CN111709459A - Cloud platform-based machine vision algorithm training data management system and method - Google Patents
Cloud platform-based machine vision algorithm training data management system and method Download PDFInfo
- Publication number
- CN111709459A CN111709459A CN202010458094.2A CN202010458094A CN111709459A CN 111709459 A CN111709459 A CN 111709459A CN 202010458094 A CN202010458094 A CN 202010458094A CN 111709459 A CN111709459 A CN 111709459A
- Authority
- CN
- China
- Prior art keywords
- image
- video
- cloud
- file
- machine vision
- 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
- 238000012549 training Methods 0.000 title claims abstract description 36
- 238000000034 method Methods 0.000 title claims description 16
- 238000013523 data management Methods 0.000 title claims description 7
- 238000007726 management method Methods 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 14
- 230000009466 transformation Effects 0.000 claims description 23
- 238000006243 chemical reaction Methods 0.000 claims description 13
- 238000013135 deep learning Methods 0.000 claims description 9
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims description 6
- 238000013473 artificial intelligence Methods 0.000 description 16
- 238000007781 pre-processing Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 230000009467 reduction Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 230000003416 augmentation Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000001965 increasing effect Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- VYZAMTAEIAYCRO-UHFFFAOYSA-N Chromium Chemical compound [Cr] VYZAMTAEIAYCRO-UHFFFAOYSA-N 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- JIYMVSQRGZEYAX-CWUUNJJBSA-N gemifloxacin mesylate Chemical compound CS(O)(=O)=O.C1C(CN)C(=N/OC)/CN1C(C(=C1)F)=NC2=C1C(=O)C(C(O)=O)=CN2C1CC1 JIYMVSQRGZEYAX-CWUUNJJBSA-N 0.000 description 1
- 238000003707 image sharpening Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/40—Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor
-
- G06T3/02—
-
- G06T3/04—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4046—Scaling the whole image or part thereof using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/60—Rotation of a whole image or part thereof
-
- G06T5/73—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/94—Hardware or software architectures specially adapted for image or video understanding
- G06V10/95—Hardware or software architectures specially adapted for image or video understanding structured as a network, e.g. client-server architectures
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/40—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using video transcoding, i.e. partial or full decoding of a coded input stream followed by re-encoding of the decoded output stream
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Signal Processing (AREA)
- Human Computer Interaction (AREA)
- Software Systems (AREA)
- Computer Networks & Wireless Communication (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a management system of machine vision algorithm training data based on a cloud platform, which comprises the following steps: a file server; a database; the cloud server is simultaneously electrically connected with the file server and the database, can receive data transmitted by the file server and the database, and is used for providing an image video transcoding service based on cloud service; and the client service module is electrically connected with the cloud server, can receive data transmitted by the cloud server, is used for providing video cutting based on cloud service, and expands and strengthens the training sample through an intelligent image processing algorithm based on cloud. A cloud-based image video management system can be provided for storing and managing image video data. The invention also provides a management method of the machine vision algorithm training data based on the cloud platform.
Description
Technical Field
The invention relates to a management system and method for machine vision algorithm training data based on a cloud platform, and belongs to the field of artificial intelligence.
Background
With the development of artificial intelligence in recent years, machine vision algorithms based on deep learning are applied in various fields. Along with the algorithms, some open-source data sets also appear, but the data sets only cover a small part of the actual application field, when the artificial intelligence algorithm is used for solving specific industry problems, a large amount of sample data is still needed for training artificial intelligence, but the cost for acquiring data is very high, so that the popularization and the use of the artificial intelligence in various industries are prevented. Most of the existing machine vision algorithm training data processing tools are off-line application program development kits, developers need to write a large amount of codes irrelevant to artificial intelligence algorithms, and the tool kits cannot expand the diversity of data. OpenCV is an open-source cross-platform computer vision library, which contains many basic machine vision algorithms and can meet the basic requirements of AI algorithm training data preprocessing, such as a series of processing of amplifying, reducing, compressing, clipping, rotating and the like on training picture data, but OpenCV is a software development library. The developer is required to write the code himself and is offline.
Pictures and videos collected from various devices such as mobile phones, cameras, video cameras in reality include various formats such as picture format: jpg pictures, png pictures, tiff pictures, etc., video coding format: x265, x264, mp4, etc., but in the cloud, the formats supported by existing browsers are limited, for example, most browsers (such as google chrome or microsoft IE) cannot parse pictures in tiff format, or x265 coded video.
The existing Ariiyun artificial intelligence platform and the data preprocessing part related to the patent are not directed at machine vision firstly, so that related functions are few. At the same time, it does not include preprocessing of images and videos, mostly character data in the database, and their data format conversion is only to convert and merge data in the database.
The existing Huashi cloud AI platform has the following problems that firstly, machine vision is not aimed at, so that image preprocessing is lacked after picture data are uploaded, then videos cannot be uploaded, and meanwhile, the existing different data sets are not combined.
The prior art has the following problems: most of them lack the support for video files, and there are two kinds of raw data commonly used in the field of machine vision: images and videos, and many related platforms only support uploading images or image sequences. This is true for some algorithms that require a sequence of consecutive images, such as: target tracking, behavior recognition and the like, and if the video is not supported and the image sequence is used for replacement when data is uploaded, the storage cost can be increased by several times or even dozens of times.
The data preprocessing methods are not complete: for example, OpenCV, which is the most commonly used library, is very common in the field of machine vision, but some of the algorithms in the library are relatively old and do not support some of the latest artificial intelligence algorithms, such as image classification and image super-resolution restoration, which are used in training data preprocessing.
The development cost is high: the developers need to write codes by themselves when using OpenCV to perform training data preprocessing, the development cost is high, and the developers cannot be closely matched with the online cloud service, so that the additional overhead of part of the cloud service is increased. (As a simple example: for example, the original image resolution is 640x480, the data required for annotation is 1920x1080, if the algorithm for enlarging the image is developed by the user, the 1920x1080 image needs to be uploaded in the subsequent uploading process, which increases the uploading flow and development workload of the network)
The sample cannot be expanded: when artificial intelligence based on a supervision algorithm is used for solving specific problems, a large number of samples can not be collected frequently, and when insufficient samples are used for artificial intelligence algorithm training, the algorithm universality is not strong, the samples are expanded to a sufficient number, and the algorithm universality and accuracy can be improved.
Support for machine vision is not professional enough: such as an artificial intelligence cloud platform provided by the arri cloud, hua shi cloud, etc., they generally provide a general platform. The method can support various data such as voice, characters, images, texts and the like, but the support of the data on machine vision algorithms is not perfect, the preprocessing of image video data is little, and the integration function of labels with different formats is lacked.
Disclosure of Invention
The invention designs and develops a management system of machine vision algorithm training data based on a cloud platform, and can provide an image video management system based on a cloud for storing and managing image video data.
The invention also designs and develops a management method of machine vision algorithm training data based on the cloud platform, which can perform video cutting and transcoding service based on cloud service, and can expand and enhance a training sample based on the cloud intelligent image processing algorithm.
The technical scheme provided by the invention is as follows:
management system of machine vision algorithm training data based on cloud platform
A file server;
a database;
the cloud server is simultaneously electrically connected with the file server and the database, can receive data transmitted by the file server and the database, and is used for providing an image video transcoding service based on cloud service;
and the client service module is electrically connected with the cloud server, can receive data transmitted by the cloud server, is used for providing video cutting based on cloud service, and expands and strengthens the training sample through an intelligent image processing algorithm based on cloud.
Preferably, the cloud server includes: the system comprises an API interface, an image expansion module, an image enhancement module, a task queue module, an image transcoding module and a video transcoding module.
Preferably, the client service module includes: image video transcoding, video cutting, file uploading and downloading, file management, sample enhancement and expansion.
A management method of machine vision algorithm training data based on a cloud platform comprises the following steps:
firstly, creating a data set, and selecting a video or image file to be cut in a data set file by a user according to interest, and uploading the video or image file to a cloud end;
step two, when a user selects a video file and cuts an interested time period from the video file, determining an interested time period selection starting point time stamp and an interested time period selection ending point time stamp, and selecting an interested area in the interested time period;
step three, carrying out standardized configuration on the region of interest, and unifying the formats and resolutions of the images and videos in the data set;
and step four, performing sample data expansion configuration, determining a final task workflow, generating a task, sending the task to a message queue, and processing the task by a cloud server.
Preferably, the third step further comprises:
selecting a target format, and unifying file formats of uploaded images and videos;
and selecting a target resolution, and unifying the resolutions of the uploaded images and videos.
Preferably, the sample data extension configuration extension mode includes: and G, generating a GAN image based on deep learning and performing image basic transformation.
Preferably, the image basis transformation includes: and generating a new sample through image rotation transformation, mirror image transformation and affine transformation.
Preferably, the image and video format conversion bar performs image format conversion using an image map library.
The invention has the following beneficial effects:
1. high efficiency: the data preprocessing algorithm does not need a developer to write codes, and meanwhile, the cloud image and video transcoding service is provided. The efficiency of artificial intelligence algorithm developers is improved.
2. Improving the accuracy of the algorithm: and the accuracy of subsequent sample labeling and the accuracy of the finally obtained AI model are improved through an image enhancement algorithm and an image sample augmentation algorithm.
3. The economic efficiency is as follows: the data set is expanded through an artificial intelligence-based algorithm, and the cost for acquiring artificial intelligence training samples is reduced.
Drawings
Fig. 1 is an overall design diagram of a cloud platform-based machine vision algorithm training data management system according to the present invention.
FIG. 2 is a video cropping process layout according to the present invention.
FIG. 3 is a diagram illustrating the process of image enhancement and sample augmentation according to the present invention.
FIG. 4 is a flow chart of the cloud AI platform in the prior art
Detailed Description
The present invention is further described in detail below with reference to the attached drawings so that those skilled in the art can implement the invention by referring to the description text.
As shown in fig. 1 to 4, the present invention provides a cloud platform-based machine vision algorithm training data management system, including: file server, database, cloud server, client service and artificial intelligence researchers and developers.
The file server is arranged at the initial position of the cloud platform-based machine vision algorithm training data management system and used for storing large files, and the file server comprises: sample images and corresponding thumbnails, video files, user images; the database and the file server are arranged at the initial position of the cloud platform-based machine vision algorithm training data management system in parallel, and the database is used for storing other data except video images and comprises: user information, image metadata (such as image resolution, image storage address, image occupation space and the like), annotation data corresponding to the sample, and task information of an image processing task and a data format conversion task.
File server and database are connected with high in the clouds server electricity respectively to give the high in the clouds server with information data transmission, the high in the clouds server includes: the system comprises an API (application programming interface) module, an image expansion module, an image enhancement module, a task queue module and a video transcoding module, and is used for providing an image video transcoding service based on cloud service.
The API interface module is used for receiving a front end API interface and communicating with a database; the image expansion module expands the existing samples by using an image style migration and GAN image generation algorithm based on deep learning; the image enhancement module improves the image resolution through a super-resolution reduction algorithm based on deep learning; the task queue module is used for queuing the asynchronous tasks generated by each module; the video transcoding module is used for transcoding the video uploaded by the user, and as a preferable choice in the invention, the video uploaded by the user is transcoded into H264 and YUV420G formats and can be compatible with common browsers.
The client service module is connected with the high in the clouds server electricity, can receive the data that the high in the clouds server carried the transmission for provide the video based on cloud and cut, and through the intelligent image processing algorithm based on the cloud, expand and strengthen the training sample, include: image video transcoding, video cutting, file uploading and downloading, file management, sample enhancement and expansion.
The image video transcoding is to transcode an image and a video format which are not supported by a browser into a supported format through a cloud service; the video cutting is provided in the form of a browser plug-in and is used for cutting out interesting segments from the video of a user and uploading the segments; the file uploading and downloading are mainly used for uploading images, videos and labeled data and the downloading function of the files; the file management module is used for the operations of querying, editing, deleting and the like of videos, images and labeled data of a user and can also create a folder; and the sample enhancement and expansion expands and enhances the uploaded samples through an artificial intelligence algorithm.
The invention also provides a management method of machine vision algorithm training data based on the cloud platform, which comprises the following steps:
creating a data set, and selecting a video or image file to cut in the data set file according to the interest of a user and uploading the video or image file to a cloud end;
when a user selects a video file and cuts out an interested time period from the video file, determining an interested time period selection starting point timestamp and an interested time period selection ending point timestamp, and selecting an interested area in the interested time period;
carrying out standardized configuration on the region of interest, and unifying the formats and resolutions of the images and videos in the data set;
and performing sample data expansion configuration, determining a final task workflow, generating a task, sending the task to a message queue, and processing the task by a cloud server.
As shown in fig. 2, video cropping includes:
1) the user installs a video processing plug-in:
after the user finishes registration and login, a video processing plug-in is installed;
2) loading a video file:
dragging the video to a cutting plug-in unit on a video browser by a user to finish video loading;
3) selecting a video segment of interest:
and clicking a play button to search the interested video clip. Dragging a starting point pointer and an end point pointer on a progress bar of the video player to the starting point and the end point of the interesting segment respectively, and then selecting the next step to carry out configuration work;
4) selecting a rectangular region of interest:
on the basis of selecting the time period of the interested video, selecting the interested rectangular area of the video, firstly clicking the upper left corner of the interested rectangular area, and dragging the mouse to move to the lower right corner of the interested area. If this step is omitted, then the rectangular region of interest is defaulted
5) Previewing video clips
Browsing the video clips, and returning to the step 3) for re-selection if the video clips are not satisfactory;
6) and (3) executing cutting: and clicking a cutting button to start video cutting. And finishing cutting when the progress bar reaches 100%.
As shown in fig. 3, the image enhancement and sample augmentation tasks:
1) uploading data:
uploading a data set to be processed;
2) configuring an intelligent image enhancement task:
selecting an image enhancement algorithm to use, the options including:
1. and (3) an image super-resolution reduction algorithm based on deep learning. The algorithm can improve the resolution of the image; and enhance the details of the image. The algorithm needs to configure the width and height of a target image, and the unit is a pixel;
2. and (3) an image deblurring algorithm based on deep learning. The algorithm may increase the sharpness of the image. The accuracy of subsequent labeling work is improved;
3. and (4) an image sharpening algorithm. And enhancing the edge gradient information of the image. Making the target object in the image more prominent;
3) configuring an intelligent image expansion task:
one or more image transformation algorithms are selected. A new image is generated, thereby increasing the number of samples. Alternative transformation algorithms include:
1. generating a GAN image;
2. image magnification, reduction, rotation, mirroring, affine transformation, etc.;
4) configuring a target data set:
configuring the name, description and storage location of the merged data set;
5) and (3) starting a task:
clicking the start button starts the above task and when the progress bar reaches 100%, a new data set is generated.
The specific implementation process of the cloud-based video image training sample enhancement system comprises the following steps:
1. a user registration account logs in a cloud system, a data set is created, then a data set file uploading page is entered, and a local picture video file is selected and uploaded to a cloud;
2. and (3) video configuration, if the video file is uploaded in the first step, the user needs to make the following selections:
1) whether a time period of interest is cut from the video and, if so, first the timestamp of the start of the time period of interest is selected and then the timestamp of the end of the time period of interest is selected. Finally, clicking to determine;
2) whether rectangular frame clipping is performed on the video time period of interest, and the region of interest is selected on the basis of the time period of interest. Firstly, clicking the upper left corner of an interested rectangular region by using a mouse, then dragging the mouse to the lower right corner of the interested region, releasing the mouse, and clicking a determined button;
3. and (3) sample standardization configuration, wherein in the step, the formats and resolutions of the images and the books in the sample data set are unified, and a user needs to make the following selections:
1) whether formats of all uploaded sample files are unified or not, if so, a target format needs to be selected;
2) whether the resolutions of all uploaded sample files are unified or not, if so, a target resolution needs to be selected;
2) clicking to determine after the selection is finished;
4. and (4) sample data expansion configuration, wherein a user selects whether to expand the sample file. If the selection is yes, one of the following three expansion modes is selected:
2) generating a GAN image based on deep learning;
3) image basic transformation;
5. determining a final task workflow, browsing the summary configured in the steps in a page, clicking to confirm if the problem does not exist, generating a task, and sending the task to a message queue;
6. the cloud server processes the tasks, reads the tasks from the message queue and respectively performs the following processing according to the content of the tasks:
1) video time section clipping and region of interest clipping, and video format conversion:
step 1, analyzing the following parameters according to task configuration: the time period of interest start timestamp t1, end timestamp t2, and the start coordinate x, y of the region of interest with width and height: w, h;
step 2: clipping and format conversion of the video are accomplished using the open source library ffmpeg, assuming the input video is: input.avi, output video is output.mp4, using the following commands:
ffmpeg-I input.avi-ss t1-c copy-to t2 output.mp4 crop＝w＝100:h＝100:x＝12:y＝34-c:v libx264；
and step 3: mp4 is stored in a dataset;
2) image format conversion (this step is relatively simple)
Step 1: reading a format of image conversion in task configuration;
step 2: calling an image map library to complete image format conversion;
3) image resolution, video resolution conversion:
step 1: reading a resolution parameter in task configuration;
step 2: judging the relation between the target resolution Res1 and the original resolution Res 0:
1. if Res0> Res1, the picture resolution is reduced to Res1 using the nearest neighbor interpolation algorithm, and for each pixel (src _ x, src _ y) of the original image, the relationship between the pixel and the corresponding target pixel (dst _ x, dst _ y) is established:
src_x＝dst_x*(Ws/Wd)
src_y＝dst_y*(Hs/Hd)
wherein Ws and Hs are the width and height of the source map respectively. Wd and Hd are the width and height of the target picture, respectively.
2. If Res0 is less than Res1, the resolution of the image is improved to Res1 by using the image super-resolution reduction algorithm based on the neural network, and the target image x to be generateddIs compared with the original image xsThe super-resolution reduction algorithm f comprises the following steps:
xd＝f(xs)；
f＝convd1·convd2…convdn·convu1·convu2…convun；
wherein convd1…convdnEncoding neural networks for images, convu1…convunDecoding neural channels and collaterals for images, wherein the layer number of the neural network can be changed differently for the images with different resolutions;
3. no operation is performed if Res0 ═ Res 1;
3) image basic transformation: generating a new sample through image rotation transformation, mirror image transformation and affine transformation:
the method comprises the following steps: reading a conversion mode in task configuration;
step two: performing an actual transformation operation according to the transformation parameters, assuming that each pixel (Xs, Ys) of the original image, each pixel (Xd, Yd) of the target image:
1. and (3) image rotation transformation:
wherein θ is the angle of rotation;
2. image mirror turning:
wherein w is the width of the image;
3. affine transformation of an image:
wherein a is1a2b1b2c1c26 parameters for affine transformation;
4) GAN image generation based on deep learning, comprising:
step 1: reading parameters of GAN image generation in the task configuration.
Step 2: generating a target image by using a GAN neural network, wherein the target image is xdThe method comprises the following steps:
xd＝fGAN(xrand)，
wherein x israndFor a randomly generated 1024-dimensional initial vector,
fGAN＝fconve·factive
wherein f isconveIs a convolution function;
wherein f isactiveActivating a function for Sigmoid;
wherein z is fupThe output result of the function;
step three: storing the generated target graph in a data set;
5) the cloud server completes the task, updates the task state in the database, and sends a message to inform a front-end user.
While embodiments of the invention have been described above, it is not limited to the applications set forth in the description and the embodiments, which are fully applicable in various fields of endeavor to which the invention pertains, and further modifications may readily be made by those skilled in the art, it being understood that the invention is not limited to the details shown and described herein without departing from the general concept defined by the appended claims and their equivalents.
Claims (8)
1. A management system of machine vision algorithm training data based on a cloud platform comprises:
a file server;
a database;
the cloud server is simultaneously electrically connected with the file server and the database, can receive data transmitted by the file server and the database, and is used for providing an image video transcoding service based on cloud service;
and the client service module is electrically connected with the cloud server, can receive data transmitted by the cloud server, is used for providing video cutting based on cloud service, and expands and strengthens the training sample through an intelligent image processing algorithm based on cloud.
2. The system for managing cloud platform based machine vision algorithm training data of claim 1, wherein the cloud server comprises: the system comprises an API interface, an image expansion module, an image enhancement module, a task queue module, an image transcoding module and a video transcoding module.
3. The cloud platform based machine vision algorithm training data management system of claim 2, wherein the client service module comprises: image video transcoding, video cutting, file uploading and downloading, file management, sample enhancement and expansion.
4. A management method of machine vision algorithm training data based on a cloud platform comprises the following steps:
creating a data set, and selecting a video or image file to cut in the data set file according to the interest of a user and uploading the video or image file to a cloud end;
when a user selects a video file and cuts out an interested time period from the video file, determining an interested time period selection starting point timestamp and an interested time period selection ending point timestamp, and selecting an interested area in the interested time period;
carrying out standardized configuration on the region of interest, and unifying the formats and resolutions of the images and videos in the data set;
and performing sample data expansion configuration, determining a final task workflow, generating a task, sending the task to a message queue, and processing the task by a cloud server.
5. The method for managing cloud platform based machine vision algorithm training data of claim 4, wherein the standardized configuration further comprises:
selecting a target format, and unifying file formats of uploaded images and videos;
and selecting a target resolution, and unifying the resolutions of the uploaded images and videos.
6. The method for managing cloud platform based machine vision algorithm training data according to claim 5, wherein the expansion manner of the sample data expansion configuration includes: and G, generating a GAN image based on deep learning and performing image basic transformation.
7. The method for managing cloud platform based machine vision algorithm training data of claim 6, wherein the image basis transformation comprises: and generating a new sample through image rotation transformation, mirror image transformation and affine transformation.
8. The method for managing cloud platform based machine vision algorithm training data as claimed in claim 7, wherein the image and video format conversion bar performs image format conversion using an image map library.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010458094.2A CN111709459A (en) | 2020-05-27 | 2020-05-27 | Cloud platform-based machine vision algorithm training data management system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010458094.2A CN111709459A (en) | 2020-05-27 | 2020-05-27 | Cloud platform-based machine vision algorithm training data management system and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111709459A true CN111709459A (en) | 2020-09-25 |
Family
ID=72538551
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010458094.2A Pending CN111709459A (en) | 2020-05-27 | 2020-05-27 | Cloud platform-based machine vision algorithm training data management system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111709459A (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103974146A (en) * | 2014-05-30 | 2014-08-06 | 深圳市同洲电子股份有限公司 | Video processing method, client and system |
| WO2016207899A1 (en) * | 2015-06-25 | 2016-12-29 | Capester Ltd | System and method for secured capturing and authenticating of video clips |
| CN106804002A (en) * | 2017-02-14 | 2017-06-06 | 北京时间股份有限公司 | A kind of processing system for video and method |
| WO2018005701A1 (en) * | 2016-06-29 | 2018-01-04 | Cellular South, Inc. Dba C Spire Wireless | Video to data |
| CN109101878A (en) * | 2018-07-01 | 2018-12-28 | 浙江工业大学 | A kind of image analysis system and image analysis method for the estimation of stalk combustion value |
| CN109635156A (en) * | 2018-12-17 | 2019-04-16 | 台州三石量子科技有限公司 | Intelligent image processing system |
| WO2020018819A1 (en) * | 2018-07-18 | 2020-01-23 | Nvidia Corporation | Virtualized computing platform for inferencing, advanced processing, and machine learning applications |
| CN111008688A (en) * | 2018-10-04 | 2020-04-14 | 国际商业机器公司 | Neural network using in-loop data augmentation during network training |
-
2020
- 2020-05-27 CN CN202010458094.2A patent/CN111709459A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103974146A (en) * | 2014-05-30 | 2014-08-06 | 深圳市同洲电子股份有限公司 | Video processing method, client and system |
| WO2016207899A1 (en) * | 2015-06-25 | 2016-12-29 | Capester Ltd | System and method for secured capturing and authenticating of video clips |
| WO2018005701A1 (en) * | 2016-06-29 | 2018-01-04 | Cellular South, Inc. Dba C Spire Wireless | Video to data |
| CN106804002A (en) * | 2017-02-14 | 2017-06-06 | 北京时间股份有限公司 | A kind of processing system for video and method |
| CN109101878A (en) * | 2018-07-01 | 2018-12-28 | 浙江工业大学 | A kind of image analysis system and image analysis method for the estimation of stalk combustion value |
| WO2020018819A1 (en) * | 2018-07-18 | 2020-01-23 | Nvidia Corporation | Virtualized computing platform for inferencing, advanced processing, and machine learning applications |
| CN111008688A (en) * | 2018-10-04 | 2020-04-14 | 国际商业机器公司 | Neural network using in-loop data augmentation during network training |
| CN109635156A (en) * | 2018-12-17 | 2019-04-16 | 台州三石量子科技有限公司 | Intelligent image processing system |
Non-Patent Citations (2)
| Title |
|---|
| CLAIRE MCQUIN等: "CellProfiler 3.0: Next-generation image processing for biology" * |
| 周礼: "面向移动视频的云编辑系统研究与实现" * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11004120B2 (en) | Method for providing real-time service of huge and high quality digital image on internet | |
| Villán | Mastering OpenCV 4 with Python: a practical guide covering topics from image processing, augmented reality to deep learning with OpenCV 4 and Python 3.7 | |
| US10318575B2 (en) | Systems and methods of building and using an image catalog | |
| CN110555795A (en) | High resolution style migration | |
| KR20190107069A (en) | Method and apparatus for image recognition based on augmented reality | |
| CN116954450A (en) | Screenshot method and device for front-end webpage, storage medium and terminal | |
| CN110633733A (en) | Intelligent image matching method and device and computer readable storage medium | |
| JP6838167B2 (en) | Web page main image recognition method and device | |
| CN111709459A (en) | Cloud platform-based machine vision algorithm training data management system and method | |
| CN108536769B (en) | Image analysis method, search method and device, computer device and storage medium | |
| JP2013008142A (en) | Image processing device, image processing method and image processing program | |
| CN113392250B (en) | Vector diagram retrieval method and system based on deep learning | |
| CN113902684A (en) | Image segmentation method and device, electronic device and storage medium | |
| JP4336813B2 (en) | Image description system and method | |
| US11914681B2 (en) | Determining and selecting operation features for digital content editing operations within an operation sequence | |
| CN114758339B (en) | Method and device for acquiring character recognition model, computer equipment and storage medium | |
| CN116680434B (en) | Image retrieval method, device, equipment and storage medium based on artificial intelligence | |
| CN116150417B (en) | Multi-scale multi-fusion image retrieval method and device | |
| KR102303335B1 (en) | Method for generating presentation content using data binding, and apparatus using said method, and system for providing presentation content using data binding | |
| KR20020046444A (en) | Sequence Diagram Generating Tool and Method for Analyzing Team-Developing Program | |
| KR20230027930A (en) | Method for generating deep learning model using feature transfer and loss transfer | |
| CN114416088A (en) | Mobile tool page generation method and device, computer equipment and storage medium | |
| CN117421604A (en) | Digital twin scene representation method and device | |
| CN115082627A (en) | 3D stage model generation method and device, electronic equipment and readable storage medium | |
| CN115619891A (en) | Method and system for generating split-mirror script |
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 |