CN109165554A - A kind of face characteristic comparison method based on cuda technology - Google Patents
A kind of face characteristic comparison method based on cuda technology Download PDFInfo
- Publication number
- CN109165554A CN109165554A CN201810816840.3A CN201810816840A CN109165554A CN 109165554 A CN109165554 A CN 109165554A CN 201810816840 A CN201810816840 A CN 201810816840A CN 109165554 A CN109165554 A CN 109165554A
- Authority
- CN
- China
- Prior art keywords
- face characteristic
- request
- aspect ratio
- thread
- cuda
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to field of biological recognition, more particularly to a kind of face characteristic comparison method based on cuda technology, this method comprises: being primarily based on the hardware architecture and memory Accessing Mechanism of cuda, by way of modifying thread accesses global memory, by one characteristic block of previous thread accesses become a byte inside one characteristic block of a thread accesses, the merging access of global memory is achieved, thread is reduced to the access times of global memory, so that improving single face characteristic compares speed;Furthermore each face characteristic is compared request to be cached, all comparisons request is combined by scheduled rule and carries out aspect ratio pair;Comparison result is finally split into result independent according to pre-defined rule and is reported to user, to improve the concurrent efficiency of face characteristic comparison.Under conditions of identical face characteristic storage capacity and request Concurrency number, this programme can effectively reduce server resource, save hardware cost.
Description
Technical field
The present invention relates to technical field of biometric identification more particularly to a kind of face characteristic based on cuda technology to compare other side
Method.
Background technique
Currently, the biological identification technology based on face characteristic mainly completes people based on the effective ratio pair of face characteristic
The identification process of face.By in present image face carry out feature extraction after with the magnanimity people in preset face characteristic library
Face feature is compared one by one, is obtained alignment similarity score respectively, then similarity score value is ranked up in descending order, most
A series of highest faces of similarity that reservation meets threshold value afterwards are exported as a result.
Development like a raging fire is built in the safe city of current National various regions, also produces magnanimity by pretreated face
Characteristic, therefore effective face characteristic comparison method is used, public security investigator can be helped quickly to identify and distinguish specific
The true identity of personnel, to extend efficient help and solve for public security video investigation, security administration, the criminal investigation work such as put on record
Method.
Prior art 1 is used based on CPU (Graphic Processing Unit) processor as hardware carrier people
Face characteristic value compares.This method is mainly that the comparison of face characteristic is completed by multithreading, specifically: 1, preset
Face characteristic library is evenly distributed to different threads;2, current feature and the feature database of all threads are compared one by one,
And export a series of highest comparison results of similarity for meeting threshold value;3, summarize the output of each thread as a result, and carrying out descending
Sequence, final output meet a series of highest comparison results of similarity of threshold value.The program uses the hardware frame of CPU processor
Structure, the limited speed of aspect ratio pair is in the limitation of CPU frequency, bus bandwidth and internal storage access speed first.Secondly face characteristic
Concurrently comparison speed it is directly proportional to feature quantity, tend to linear increase.Therefore, allow in face characteristic storage capacity and user maximum
Comparison time-consuming it is all fixed under conditions of, if it is desired to improving the number of concurrent that user carries out face characteristic comparison, then need to increase ratio
Pair server carry out lateral dilatation, increase hardware cost.
Prior art 2 is used based on GPU as the comparison of hardware carrier face characteristic value.This method mainly passes through GPU
Powerful floating-point operation ability and high performance parallel computing complete the comparison of face characteristic, specifically: 1, by writing
The kernel function of GPU equipment end completes the aspect ratio pair in default face characteristic library;2,1 thread completes the comparison of 1 face characteristic,
Ultra-large parallel computation;3, GPU overall situation video memory is written in all comparison results;4, the comparison result of GPU is copied to
CPU, and descending sort is carried out, output meets a series of highest comparison results of similarity of threshold value eventually;5, each aspect ratio pair
Request is serial to be executed.In scheme 2, the comparison of 1 face characteristic of single GPU thread completion first causes GPU memory not merge
Access, not no being optimal of global memory's bandwidth cause global memory to access slack-off;Secondly with as prior art 1,
All it is to execute aspect ratio using serial order to task, causes to compare concurrent inefficient.
Summary of the invention
The purpose of the present invention is to provide a kind of face characteristic comparison methods based on cuda technology, to solve the prior art
Middle face characteristic comparison is concurrent inefficient, and the server for needing to increase comparison carries out lateral dilatation, and hardware cost is caused to improve
The problem of.
The invention is realized by the following technical scheme:
A kind of face characteristic comparison method based on cuda technology, comprising: in advance add all face characteristics of object library
Be downloaded to GPU video memory, and guarantee memory continuously be aligned, by GPU thread accesses feature memory, realize the merging access of memory;
Under the premise of hardware architecture and concurrent technology based on cuda, by the aspect ratio of merging user to request, by multiple spies
Sign compares request and is merged into a kernel function, finally the separation of feature comparison result is carried out according still further to default rule, so that special
Sign comparison result matches with request.
The merging access method of memory specifically comprises the following steps:
A, effective face characteristic is obtained;
B, per thread only calculates the characteristic for meeting serial number condition;Serial number condition=start sequence number+the number of step-length * times,
Wherein, start sequence number is the respective initial value of thread, and step-length is the Thread Count inside thread block (Block), and number is initial value 0
From number is increased, the maximum value of number is equal to characteristic length divided by the Thread Count of thread block (Block);
C, phase knowledge and magnanimity are calculated, memory is write back;
D, whether terminate, if the determination result is YES, then terminate if judging to compare;Otherwise return step a.
Wherein, the calculation of the face characteristic is that a thread block runs a face characteristic value.
Preferably, each aspect ratio caches request according to the first pre-defined rule.
The aspect ratio specifically comprises the following steps: the caching method of request
A1, aspect ratio is received to request, cached according to first pre-defined rule;
B1, feature comparison result is waited to return;
C1, aspect ratio pair is reported as a result, terminating.
Further comprise obtaining the aspect ratio of certain amount caching to request, then presses the aspect ratio of caching to request
It is merged according to the second pre-defined rule.
It further comprise after merging treatment, carrying out aspect ratio pair, then carrying out aspect ratio according to task unique identifier SN
Result is separated.
Preferably, it then follows the hardware architecture and memory Accessing Mechanism of cuda, it is special using GPU thread connected reference face
The internal storage data of sign realizes the merging access of global characteristics memory, reduces internal storage access number, improves aspect ratio to speed.
Preferably, multiple aspect ratio is merged into primary request to request, so that reducing GPU compares kernel function function
Call number improves the number of concurrent of aspect ratio pair.
It applies the technical scheme of the present invention, is primarily based on the hardware architecture and memory Accessing Mechanism of cuda, passes through
Modify thread accesses global memory mode, by one characteristic block of previous thread accesses become one spy of a thread accesses
A byte inside block is levied, the merging access of global memory is achieved, reduces thread to the access times of global memory, from
And it improves single face characteristic and compares speed.Furthermore each face characteristic is compared request to be cached, passes through scheduled rule
Then all comparisons request is combined and carries out aspect ratio pair, comparison result is finally split into respective independence according to pre-defined rule
Result be reported to user, thus improve face characteristic comparison concurrent efficiency.
Technical solution of the present invention uses the hardware architecture of cuda, and parallel processing capability and floating-point operation ability are bright
The powerful and influential hardware configuration for being better than CPU, improves face characteristic comparison efficiency.By changing the calculation of face characteristic, by it
One face characteristic of previous thread operation becomes a thread block (Block) and runs a feature, realizes the conjunction of GPU memory
And access, it compares speed and promotes 1 times or so.And the method for comparing request merging treatment is increased, multiple kernel function tune
With becoming once to call, to promote the number of concurrent of face characteristic comparison.In identical face characteristic storage capacity and request Concurrency number
Under conditions of, this programme can effectively reduce server resource, save hardware cost.
Detailed description of the invention
The present invention is described in further details below with reference to attached drawing;
Fig. 1 is the flow chart of the merging access method of GPU global memory of the invention;
Fig. 2 is the flow chart that face characteristic of the invention compares request caching;
Fig. 3 is that face characteristic of the invention compares the combined data structure diagram of request;
Fig. 4 is that face characteristic of the invention compares the combined flow chart of request.
Specific embodiment
Present invention is further described in detail combined with specific embodiments below and referring to attached drawing.
This programme needs are loaded into all face characteristics in face characteristic library in the global memory of GPU in advance, and guarantee
Memory block is continuous, alignment.
Hardware architecture and memory based on cuda (Compute United Device Architecture) are visited
It asks mechanism, in such a way that a thread block calculates a feature, solves the merging access of GPU global memory, reduce thread pair
The access times of global memory make global memory's being optimal of bandwidth.To improve the comparison speed that single feature compares request
Degree, specific steps are as shown in Figure 1, comprising:
Step a, validity feature is obtained;Validity feature is the face characteristic for referring to effectively detect and recognize;
Step b, per thread (Thread) only calculates the characteristic for meeting serial number condition (tid serial number), serial number condition
=start sequence number+the number of step-length * times.Wherein, start sequence number is the respective initial value of thread, and step-length is inside thread block (Block)
Thread Count;Number is initial value 0 from number is increased, and the maximum value of number is equal to characteristic length divided by thread block (Block)
Thread Count;Characteristic length is the numerical value length of face characteristic.
Step c, phase knowledge and magnanimity are calculated, memory is write back;
Step d, whether terminate, if the determination result is YES, then terminate this process if judging to compare;Otherwise return step a.
Since the access speed of global memory is slow, 400~600 clock cycle are needed, therefore optimize global memory
Access speed just seem increasingly important.In addition to guaranteeing that the memory address of face feature database is continuous and is aligned it in this programme
Outside, also one feature is calculated by a thread and becomes thread block one spy of calculating by modification face characteristic calculation
Sign ensure that per thread in access global memory is one-to-one continuous alignment access, therefore the access address of per thread can
To be combined, reach the number for reducing internal storage access, to improve single aspect ratio to the speed of request.Pass through this plan
Slightly, under the face characteristic storage capacity of 100w, single face characteristic compares speed and improves nearly 1 times.
Concurrent technology based on cuda, by carried out in a kernel function multiple characteristic values comparison we obtain execution when
Between as shown in table 1.With the increase for comparing characteristic value, effect of optimization is further obvious.
Table 1
Compare characteristic value number | Time-consuming (ms) is called in nonjoinder | Merge and calls time-consuming (ms) |
1 | 31 | 31 |
2 | 60 | 46 |
10 | 929 | 494 |
12 | 1213 | 605 |
Based on the test data of table 1, request is compared using the face characteristic for merging single, calls face characteristic ratio multiple
Core (kernal) function is become once to call, to improve the number of concurrent of face characteristic comparison.The realization packet of the technical program
Include following 2 sub-processes.
1, caching process of the aspect ratio to request
After receiving the request of aspect ratio pair, unique identifier (SN) is assigned to each request automatically and is cached in order
Get up, finally obstruction waits aspect ratio to the task completion notice of thread.The comparison of thread is completed to lead to when receiving aspect ratio
Know and then comparison result is reported to user, to complete this aspect ratio to request.Specific steps are as shown in Fig. 2, packet
It includes:
Step a1, aspect ratio is received to request, according to rule cache;
Step b1, comparison result is waited to return;
Step c1, comparison result is reported, this process is terminated.
2, merge aspect ratio to request process
A certain number of face characteristics are obtained from buffer queue and compare request, according to the unique identification SN of Fig. 3, characteristic
It carries out face characteristic according to, the data structure that successively forms of channel information and timestamp and compares the merging of task, and set merging and appoint
The upper limit of business.Then it calls the kernel function of aspect ratio pair and carries out task requests according to the unique identification SN of request task and compare
As a result matching, finally notice waits this aspect ratio of thread to complete request, and specific steps are as shown in Figure 4, comprising:
Step a2, the aspect ratio of certain amount caching is obtained to task;
Step b2, it is merged according to pre-defined rule;
Step c2, aspect ratio pair is carried out;
Step d2, characteristic results separation is carried out according to task SN;
Step e2, notice waits thread comparison to terminate.
Using the hardware architecture of cuda, the parallel processing capability and floating-point operation ability of this programme be obviously eager to excel with
The hardware configuration of CPU, for example under conditions of 100w face characteristic storage capacity, retrieval is carried out under cpu processor and needs time-consuming
113ms, and retrieved under the hardware based on cuda, need time-consuming 31ms.Obviously, this programme aspect ratio is higher than efficiency existing
There is technical solution 1.
Compared with prior art 2, change the calculation of face characteristic, by one spy of previous thread operation
Sign becomes a thread block and runs a feature, realizes the merging access of GPU memory, compares speed and promote 1 times or so.And increase
The method for comparing request merging treatment is added, multiple kernel function function call is become once to call, to promote face spy
Levy the number of concurrent compared.Obviously, the technical program can more save server resource under conditions of identical storage capacity and number of concurrent,
Save hardware cost.
Particular embodiments described above has carried out further in detail the purpose of the present invention, technical scheme and beneficial effects
Describe in detail it is bright, it should be understood that the above is only a specific embodiment of the present invention, the guarantor being not intended to limit the present invention
Protect range.Without departing from the spirit and scope of the invention, any modification, equivalent substitution, improvement and etc. done also belong to this
Within the protection scope of invention.
Claims (10)
1. a kind of face characteristic comparison method based on cuda technology characterized by comprising in advance by the owner of object library
Face feature is loaded into GPU video memory, and guarantee memory continuously be aligned, by GPU thread accesses feature memory, realize the conjunction of memory
And it accesses;It, will by the aspect ratio of merging user to request under the premise of hardware architecture and concurrent technology based on cuda
Multiple aspect ratios are merged into a kernel function request, finally carry out the separation of feature comparison result according still further to default rule,
So that feature comparison result matches with request.
2. the face characteristic comparison method according to claim 1 based on cuda technology, which is characterized in that the merging of memory
Access method specifically comprises the following steps:
A, effective face characteristic is obtained;
B, per thread only calculates the characteristic for meeting serial number condition;
C, phase knowledge and magnanimity are calculated, memory is write back;
D, whether terminate, if the determination result is YES, then terminate if judging to compare;Otherwise return step a.
3. the face characteristic comparison method according to claim 2 based on cuda technology, which is characterized in that the face
The calculation of feature is that a thread block runs a face characteristic value.
4. the face characteristic comparison method according to claim 2 based on cuda technology, which is characterized in that the serial number
Condition=start sequence number+the number of step-length * times;Wherein, start sequence number is the respective initial value of thread, and step-length is the line inside thread block
Number of passes, number are initial value 0 from integer is increased, and the maximum value of number is equal to characteristic length divided by the Thread Count of thread block.
5. the face characteristic comparison method according to claim 1 based on cuda technology, which is characterized in that by each feature
Request is compared to be cached according to the first pre-defined rule.
6. the face characteristic comparison method according to claim 5 based on cuda technology, which is characterized in that the feature
The caching method for comparing request specifically comprises the following steps:
A1, aspect ratio is received to request, cached according to first pre-defined rule;
B1, feature comparison result is waited to return;
C1, aspect ratio pair is reported as a result, terminating.
7. the face characteristic comparison method according to claim 1 based on cuda technology, which is characterized in that follow cuda's
Hardware architecture and memory Accessing Mechanism realize global characteristics using the internal storage data of GPU thread connected reference face characteristic
The merging of memory accesses, and reduces internal storage access number, improves aspect ratio to speed.
8. the face characteristic comparison method according to claim 1 based on cuda technology, which is characterized in that by multiple spy
Sign compares request and is merged into primary request, to reduce the call number that GPU compares kernel function, improves the concurrent of aspect ratio pair
Number.
9. the face characteristic comparison method according to claim 5 based on cuda technology, which is characterized in that further packet
Include, obtain certain amount caching aspect ratio to request, then by the aspect ratio of caching to request according to the second pre-defined rule into
Row merges.
10. the face characteristic comparison method according to claim 9 based on cuda technology, which is characterized in that further packet
It includes, after merging treatment, carries out aspect ratio pair, then carry out the separation of feature comparison result according to task SN.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810816840.3A CN109165554B (en) | 2018-07-24 | 2018-07-24 | Human face feature comparison method based on cuda technology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810816840.3A CN109165554B (en) | 2018-07-24 | 2018-07-24 | Human face feature comparison method based on cuda technology |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109165554A true CN109165554A (en) | 2019-01-08 |
CN109165554B CN109165554B (en) | 2021-09-24 |
Family
ID=64898241
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810816840.3A Active CN109165554B (en) | 2018-07-24 | 2018-07-24 | Human face feature comparison method based on cuda technology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109165554B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111368020A (en) * | 2020-02-10 | 2020-07-03 | 浙江大华技术股份有限公司 | Feature vector comparison method and device and storage medium |
CN113326714A (en) * | 2020-02-28 | 2021-08-31 | 杭州海康威视数字技术股份有限公司 | Target comparison method and device, electronic equipment and readable storage medium |
WO2023216444A1 (en) * | 2022-05-10 | 2023-11-16 | 上海登临科技有限公司 | Processor, multi-thread merging method and electronic device |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102521581A (en) * | 2011-12-22 | 2012-06-27 | 刘翔 | Parallel face recognition method with biological characteristics and local image characteristics |
CN104063714A (en) * | 2014-07-20 | 2014-09-24 | 詹曙 | Fast human face recognition algorithm used for video monitoring and based on CUDA parallel computing and sparse representing |
US20160110590A1 (en) * | 2014-10-15 | 2016-04-21 | University Of Seoul Industry Cooperation Foundation | Facial identification method, facial identification apparatus and computer program for executing the method |
CN106228628A (en) * | 2016-07-15 | 2016-12-14 | 腾讯科技(深圳)有限公司 | System, the method and apparatus of registering based on recognition of face |
KR20170137273A (en) * | 2016-06-02 | 2017-12-13 | 중앙대학교 산학협력단 | Apparatus and Method for Pedestrian Detection using Deformable Part Model |
-
2018
- 2018-07-24 CN CN201810816840.3A patent/CN109165554B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102521581A (en) * | 2011-12-22 | 2012-06-27 | 刘翔 | Parallel face recognition method with biological characteristics and local image characteristics |
CN104063714A (en) * | 2014-07-20 | 2014-09-24 | 詹曙 | Fast human face recognition algorithm used for video monitoring and based on CUDA parallel computing and sparse representing |
US20160110590A1 (en) * | 2014-10-15 | 2016-04-21 | University Of Seoul Industry Cooperation Foundation | Facial identification method, facial identification apparatus and computer program for executing the method |
KR20170137273A (en) * | 2016-06-02 | 2017-12-13 | 중앙대학교 산학협력단 | Apparatus and Method for Pedestrian Detection using Deformable Part Model |
CN106228628A (en) * | 2016-07-15 | 2016-12-14 | 腾讯科技(深圳)有限公司 | System, the method and apparatus of registering based on recognition of face |
Non-Patent Citations (3)
Title |
---|
LI CHAO SUN,ET AL: "《Acceleration algrithm for CUDA-based face detection》", 《ICSPCC 2013》 * |
WEIXIN_30951231: "《hystrix源码之请求合并》", 《HTTPS://BLOG.CSDN.NET/WEIXIN_30951231/ARTICLE/DETAILS/98317126》 * |
武泗海: "《CUDA全局内存-对齐与合并》", 《HTTPS://BLOG.CSDN.NET/QQ_17239003/ARTICLE/DETAILS/79038102》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111368020A (en) * | 2020-02-10 | 2020-07-03 | 浙江大华技术股份有限公司 | Feature vector comparison method and device and storage medium |
CN113326714A (en) * | 2020-02-28 | 2021-08-31 | 杭州海康威视数字技术股份有限公司 | Target comparison method and device, electronic equipment and readable storage medium |
CN113326714B (en) * | 2020-02-28 | 2024-03-22 | 杭州海康威视数字技术股份有限公司 | Target comparison method, target comparison device, electronic equipment and readable storage medium |
WO2023216444A1 (en) * | 2022-05-10 | 2023-11-16 | 上海登临科技有限公司 | Processor, multi-thread merging method and electronic device |
Also Published As
Publication number | Publication date |
---|---|
CN109165554B (en) | 2021-09-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US9298768B2 (en) | System and method for the parallel execution of database queries over CPUs and multi core processors | |
CN109165554A (en) | A kind of face characteristic comparison method based on cuda technology | |
CN110909025A (en) | Database query method, query device and terminal | |
CN111913955A (en) | Data sorting processing device, method and storage medium | |
KR20040084893A (en) | Multi-core multi-thread processor | |
CN110889754B (en) | Method for improving processing efficiency of non-overdraft hot spot account | |
EP2880566A1 (en) | A method for pre-processing and processing query operation on multiple data chunk on vector enabled architecture | |
WO2022218248A1 (en) | User portrait data processing method, apparatus, electronic device, and readable storage medium | |
US20220358178A1 (en) | Data query method, electronic device, and storage medium | |
US20210233027A1 (en) | Method for conducting statistics on insurance type state information of policy, terminal device and storage medium | |
CN108829740A (en) | Date storage method and device | |
CN112181948A (en) | Database operation statement processing method and device, electronic equipment and medium | |
CN111949681A (en) | Data aggregation processing device and method and storage medium | |
CN112783620A (en) | Optimization method for timing task scheduling, computer equipment and storage medium | |
CN108932258A (en) | Data directory processing method and processing device | |
CN116097222A (en) | Memory arrangement optimization method and device | |
CN111339210A (en) | Data clustering method and device | |
CN115065366A (en) | Compression method, device and equipment of time sequence data and storage medium | |
CN114925067A (en) | Data processing method, device, equipment and storage medium | |
CN114372071A (en) | Table data deleting method and device, computer equipment and storage medium | |
CN108255825A (en) | For dynamically distributing the method and apparatus of database connection | |
US7120749B2 (en) | Cache mechanism | |
CN117149781B (en) | Group-associative self-adaptive expansion cache architecture and access processing method thereof | |
CN116700999B (en) | Data processing method, device, computer equipment and storage medium | |
CN109710884B (en) | Real-time index configuration method and system supporting multiple complex calculation modes |
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 |