2019-06-06 | 2021-07-27 | 华中科技大学 | Instance segmentation method fusing void convolution and edge information |
CN110443808A (en) * | 2019-07-04 | 2019-11-12 | 杭州深睿博联科技有限公司 | Medical image processing method and device, equipment, storage medium for the detection of brain middle line |
CN110443808B (en) * | 2019-07-04 | 2022-04-01 | 杭州深睿博联科技有限公司 | Medical image processing method and device for brain midline detection, equipment and storage medium |
CN112418239A (en) * | 2019-08-21 | 2021-02-26 | 青岛海尔智能技术研发有限公司 | Method and device for positioning clothing key points and clothes folding machine |
CN112418239B (en) * | 2019-08-21 | 2023-01-24 | 青岛海尔智能技术研发有限公司 | Method and device for positioning clothing key points and clothes folding machine |
CN110610499A (en) * | 2019-08-29 | 2019-12-24 | 杭州光云科技股份有限公司 | Method for automatically cutting local detail picture in image |
CN110610499B (en) * | 2019-08-29 | 2020-10-20 | 杭州光云科技股份有限公司 | Method for automatically cutting local detail picture in image |
CN110689061B (en) * | 2019-09-19 | 2023-04-28 | 小米汽车科技有限公司 | Image processing method, device and system based on alignment feature pyramid network |
CN110689061A (en) * | 2019-09-19 | 2020-01-14 | 深动科技(北京)有限公司 | Image processing method, device and system based on alignment feature pyramid network |
CN110674815A (en) * | 2019-09-29 | 2020-01-10 | 四川长虹电器股份有限公司 | Invoice image distortion correction method based on deep learning key point detection |
CN111339883A (en) * | 2020-02-19 | 2020-06-26 | 国网浙江省电力有限公司 | Method for identifying and detecting abnormal behaviors in transformer substation based on artificial intelligence in complex scene |
CN112132013A (en) * | 2020-09-22 | 2020-12-25 | 中国科学技术大学 | Vehicle key point detection method |
CN112132013B (en) * | 2020-09-22 | 2022-07-15 | 中国科学技术大学 | Vehicle key point detection method |
CN112418046A (en) * | 2020-11-17 | 2021-02-26 | 武汉云极智能科技有限公司 | Fitness guidance method, storage medium and system based on cloud robot |
CN112418046B (en) * | 2020-11-17 | 2023-06-23 | 武汉云极智能科技有限公司 | Exercise guiding method, storage medium and system based on cloud robot |
CN112529768B (en) * | 2020-12-04 | 2023-01-06 | 中山大学 | Garment editing and generating method based on generation countermeasure network |
CN112529768A (en) * | 2020-12-04 | 2021-03-19 | 中山大学 | Garment editing and generating method based on generation countermeasure network |
CN113610070A (en) * | 2021-10-11 | 2021-11-05 | 中国地质环境监测院(自然资源部地质灾害技术指导中心) | Landslide disaster identification method based on multi-source data fusion |
CN114428877A (en) * | 2022-01-27 | 2022-05-03 | 西南石油大学 | Intelligent clothing matching method and system |
CN114428877B (en) * | 2022-01-27 | 2023-09-15 | 西南石油大学 | Intelligent clothing matching method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109784350A (en) | In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network | |
CN110427867B (en) | Facial expression recognition method and system based on residual attention mechanism | |
CN109359538B (en) | Training method of convolutional neural network, gesture recognition method, device and equipment | |
CN113221639B (en) | Micro-expression recognition method for representative AU (AU) region extraction based on multi-task learning | |
Zhang et al. | Mask SSD: An effective single-stage approach to object instance segmentation | |
Arora et al. | Recognition of non-compound handwritten devnagari characters using a combination of mlp and minimum edit distance | |
CN112836597B (en) | Multi-hand gesture key point estimation method based on cascade parallel convolution neural network | |
CN107748858A (en) | A kind of multi-pose eye locating method based on concatenated convolutional neutral net | |
CN109598234A (en) | Critical point detection method and apparatus | |
Hsu et al. | Pedestrian detection using stationary wavelet dilated residual super-resolution | |
CN112069900A (en) | Bill character recognition method and system based on convolutional neural network | |
Ronchetti et al. | Handshape recognition for argentinian sign language using probsom | |
Xu et al. | Robust hand gesture recognition based on RGB-D Data for natural human–computer interaction | |
Sun et al. | Convolutional multi-directional recurrent network for offline handwritten text recognition | |
Mahdavi et al. | Visual parsing with query-driven global graph attention (QD-GGA): preliminary results for handwritten math formula recognition | |
Bengamra et al. | A comprehensive survey on object detection in Visual Art: taxonomy and challenge | |
CN111862031A (en) | Face synthetic image detection method and device, electronic equipment and storage medium | |
CN113780140B (en) | Gesture image segmentation and recognition method and device based on deep learning | |
Liu et al. | Double Mask R‐CNN for Pedestrian Detection in a Crowd | |
Dhore et al. | Human Pose Estimation And Classification: A Review | |
Vaishali | Real-time object detection system using caffe model | |
Zhao et al. | Micro-expression recognition based on pixel residual sum and cropped gaussian pyramid | |
AU2021104479A4 (en) | Text recognition method and system based on decoupled attention mechanism | |
CN114862716A (en) | Image enhancement method, device and equipment for face image and storage medium | |
Cao et al. | A method based on faster RCNN network for object detection |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190521 |