CN107247944B - Face detection speed optimization method and device based on deep learning - Google Patents
Face detection speed optimization method and device based on deep learning Download PDFInfo
- Publication number
- CN107247944B CN107247944B CN201710508939.2A CN201710508939A CN107247944B CN 107247944 B CN107247944 B CN 107247944B CN 201710508939 A CN201710508939 A CN 201710508939A CN 107247944 B CN107247944 B CN 107247944B
- Authority
- CN
- China
- Prior art keywords
- data
- deep learning
- target frame
- fixed
- point data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 30
- 238000013135 deep learning Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000005457 optimization Methods 0.000 title claims abstract description 16
- 238000013136 deep learning model Methods 0.000 claims abstract description 44
- 238000012937 correction Methods 0.000 claims abstract description 21
- 238000012545 processing Methods 0.000 claims abstract description 10
- 230000004807 localization Effects 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 abstract description 3
- 238000007792 addition Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 230000006872 improvement Effects 0.000 description 3
- 238000003909 pattern recognition Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 101100285899 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) SSE2 gene Proteins 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000012567 pattern recognition method Methods 0.000 description 1
- 230000008093 supporting effect Effects 0.000 description 1
- 238000002054 transplantation Methods 0.000 description 1
Images
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
- G06V40/161—Detection; Localisation; Normalisation
Abstract
The invention discloses a face detection speed optimization method and a face detection speed optimization device based on deep learning, wherein the method comprises the following steps: floating point data in the deep learning model are fixed-point; the data in the face image is fixed-point; performing parallel operation on the fixed-point data in the deep learning model and the fixed-point data in the face image; repeating the step of parallel operation until the whole deep learning model is operated and outputting the coordinate information of the human face target frame and the correction information of the human face target frame; restoring the correction information of the face target frame into corresponding floating point data; and combining the coordinate information of the face target frame with the restored correction information of the face target frame, adjusting the face target frame, and finally obtaining the position of the face real target frame. The invention can save the data occupation space, and the fixed-point operation can improve the data processing efficiency; the parallel operation can reduce the total operation times, improve the efficiency, save the time and achieve the effects of reducing the hardware cost and improving the human face detection speed.
Description
Technical Field
The invention relates to the field of face detection, in particular to a face detection speed optimization method and device based on deep learning.
Background
The face recognition detection research mainly comprises the research of a face detection technology and a face recognition technology. The face recognition detection means that for any given image, a certain strategy is adopted to search the image to determine whether the image contains a face, if so, the position, size and posture of the face are returned, and then the face is recognized. It is a complex and challenging pattern detection problem. Meanwhile, the face detection needs to be put to practical application, the accuracy and the speed are two key problems which need to be solved urgently, the accuracy of the face detection is greatly improved since the 90 s of the 20 th century, but the speed is not up to the satisfaction degree of an application system user, and therefore, the research staff pay hard efforts.
Deep learning is one of the most important breakthroughs that the field of artificial intelligence has taken in the last decade. It has achieved great success in many fields such as speech recognition, natural language processing, computer vision, image and video analysis, multimedia, etc.
The most difference between deep learning and traditional pattern recognition method is that the features adopted by the deep learning method are obtained by automatic learning from big data, and are not designed by hand. Good features may improve the performance of the pattern recognition system. Over the past several decades, hand-designed features have been dominating in various applications of pattern recognition. The manual design mainly depends on the prior knowledge of a designer, and the advantage of big data is difficult to utilize. The number of parameters allowed to occur in the design of a feature is quite limited due to the reliance on manual tuning of the parameters. Deep learning can automatically learn representations of features from large data, which can contain thousands of parameters.
Manually designed effective features often take five to ten years, while deep learning can quickly learn new effective feature representations from training data for new applications.
A pattern recognition system includes two parts, a feature and a classifier. In the conventional approach, the optimization of features and classifiers is separated. In the neural network framework, the feature representation and the classifier are jointly optimized, and the performance of joint cooperation of the feature representation and the classifier can be exerted to the maximum extent.
The deep learning adopts a layered structure similar to that of the traditional neural network, the system is a multilayer network consisting of an input layer, a hidden layer (multilayer) and an output layer, only the nodes of adjacent layers are connected, the nodes of the same layer and cross-layer are not connected, and each layer can be regarded as a logistic regression model; the layered structure is relatively close to the structure of the human brain.
The deep learning has a large calculation amount and complex calculation, so that the face detection speed based on the deep learning has certain disadvantages compared with the traditional method. For the optimization of the deep learning face detection speed, many schemes exist, such as changing hardware conditions, using a high-performance hardware GPU, and adopting rapid acceleration software ATBLAS, Openblas, NNPACK and the like. These existing technical solutions all require hardware to have corresponding support units, and if there is no corresponding support unit, there will be a corresponding bottleneck in speed increase; in addition, on low-end CPUs, the supporting effect is not ideal.
Disclosure of Invention
The invention provides a face detection speed optimization method and device based on deep learning, and aims to solve the technical problems that the existing scheme is high in hardware cost and cannot realize speed improvement on a low-end CPU.
The technical scheme adopted by the invention is as follows:
according to an aspect of the present invention, a method for optimizing a face detection speed based on deep learning is provided, the method comprising: floating point data in the deep learning model are fixed-point; the data in the face image is fixed-point; performing parallel operation on the fixed-point data in the deep learning model and the fixed-point data in the face image; repeating the step of parallel operation until the whole deep learning model is operated and outputting the coordinate information of the human face target frame and the correction information of the human face target frame; restoring the correction information of the face target frame into corresponding floating point data; and combining the coordinate information of the face target frame with the restored correction information of the face target frame, adjusting the face target frame, and finally obtaining the position of the face real target frame.
Further, the step of fixing the floating point data in the deep learning model comprises the following steps: reading a deep learning model; converting floating point data in the deep learning model into first fixed point data; the first anchor data is stored as first short type data.
Optionally, the data range of the first fixed point data is-255 to + 255.
Further, the step of spotting the data of the face image comprises: reading a face image; preprocessing a face image; converting the preprocessed data into second fixed point data; the second fixed-point data is stored as second short type data.
Optionally, the data range of the second fixed-point data is-255 to + 255.
Further, the step of parallel operation comprises: loading the first short type data and the second short type data; and carrying out parallel multiply-add operation on the loaded first short type data and the loaded second short type data.
According to another aspect of the present invention, there is also provided a face detection speed optimization apparatus based on deep learning, the apparatus including: the first localization module is used for localizing floating point data in the deep learning model; the second fixed-point module is used for fixing the data of the preprocessed face image; the parallel operation module is used for repeatedly performing parallel operation on the fixed-point data in the deep learning model and the fixed-point data in the face image until the whole deep learning model is operated and outputting face target frame coordinate information and face target frame correction information; the restoring module is used for restoring the correction information of the face target frame into corresponding floating point data; and the adjusting module is used for combining the coordinate information of the face target frame with the restored face target frame correction information to adjust the face target frame and finally obtain the position of the real face target frame.
Further, the first localization module includes: the first reading submodule is used for reading the deep learning model; the first conversion submodule is used for converting floating point data in the deep learning model into first fixed point data; and the first storage submodule is used for storing the first fixed point data as the first short type data.
Further, the second spotting module comprises: the second reading submodule is used for reading the face image; the preprocessing submodule is used for preprocessing the face image read by the second reading module; the second conversion submodule is used for converting the preprocessed data into second fixed point data; and the second storage submodule is used for storing the second fixed-point data as second short type data.
Further, the parallel operation module comprises: the loading sub-module is used for loading the first short type data and the second short type data; and the parallel multiplication and addition sub-module is used for carrying out parallel multiplication and addition operation on the loaded first short type data and the loaded second short type data.
According to the face detection optimization method and device based on deep learning, floating point data in a deep learning model and a face image are fixed in point, parallel operation is conducted on the fixed point data, the occupied space of the data can be saved, the situation that a large number of operation resources and operation time are consumed in the floating point data operation process is avoided, the efficiency of processing the data can be improved through operation on the fixed point data, multiple operations on multiple data can be executed through parallel operation processing, the overall operation times are reduced, the operation efficiency is improved, the time is saved, the effects of reducing hardware cost and improving the face detection speed are achieved, and the face detection speed can be improved on a low-end CPU.
In addition to the objects, features and advantages described above, other objects, features and advantages of the present invention are also provided. The present invention will be described in further detail below with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a flow chart of a face detection speed optimization method based on deep learning according to a preferred embodiment of the present invention;
FIG. 2 is a detailed flow chart of the present invention for pinning floating point data in the deep learning model;
FIG. 3 is a detailed flow chart of the preferred embodiment of the present invention for spotting data of face images;
fig. 4 is a schematic block diagram of a face detection speed optimization device based on deep learning according to a preferred embodiment of the present invention.
Detailed Description
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present invention will be described in detail below with reference to the embodiments with reference to the attached drawings.
The invention is mainly based on the network structure obtained by deep learning training and the data obtained by training, and the invention process will be elaborated below.
Referring to fig. 1, a preferred embodiment of the present invention provides a method for optimizing a face detection speed based on deep learning, including:
s100, floating point data in the deep learning model are fixed in point;
step S200, data in the face image is fixed in point;
step S300, performing parallel operation on the fixed-point data in the deep learning model and the fixed-point data in the face image;
repeating the step S300 of parallel operation until the whole deep learning model is operated and outputting the coordinate information of the human face target frame and the correction information of the human face target frame;
s400, restoring the correction information of the face target frame into corresponding floating point data;
and step S500, combining the coordinate information of the face target frame with the restored correction information of the face target frame, adjusting the face target frame, and finally obtaining the position of the face real target frame.
Further, referring to fig. 2, the step S100 of spotting the floating point data in the deep learning model includes:
step S110, reading a deep learning model;
in the preferred embodiment, the deep learning model file is read by mainly adopting the cross-platform computer vision library opencv based on the open source style, the deep learning framework cafe and software depended by the cafe.
Step S120, floating point data in the deep learning model is converted into first floating point data; in the preferred embodiment, the floating point data is fixed-point-255 to +255, that is, the data range of the first fixed point data is-255 to + 255.
Step S130, storing the first fixed point data as first short type data. After the deep learning model file is subjected to floating point fixed point conversion, in order to avoid that the result data after the following parallel operation (multiplication, addition and accumulation) is out of bounds, the method saves the fixed-point data into short type fixed point data, and because the short type fixed point data occupies 16 bytes of data space and occupies half less space than floating point (float) data, the space can be saved in storage; meanwhile, when data is loaded in the subsequent parallel operation step, more data can be loaded at one time.
Further, referring to fig. 3, the step S200 of spotting the data of the face image includes:
step S210, reading a face image;
step S220, preprocessing the face image;
step S230, converting the preprocessed data into second fixed point data; also optionally, the data range of the second fixed-point data is-255 to + 255.
Step S240, the second fixed-point data is stored as the second short type data. In the same way as step S130, this step can save the storage space and facilitate subsequent efficient data reading, which is beneficial to subsequent efficient parallel operation processing.
Further, the step S300 of parallel operation includes:
step S310, loading the first short type data and the second short type data;
in the preferred embodiment, a CPU computing unit is used to load data, 256 bytes of data are loaded at a time (256 bytes of data may represent 8 short type data), and the 8 first short type data after the deep learning model is fixed and the 8 second short type data after the face image is fixed are loaded.
And step S320, performing parallel multiply-add operation on the loaded first short type data and the loaded second short type data.
In the preferred embodiment, the loaded data is subjected to parallel multiply-add operation by using the CPU data unit AVX2 parallel multiply-add operation logic unit, such as the function _ mm256_ hadd _ epi32(), so that multiple multiplications and multiple additions can be completed once, and from the operation process, one operation can represent the previous multiple operations, so that the processing benefits also save the operation time and improve the working efficiency. Of course, in other embodiments, other CPU data units, such as SSE, SSE2, AVX, etc., may be used to perform the parallel operations. The operation mode of performing multiple operations on one instruction at a time can increase cache hit and reduce memory access.
In consideration of hardware cost, the GPU and the FPGA are limited in application scenarios due to excessively high price and power consumption, and are not basically used in common consumer-grade products. The method of the invention can reduce the hardware condition, realize the deep learning on the common consumption-level product, and is beneficial to realizing the face detection speed improvement on the low-end CPU.
According to another aspect of the present invention, there is also provided a face detection speed optimization apparatus based on deep learning, referring to fig. 4, the apparatus including:
a first stationing module 100, configured to statione floating point data in the deep learning model;
the second stationing module 200 is configured to perform stationing on the preprocessed data of the face image;
the parallel operation module 300 is configured to repeatedly perform parallel operation on the fixed-point data in the deep learning model and the fixed-point data in the face image until the whole deep learning model is operated and output face target frame coordinate information and face target frame correction information;
the restoring module 400 is configured to restore the face target frame correction information to corresponding floating point data;
the adjusting module 500 is configured to combine the coordinate information of the face target frame with the restored face target frame correction information to adjust the face target frame, and finally obtain a real face target frame position.
Further, the first localization module 100 includes:
a first reading sub-module 110 for reading the deep learning model;
the first conversion submodule 120 is configured to convert floating point data in the deep learning model into first fixed point data;
a first storage sub-module 130 for storing the first anchor data as first short type data.
Further, the second spotting module 200 includes:
a second reading sub-module 210 for reading the face image;
the preprocessing submodule 220 is configured to preprocess the face image read by the second reading module;
a second conversion sub-module 230, configured to convert the preprocessed data into second fixed-point data;
a second storage sub-module 240 for storing the second fixed-point data as second short type data.
Further, the parallel operation module 300 includes:
a load sub-module 310 for loading the first short type data and the second short type data;
and the parallel multiplication and addition sub-module 320 is used for performing parallel multiplication and addition operation on the loaded first short type data and the loaded second short type data.
The method and the device have the advantage of high detection speed, can reduce the storage space of hardware, and lay a foundation for the transplantation optimization of later-stage hardware platforms (ARM, DSP and the like). According to the invention, floating point data in the deep learning model and the face image are fixed in point, and then parallel operation is performed on the fixed point data, so that the occupied space of the data can be saved, a large amount of operation resources and operation time are prevented from being consumed in the operation process of the floating point data, the efficiency of processing the data can be improved by the operation on the fixed point data, and multiple operations of multiple data can be executed once by the parallel operation processing, so that the overall operation times are reduced, the operation efficiency is improved, the time is saved, and the effects of reducing hardware cost and improving the face detection speed are achieved. Meanwhile, the data after fixed-point processing is stored as short type data, and the result data after parallel operation can be prevented from crossing the boundary. The parallel operation mode of performing a plurality of operations one instruction at a time can increase cache hit and reduce memory access.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (2)
1. A face detection speed optimization method based on deep learning is characterized by comprising the following steps:
floating point data in the deep learning model are fixed-point;
the data in the face image is fixed-point;
performing parallel operation on the fixed-point data in the deep learning model and the fixed-point data in the face image;
repeating the step of parallel operation until the whole deep learning model is operated and outputting the coordinate information of the human face target frame and the correction information of the human face target frame;
restoring the human face target frame correction information into corresponding floating point data;
combining the coordinate information of the face target frame with the restored correction information of the face target frame, and adjusting the face target frame to finally obtain the position of the real face target frame;
the step of fixing the floating point data in the deep learning model comprises the following steps:
reading the deep learning model by adopting a cross-platform computer vision library opencv based on an open-source hairstyle, a deep learning framework cafe and software depended by the cafe;
converting floating point data in the deep learning model into first fixed point data, wherein the data range of the first fixed point data is-255 to + 255;
storing the first fixed point data as first short type data;
the step of stationing the data of the face image comprises the following steps:
reading the face image;
preprocessing the face image;
converting the preprocessed data into second fixed point data, wherein the data range of the second fixed point data is-255 to + 255;
storing the second fixed-point data as second short type data;
the step of parallel operation comprises:
loading the first short type data and the second short type data, loading the data by using a CPU (central processing unit) computing unit, loading 256 bytes of data at a time, and completely loading 8 first short type data after the localization of the deep learning model and 8 second short type data after the localization of the face image;
and performing parallel multiply-add operation on the loaded first short type data and the loaded second short type data by utilizing a CPU data unit AVX2 parallel multiply-add operation logic unit.
2. A face detection speed optimization device based on deep learning is characterized by comprising:
the first localization module is used for localizing floating point data in the deep learning model;
the second fixed-point module is used for fixing the data of the preprocessed face image;
the parallel operation module is used for repeatedly performing parallel operation on the fixed-point data in the deep learning model and the fixed-point data in the face image until the whole deep learning model is operated and outputting face target frame coordinate information and face target frame correction information;
the restoring module is used for restoring the human face target frame correction information into corresponding floating point data;
the adjusting module is used for combining the coordinate information of the face target frame with the restored face target frame correction information to adjust the face target frame and finally obtain the position of the real face target frame;
the first localization module comprises:
the first reading submodule is used for reading the deep learning model by adopting a cross-platform computer vision library opencv based on an open source style, a deep learning framework caffe and software depended by the caffe;
the first conversion submodule is used for converting floating point data in the deep learning model into first fixed point data, and the data range of the first fixed point data is-255 to + 255;
the first storage submodule is used for storing the first fixed point data into first short type data;
the second spotting module comprises:
the second reading submodule is used for reading the face image;
the preprocessing submodule is used for preprocessing the face image read by the second reading module;
the second conversion submodule is used for converting the preprocessed data into second fixed point data, and the data range of the second fixed point data is-255 to + 255;
a second storage submodule for storing the second fixed-point data as second short type data;
the parallel operation module comprises:
the loading sub-module is used for loading the first short type data and the second short type data, loading the data by using a CPU (central processing unit) computing unit, loading 256 bytes of data at a time, and completely loading 8 first short type data after the localization of the deep learning model and 8 second short type data after the localization of the face image;
and the parallel multiply-add sub-module is used for performing parallel multiply-add operation on the loaded first short type data and the loaded second short type data by utilizing a CPU data unit AVX2 parallel multiply-add operation logic unit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710508939.2A CN107247944B (en) | 2017-06-28 | 2017-06-28 | Face detection speed optimization method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710508939.2A CN107247944B (en) | 2017-06-28 | 2017-06-28 | Face detection speed optimization method and device based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107247944A CN107247944A (en) | 2017-10-13 |
| CN107247944B true CN107247944B (en) | 2020-11-10 |
Family
ID=60013649
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710508939.2A Active CN107247944B (en) | 2017-06-28 | 2017-06-28 | Face detection speed optimization method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107247944B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP7188237B2 (en) * | 2019-03-29 | 2022-12-13 | 富士通株式会社 | Information processing device, information processing method, information processing program |
| CN113298244B (en) * | 2021-04-21 | 2023-11-24 | 上海安路信息科技股份有限公司 | Neural network post-processing implementation method, device, terminal and medium in target detection |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106485230A (en) * | 2016-10-18 | 2017-03-08 | 中国科学院重庆绿色智能技术研究院 | Based on the training of the Face datection model of neutral net, method for detecting human face and system |
| CN106502626A (en) * | 2016-11-03 | 2017-03-15 | 北京百度网讯科技有限公司 | Data processing method and device |
| CN106570559A (en) * | 2015-10-09 | 2017-04-19 | 阿里巴巴集团控股有限公司 | Data processing method and device based on neural network |
| CN106874883A (en) * | 2017-02-27 | 2017-06-20 | 中国石油大学(华东) | A kind of real-time face detection method and system based on deep learning |
-
2017
- 2017-06-28 CN CN201710508939.2A patent/CN107247944B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106570559A (en) * | 2015-10-09 | 2017-04-19 | 阿里巴巴集团控股有限公司 | Data processing method and device based on neural network |
| CN106485230A (en) * | 2016-10-18 | 2017-03-08 | 中国科学院重庆绿色智能技术研究院 | Based on the training of the Face datection model of neutral net, method for detecting human face and system |
| CN106502626A (en) * | 2016-11-03 | 2017-03-15 | 北京百度网讯科技有限公司 | Data processing method and device |
| CN106874883A (en) * | 2017-02-27 | 2017-06-20 | 中国石油大学(华东) | A kind of real-time face detection method and system based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107247944A (en) | 2017-10-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110546611B (en) | Reducing power consumption in a neural network processor by skipping processing operations | |
| WO2020073211A1 (en) | Operation accelerator, processing method, and related device | |
| CN107895191B (en) | Information processing method and related product | |
| US11307864B2 (en) | Data processing apparatus and method | |
| US20200104573A1 (en) | Data processing apparatus and method | |
| US20180260710A1 (en) | Calculating device and method for a sparsely connected artificial neural network | |
| WO2021022521A1 (en) | Method for processing data, and method and device for training neural network model | |
| CN111047602A (en) | Image segmentation method and device and terminal equipment | |
| Sheng et al. | Improving video temporal consistency via broad learning system | |
| US20220067354A1 (en) | Dynamic gesture recognition method, device and computer-readable storage medium | |
| CN108171328A (en) | A kind of convolution algorithm method and the neural network processor based on this method | |
| CN107247944B (en) | Face detection speed optimization method and device based on deep learning | |
| Zhang et al. | Summary of convolutional neural network compression technology | |
| CN114925320B (en) | Data processing method and related device | |
| CN110968235B (en) | Signal processing device and related product | |
| CN115221102A (en) | Method for optimizing convolution operation of system on chip and related product | |
| CN111652349A (en) | Neural network processing method and related equipment | |
| CN117011118A (en) | Model parameter updating method, device, computer equipment and storage medium | |
| WO2021238734A1 (en) | Method for training neural network, and related device | |
| CN116246110A (en) | Image classification method based on improved capsule network | |
| US20230196093A1 (en) | Neural network processing | |
| CN110069647A (en) | Image tag denoising method, device, equipment and computer readable storage medium | |
| He et al. | ECS-SC: Long-tailed classification via data augmentation based on easily confused sample selection and combination | |
| US11086634B2 (en) | Data processing apparatus and method | |
| CN114519425A (en) | Convolution neural network acceleration system with expandable scale |
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 | ||
| CB02 | Change of applicant information |
Address after: 100097 Beijing Haidian District Kunming Hunan Road 51 C block two floor 207. Applicant after: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD. Address before: 100193 4, 403, block A, 14 building, 10 East North Road, Haidian District, Beijing. Applicant before: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD. |
|
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 410205 14 Changsha Zhongdian Software Park Phase I, 39 Jianshan Road, Changsha High-tech Development Zone, Yuelu District, Changsha City, Hunan Province Applicant after: Wisdom Eye Technology Co.,Ltd. Address before: 100097 2nd Floor 207, Block C, 51 Hunan Road, Kunming, Haidian District, Beijing Applicant before: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD. |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: Optimization method and device for face detection speed based on depth learning Effective date of registration: 20221205 Granted publication date: 20201110 Pledgee: Agricultural Bank of China Limited Hunan Xiangjiang New Area Branch Pledgor: Wisdom Eye Technology Co.,Ltd. Registration number: Y2022430000107 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20231220 Granted publication date: 20201110 Pledgee: Agricultural Bank of China Limited Hunan Xiangjiang New Area Branch Pledgor: Wisdom Eye Technology Co.,Ltd. Registration number: Y2022430000107 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| CP03 | Change of name, title or address |
Address after: No. 205, Building B1, Huigu Science and Technology Industrial Park, No. 336 Bachelor Road, Bachelor Street, Yuelu District, Changsha City, Hunan Province, 410000 Patentee after: Wisdom Eye Technology Co.,Ltd. Address before: 410205 building 14, phase I, Changsha Zhongdian Software Park, No. 39, Jianshan Road, Changsha high tech Development Zone, Yuelu District, Changsha City, Hunan Province Patentee before: Wisdom Eye Technology Co.,Ltd. |
|
| CP03 | Change of name, title or address |