CN112825144A - Picture labeling method and device, electronic equipment and storage medium - Google Patents
Picture labeling method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN112825144A CN112825144A CN201911141366.XA CN201911141366A CN112825144A CN 112825144 A CN112825144 A CN 112825144A CN 201911141366 A CN201911141366 A CN 201911141366A CN 112825144 A CN112825144 A CN 112825144A
- Authority
- CN
- China
- Prior art keywords
- model
- labeling
- picture data
- training
- data set
- 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
- 238000002372 labelling Methods 0.000 title claims abstract description 146
- 238000000034 method Methods 0.000 claims abstract description 40
- 238000007635 classification algorithm Methods 0.000 claims abstract description 26
- 230000006870 function Effects 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 12
- 238000010586 diagram Methods 0.000 description 10
- 241000220225 Malus Species 0.000 description 7
- 235000013399 edible fruits Nutrition 0.000 description 7
- 235000011430 Malus pumila Nutrition 0.000 description 6
- 235000015103 Malus silvestris Nutrition 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 6
- 241000234295 Musa Species 0.000 description 4
- 235000018290 Musa x paradisiaca Nutrition 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 244000294611 Punica granatum Species 0.000 description 2
- 235000014360 Punica granatum Nutrition 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000002349 favourable effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 241000249058 Anthracothorax Species 0.000 description 1
- 235000004936 Bromus mango Nutrition 0.000 description 1
- 241000196324 Embryophyta Species 0.000 description 1
- 235000014826 Mangifera indica Nutrition 0.000 description 1
- 241001465754 Metazoa Species 0.000 description 1
- 241000220324 Pyrus Species 0.000 description 1
- 235000009184 Spondias indica Nutrition 0.000 description 1
- 235000021016 apples Nutrition 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 235000021017 pears Nutrition 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
- 235000013311 vegetables Nutrition 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
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Abstract
The invention relates to the technical field of picture processing, and provides a picture marking method, a picture marking device, electronic equipment and a storage medium, wherein the method comprises the following steps: classifying and labeling the obtained pictures according to a classification algorithm model to obtain different types of picture data sets; training a model of the same type as the picture data set by using the picture data set to obtain a model threshold value of the model; and if the model threshold value of the model does not reach the preset threshold value, creating a task interface corresponding to the model, selecting a new picture data set for retraining, and stopping training the model reaching the preset threshold value when the model threshold value of the model reaches the preset threshold value. The invention can reduce the labor cost and improve the efficiency of marking the picture.
Description
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to an image labeling method and apparatus, an electronic device, and a storage medium.
Background
In the prior art, before information in an image is identified or model training is performed based on a plurality of pictures, the image identification technology often needs to label the information to be identified, and needs a large amount of manpower to label, and because the labeling through the manpower has large subjectivity, the result is generally greatly deviated, so that the labeling quality is difficult to guarantee, and a large amount of resources are needed for the training of the early period of the manpower memorability. Therefore, in the prior art, the problems of high manual labeling cost and low efficiency exist in the process of labeling the pictures.
Disclosure of Invention
The embodiment of the invention provides a method and a device for marking pictures, electronic equipment and a storage medium, which can reduce labor cost and improve the efficiency of marking pictures.
In a first aspect, an embodiment of the present invention provides a method for labeling a picture, including the following steps:
classifying and labeling the obtained pictures according to a classification algorithm model to obtain different types of picture data sets;
training a model of the same type as the picture data set by using the picture data set to obtain a model threshold value of the model;
and if the model threshold value of the model does not reach the preset threshold value, creating a task interface corresponding to the model, selecting a new picture data set for retraining, and stopping training the model reaching the preset threshold value when the model threshold value of the model reaches the preset threshold value.
In a second aspect, an embodiment of the present invention further provides an apparatus for labeling a picture, including:
the training module is used for training the image data set to a model with the same type as the image data set to obtain a model threshold value of the model;
and the creating module is used for creating a task interface corresponding to the model to select a new picture data set for retraining if the model threshold of the model does not reach the preset threshold, and stopping the training of the model reaching the preset threshold until the model threshold of the model reaches the preset threshold.
In a third aspect, an embodiment of the present invention further provides an electronic device, including: the image annotation method comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to realize the steps in the image annotation method provided by the embodiment.
In a fourth aspect, a computer-readable storage medium has a computer program stored thereon, and the computer program, when executed by a processor, implements the steps in the method for labeling pictures provided in the foregoing embodiments.
In the embodiment of the invention, the obtained pictures are classified and labeled according to a classification algorithm model to obtain different types of picture data sets; training a model of the same type as the picture data set by using the picture data set to obtain a model threshold value of the model; and if the model threshold value of the model does not reach the preset threshold value, creating a task interface corresponding to the model, selecting a new picture data set for retraining, and stopping training the model reaching the preset threshold value when the model threshold value of the model reaches the preset threshold value. According to the embodiment of the invention, the pictures are classified and labeled through the classification algorithm to obtain different types of picture data sets, the models with the same types as the picture data sets are trained, the model threshold corresponding to the trained models is compared with the preset threshold, and the new picture data sets are selected to retrain the substandard models through newly establishing the task interfaces, so that the efficiency of labeling the pictures by the models can be improved and the manual use cost is reduced through continuously optimizing the models.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of a method for annotating a picture according to an embodiment of the present invention;
FIG. 2 is a flowchart of another method for annotating pictures according to an embodiment of the present invention;
FIG. 3 is a flowchart of another method for annotating pictures according to an embodiment of the present invention;
FIG. 4 is a flowchart of another method for annotating pictures according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a picture labeling apparatus according to an embodiment of the present invention;
FIG. 6 is a schematic structural diagram of an apparatus for labeling another picture according to an embodiment of the present invention;
FIG. 7 is a schematic structural diagram of an apparatus for labeling another picture according to an embodiment of the present invention;
FIG. 8 is a schematic structural diagram of an apparatus for labeling another picture according to an embodiment of the present invention;
fig. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The terms "including" and "having," and any variations thereof, in the description and claims of this application and the description of the above figures are intended to cover non-exclusive inclusions. The terms "first," "second," and the like in the description and claims of this application or in the above-described drawings are used for distinguishing between different objects and not for describing a particular order. Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
As shown in fig. 1, fig. 1 is a flowchart of a method for tagging a picture according to an embodiment of the present invention, where the method for tagging a picture includes the following steps:
s101, classifying and labeling the obtained pictures according to a classification algorithm model to obtain different types of picture data sets.
In this embodiment, the electronic device on which the image annotation method is executed may obtain the image and the like through a wired connection manner or a wireless connection manner. It should be noted that the Wireless connection manner may include, but is not limited to, a 3G/4G connection, a WiFi (Wireless-Fidelity) connection, a bluetooth connection, a wimax (worldwide Interoperability for Microwave access) connection, a Zigbee (low power local area network protocol), a uwb (ultra wideband) connection, and other Wireless connection manners known now or developed in the future.
It should be noted that the electronic device may be various electronic devices having a display screen and supporting web browsing, including but not limited to a smart phone, a tablet computer, a laptop portable computer, a desktop computer, and the like.
The acquired picture may be a picture pre-stored in a picture database, an online picture may be actively acquired through a network, or a picture manually input, and the like. The acquired picture may be an unlabeled picture. The classification algorithm model can be used for performing multi-level label classification and labeling on a single picture, for example, if the picture is displayed as a banana, the first-level label of the picture is fruit and the second-level label of the picture is a banana during classification and labeling. It may also be a single label classification labeling for multiple pictures, for example: the multiple pictures are respectively displayed as banana, apple, pomegranate and orange, and the corresponding single label is banana, apple, pomegranate and orange. The labeling may be performed by a designated labeling person, or may be performed by an Active learning (Active learning) algorithm of a labeling system, where the Active learning algorithm is an algorithm that has been packaged by a third party.
The acquired pictures can include a plurality of pictures with small differences, which cause difficulty in distinguishing by a labeling system for manual or active learning, and the pictures can be labeled for many times. The system can also comprise a plurality of pictures with large difference and easy recognition, and for the pictures with large difference and easy recognition, the marking system can automatically mark after active learning.
The picture data set may include a plurality of pictures with finished classification, that is, pictures with different types, for example, a fruit is stored as a type, a vegetable is stored as a type, and a living article is stored as a type.
S102, training the picture data set on the model with the same type as the picture data set to obtain the model threshold of the model.
The model may also be referred to as a training model, and training the different types of picture data sets may be to input the different types of picture data sets into training models having the same type respectively for training, so that each model obtained by training can have high recognition capability for one type of picture. The training model can be a model which is preliminarily established by training in advance according to a small amount of manually labeled picture data sets and optimized after one or more times of training, and model thresholds corresponding to different models can be obtained according to the labeling condition of the input picture data sets.
The model threshold is a threshold obtained by training the image data set input to the training model. The above-mentioned picture data set is used for the model training can be that let the training model carry out automatic marking to the picture in the picture data set of input, when the training model can both carry out correct marking to the picture of certain quantity in the picture data set, alright in order to output corresponding model threshold value, for example: for 100 pictures, if the training model can be actively labeled with 98 pictures, a model threshold value of 0.98 is output, and if the training model can be actively labeled with 60 pictures, a model threshold value of 0.6 is output. The higher the output model threshold value is, the better the optimization degree of the training model can be represented, and the labeling efficiency is higher. Of course, the maximum value of the model threshold may also be set, for example: the maximum value is 1. The model threshold may also be a percentage or other expression, and the functional relationship between the model threshold and the labeled number may be artificially defined, which is not limited in the embodiment of the present invention.
The image data sets comprise various types of data sets, and when training is performed, the various types of data sets can be selected, and the selected data sets are loaded into corresponding training models for training.
As a feasible embodiment, different training models can be set for different types of picture data sets, and the picture data sets of the types corresponding to the training models are input into the training models for training, so that the training speed can be increased, and the accuracy of training results can be improved. Of course, all the picture data sets can be input into the same training model for training, and the process of setting a plurality of training models can be omitted. Whereas the above-mentioned picture data sets for different categories may be identified by means of a classifier.
S103, if the model threshold value of the model does not reach the preset threshold value, a task interface corresponding to the model is created, a new picture data set is selected for retraining, and when the model threshold value of the model reaches the preset threshold value, the training of the model reaching the preset threshold value is stopped.
The preset threshold value can be a threshold value set manually after the active learning of the annotation system. The size of the set threshold can be adjusted according to the number of the picture data sets, and one threshold does not need to be used for all models. If the model threshold obtained after the corresponding picture data set is trained is compared with the preset threshold, and the model threshold obtained by training does not reach the preset threshold, it can be shown that the training model has not reached the optimization yet, and the training needs to be continued.
By comparing the model threshold obtained by training the picture data set with the preset threshold, the method can label pictures based on an active learning algorithm, and screens out pictures which cannot automatically identify labels and appoint manual labeling based on created task interfaces. The manual labeling can be labeling by professional labeling personnel. The artificially labeled picture can be used as a new picture data set of the corresponding training model again, and the training model is trained for multiple times until the obtained model threshold value of the corresponding training model reaches a preset threshold value, so that the training model reaches a relative optimal state.
In the embodiment of the invention, the obtained pictures are classified and labeled according to a classification algorithm model to obtain different types of picture data sets; applying different types of picture data sets to corresponding model training to obtain a model threshold value of the model; and if the model threshold value of the model does not reach the preset threshold value, creating a task interface to select a new picture data set for retraining, and stopping training the model reaching the preset threshold value when the model threshold value of the model reaches the preset threshold value. According to the embodiment of the invention, the pictures are classified and labeled through the classification algorithm to obtain different types of picture data sets, the models with the same types as the picture data sets are trained, the model threshold corresponding to the trained models is compared with the preset threshold, and the new picture data sets are selected to retrain the substandard models through newly establishing the task interfaces, so that the efficiency of labeling the pictures by the models can be improved and the manual use cost is reduced through continuously optimizing the models.
As shown in fig. 2, fig. 2 is a flowchart of another method for annotating pictures according to the present invention.
The method specifically comprises the following steps:
s201, classifying and labeling the obtained pictures according to the classification algorithm model to obtain different types of picture data sets.
Optionally, the step S201 includes:
and acquiring a single label classification labeling task.
The single-label classification labeling task can be a task generated by manually inputting an instruction, and the single-label classification labeling task can be a task created according to different types of pictures and used for labeling the pictures. In the task of single label classification and labeling, besides the single label classification and labeling of the multiple pictures, the task may further include attached labeling information related to the multiple pictures, for example: task name, annotation demanders, deadline, annotation type, data set, annotation team, task expression, annotation attribute setting and the like.
And performing single-label classification labeling on the multiple acquired pictures according to the single-label classification labeling task and the classification algorithm model to obtain multiple single-label pictures, wherein each single-label picture corresponds to one label type.
The above single-label classification labeling of multiple acquired pictures may refer to labeling multiple pictures at the same time, and each obtained single-label picture has only one label. The single-label classification labeling of the multiple pictures can be actively performed after active learning through a labeling system, and can also be manually performed. The single label may be a first-level label, a second-level label, a third-level label, or the like, and the previous-level label may include a corresponding next-level label, for example: the picture shows for the apple, then marks the primary label for fruit, perhaps can mark the secondary label, and the secondary label is the apple, and wherein, the primary label fruit includes the secondary label fruit.
And merging the single-label pictures of the same label type into the same picture data set to obtain different types of picture data sets.
The picture data set may include a label type of the picture, a number corresponding to the label type of the picture, and the like, for example, 03-fruit identification. The above-mentioned label attribute setting may include a plurality of options, one or more options may be arbitrarily selected, or a full-selection option may be selected. The selection items may be corresponding tags, for example: the selection items comprise fruits, the pull-down selection items of the fruits comprise apples, pears, mangos and the like, and the selection items can also comprise animals, plants, airplanes, motorcycles, ages and the like. The contents of the selection items are not limited in the present invention.
The merging may be to merge and store the pictures with the same type of the single labels, for example: the single labels of the multiple pictures are all apple pictures. Categorised mark can effectually distinguish the picture of different grade type, concentrates the picture of same type, when training in inputing the model that corresponds, can let the model discern the picture of the different forms of same type fast, improves the rate of accuracy of discernment, is favorable to when using the training model, can mark more pictures fast, reduces artifical threshold and the cost of marking.
Optionally, the step S201 includes:
and acquiring a multi-label classification labeling task.
The multi-label classification labeling task may include labeling a plurality of labels on the same picture, for example: the picture is a bicycle, then the tag may be a type or signal of a vehicle, a ride, a bicycle, even a bicycle, etc.
Of course, when labeling a picture, different types of objects may appear on the same picture, for example: the objects shown on the picture comprise an apple and a plate. Therefore, when the marking is carried out manually, the marking can be carried out according to the size of the space occupied by the object, the object occupying large space is preferably used as a main body for marking, pictures with different types of objects on one picture can be discarded, confusion is avoided, training difficulty is increased, and the pictures needing to be marked can be based on some clear and representative pictures.
If the marking system actively marks after actively learning, the size of the space occupied by the objects in the picture can be judged to mark by identifying the picture, and certainly, the picture can be marked in multiple ways. If the marking system cannot identify the pictures after active learning, the pictures can be sent to professional marking personnel for marking through creating a task window, and then the pictures can be used as a data set of a training model for training again, so that the identification and marking capability of the corresponding training model on the pictures can be improved.
And performing multi-level classification labeling on each picture according to the multi-label classification labeling task and the classification algorithm model to obtain a plurality of multi-label pictures, wherein each multi-label picture corresponds to a plurality of label types, and the multi-level classification labeling comprises orderly arranged multi-level labels.
Wherein, performing multi-level classification labeling on each picture may include labeling the same picture with a plurality of labels through a classification algorithm model, and the plurality of labels include labels of different levels to obtain a multi-label picture, for example: the same time scale is marked with three labels of vehicle, land and car. The above ordered multi-level tags may preferably be sequentially descending multi-level tags, and the descending order may indicate that the tag ranges are sequentially reduced, for example: the first-level label is larger than the second-level label, the first-level label is a vehicle, and the second-level label is an automobile. Of course, it can also be a multi-level tag in ascending order, for example: the first-level label is smaller than the second-level label, the second-level label is a vehicle, and the first-level label is an automobile. The ordering of the multi-level tags may be set by itself, and is not limited in the embodiment of the present invention.
And merging the multi-label pictures with the same multi-level label types into the same picture data set to obtain different types of picture data sets.
After the multiple pictures are subjected to multi-level label labeling, the pictures with the same multi-level labels can be combined to obtain a picture data set corresponding to the multiple multi-label pictures, and the picture data set can be loaded into a corresponding training model to train the model.
S202, selecting to newly build or select the created data channel corresponding to the picture data set.
Wherein, a new data channel (Pipeline) can be created with a new name, and a preset Pipeline can be selected. The previously established Pipeline may include a plurality of items such as a corresponding Pipeline name, a related description, an algorithm type, a current state, a creator, a creation time, an operation, and the like, and the Pipeline name to be created may be directly selected or input to be performed. The Pipeline may be equivalent to a Pipeline, and after loading the corresponding data set, each action is sequentially executed, and a result is output.
S203, adding the picture data set in the data channel, the pre-configured model and the function into a training frame corresponding to the data channel for training, and calculating the model threshold of the model corresponding to the picture data set in the data channel.
After the Pipeline is created, the image data set which is labeled and merged is loaded into the training resource of the Pipeline. The training resources also include pre-configured training models and functions, and the training models may be of various types, and the functions may also be of various types. The function may be a multi-stage function, and the upper-stage function may further include a lower-stage function. In Pipeline, a plurality of training frames corresponding to Pipeline are configured, corresponding association exists among the plurality of training frames, and different contents can be loaded in each training frame. The picture data set, the training model and the function in the training resource can be input into the corresponding training frame, and the model threshold value of the corresponding training model is obtained by calculating through clicking operation. In addition, operations such as pause, resume, terminate, withdraw, save, clone, etc. may also be performed in Pipeline.
By inputting different image data sets, corresponding functions and training models into the training frame for training, model thresholds of a plurality of training models can be obtained.
And S204, if the model threshold of the model does not reach the preset threshold, creating a task interface corresponding to the model, selecting a new picture data set for retraining, and stopping training the model reaching the preset threshold when the model threshold of the model reaches the preset threshold.
In the embodiment of the invention, a batch of unmarked pictures are selected for classification marking to obtain a picture data set, then corresponding Pipeline is created or newly built, the picture data set is loaded into the Pipeline, the picture data set, corresponding functions and training models in training resources in the Pipeline are extracted and added to corresponding positions in training frames, after clicking operation, model thresholds corresponding to the training models can be calculated according to corresponding relations among the training frames, namely after training is carried out based on a plurality of picture data sets, a quantitative numerical value corresponding to the training models can be converted according to the relation between the manually set marking number and the model thresholds. And when the model threshold value does not reach the preset threshold value, a task interface is created, and a new data set is reselected for classification and labeling and then is used for training the unqualified training model. Therefore, the training model is trained for many times by combining the autonomous learning and the manual labeling of the labeling system for many times, so that the training model reaches a preset threshold value, the accuracy of the training model is improved, and the cost of the manual labeling is reduced.
As shown in fig. 3, fig. 3 is a flowchart of another method for annotating a picture according to an embodiment of the present invention, including the following steps:
s301, classifying and labeling the obtained pictures according to the classification algorithm model to obtain different types of picture data sets.
S302, training the picture data set on the model with the same type as the picture data set to obtain the model threshold of the model.
S303, judging whether the model threshold of the model corresponding to the picture data set reaches a preset threshold.
After obtaining the model threshold of the model, extracting the model threshold and comparing the model threshold with a preset threshold, for example: the preset threshold value is 0.8, and the calculated model threshold value is 0.6, which indicates that the model threshold value does not reach the preset threshold value; and if the calculated model threshold is 0.9, the model threshold reaches a preset threshold.
And S304, if the number of the models does not reach the preset threshold value, triggering the digital light processing to create a task interface corresponding to the models, and selecting a new image data set to create a new labeling task for the models which do not reach the preset threshold value.
Among them, Digital Light Processing (DLP) is a technology that requires Digital Processing of an image signal and then projecting Light. The task interface can trigger DLP automatic creation inside the annotation system, and after the task interface is created, a new annotation task can be created again based on the task interface.
In a new labeling task, more new picture data sets can be included, and the new picture data sets are used for continuously training the training models which do not reach the preset threshold value. Of course, the new picture data set may include different types of new, re-selected and untrained pictures or pictures that have been trained but not successfully labeled.
S305, classifying and labeling the new labeling tasks until the model threshold value corresponding to the model which does not reach the preset threshold value in the new labeling tasks reaches the preset threshold value, and stopping training the model which reaches the preset threshold value.
After a new labeling task is formed, the pictures in the new labeling task can be used for training the models of the corresponding types again until the threshold value of the model reaches the preset threshold value, and the training is stopped. The model is also shown to be trained completely, the set requirement is met, and the model does not need to be trained continuously by using the picture data set of the corresponding type.
In the embodiment of the invention, different types of picture data sets are obtained by classifying and labeling pictures in advance, then the obtained different types of picture data sets are trained to obtain the model threshold of the model corresponding to the picture data sets, the model threshold is compared with the preset threshold, and when the model threshold obtained by training reaches the manually preset threshold, the model corresponding to the model threshold obtained by training can be considered to meet the requirements without continuing the training. Therefore, the model meeting the requirements after training can be used more effectively, the high-precision identification capability is realized, and when the model is in operation, the optimized model can reduce the labor cost and the threshold of manual use functions, so that the labeling efficiency is improved.
As shown in fig. 4, fig. 4 is a flowchart of another method for labeling a picture according to the embodiment of the present invention, which specifically includes the following steps:
s401, classifying and labeling the obtained pictures according to the classification algorithm model to obtain different types of picture data sets.
S402, training the picture data set on the model with the same type as the picture data set to obtain the model threshold of the model.
And S403, judging whether the model threshold value of the model corresponding to the plurality of picture data sets reaches a preset threshold value.
After model thresholds of a plurality of training models are obtained, the model thresholds can be compared with preset thresholds. One model threshold may be compared with a preset threshold at a time, or whether a plurality of model thresholds reach a preset threshold may be determined simultaneously, for example: model thresholds corresponding to the trained model A, B, C are 0.5, 0.6 and 0.8 in sequence, the preset threshold is 0.7, the training model C reaches the preset threshold, which indicates that the training model C meets the requirements of manual work and reaches an optimized state, and the training model A, B does not reach the preset threshold, so that the picture data set needs to be continuously obtained for training.
S404, if only part of the plurality of model thresholds reach the preset threshold, suspending continuous labeling of the model reaching the preset threshold, and retraining the model corresponding to the model threshold which does not reach the preset threshold based on a new labeling task.
If the system detects that only part of the model threshold values meet the preset threshold values through the active learning algorithm, the met models are temporarily marked, and the unsatisfied models are marked again, for example: the model thresholds corresponding to the trained model A, B, C are 0.5, 0.6 and 0.8 in sequence, the preset threshold is 0.7, and when the trained model C reaches the preset threshold, the training of the model C is suspended, and the model A, B reselects a new picture data set for retraining through a newly established tagging task.
Of course, it is also possible to compare the model threshold values of the models with the preset threshold values, and none of the model threshold values satisfies the preset threshold value, and all models need to be trained again. If the model threshold values of the models are compared with the preset threshold values, and all the model threshold values meet the preset threshold values, the continuous training of all the models can be suspended, and the training is finished.
S405, classifying and labeling the new labeling tasks until a model threshold value corresponding to the model which does not reach the preset threshold value in the new labeling tasks reaches the preset threshold value, and stopping training the model reaching the preset threshold value.
And classifying and labeling the new picture data set in the new labeling task, executing the same step as the step S101, loading the classified new picture data set into new Pipeline, continuing training the training model which does not reach the preset threshold value, and repeating the training process until all model threshold values corresponding to all training models reach the preset threshold value, which indicates that all training models have finished training. Therefore, when the obtained training model is used, the marking accuracy is improved, and meanwhile, the manual use cost is reduced.
In the implementation of the invention, a batch of unmarked pictures are selected for classification marking to obtain a picture data set, then the picture data set, the corresponding function and the training model are extracted and added to the corresponding positions in the training frames for operation, and then the model threshold value of the corresponding training model is calculated according to the corresponding relation between the training frames. When the model threshold value does not reach the preset threshold value, the digital optical processing creation task interface is triggered, a new data set is reselected, a new labeling task is created, the new labeling task is redistributed to labeling personnel for classification labeling, and then the new labeling task is used for training the training model which does not reach the standard. Like this, through establishing new mark task, select new picture data set and train the training model that does not reach predetermined threshold value until it reaches predetermined threshold value, be favorable to improving the accurate rate of training model to the accurate rate of training model improves and can reduce the cost of manual use.
As shown in fig. 5, fig. 5 is a schematic structural diagram of a picture annotation apparatus according to an embodiment of the present invention, where the picture annotation apparatus 500 includes:
the labeling module 501 is configured to perform classification labeling on the obtained pictures according to the classification algorithm model to obtain different types of picture data sets;
a training module 502, configured to train the image data set with a model of the same type as the image data set, to obtain a model threshold of the model;
a creating module 503, configured to create a task interface corresponding to the model to select a new picture data set for retraining if the model threshold of the model does not reach the preset threshold, and stop training the model reaching the preset threshold until the model threshold of the model reaches the preset threshold.
Optionally, as shown in fig. 6, fig. 6 is a schematic structural diagram of another picture labeling apparatus provided in the embodiment of the present invention, and the training module 502 includes:
a selecting unit 5021, configured to select a newly created or selected data channel corresponding to the picture data set;
the training unit 5022 is configured to add the picture data set in the data channel, the preconfigured model and the function to a training frame corresponding to the data channel for training, and calculate a model threshold of the model corresponding to the picture data set in the data channel.
Optionally, as shown in fig. 7, fig. 7 is a schematic structural diagram of another picture labeling apparatus provided in the embodiment of the present invention, and the creating module 503 includes:
a determining unit 5031, configured to determine whether a model threshold of a model corresponding to the picture data set reaches a preset threshold;
a first creating unit 5032, configured to trigger digital light processing to create a task interface corresponding to the model if the task interface does not reach the preset threshold, and select a new image dataset to create a new annotation task for the model that does not reach the preset threshold;
an allocating unit 5033, configured to classify and label the new labeling task until a model threshold corresponding to a model that does not reach the preset threshold in the new labeling task reaches the preset threshold, stop training the model that reaches the preset threshold.
Optionally, the determining unit 5031 is further configured to determine whether a model threshold of a model corresponding to the multiple image data sets reaches a preset threshold;
the allocating unit 5033 is further configured to, if only part of the plurality of model thresholds is detected to reach the preset threshold, suspend continuous labeling of the model reaching the preset threshold, and retrain the model corresponding to the model threshold that does not reach the preset threshold based on a new labeling task.
Optionally, as shown in fig. 8, fig. 8 is a schematic structural diagram of another picture annotation device provided in the embodiment of the present invention, and the annotation module 501 includes:
the acquiring unit 5011 is configured to acquire a multi-label classification labeling task;
the labeling unit 5012 is configured to perform multi-level classification labeling on each picture according to the multi-label classification labeling task and the classification algorithm model to obtain multiple multi-label pictures, where each multi-label picture corresponds to multiple label types, and the multi-level classification labeling includes orderly arranged multi-level labels;
the merging unit 5013 is configured to merge multiple tagged pictures of the same multi-level tag type into the same picture data set to obtain different types of picture data sets.
Optionally, the obtaining unit 5011 is further configured to obtain a multi-label classification labeling task;
the labeling unit 5012 is further configured to perform multi-level classification labeling on each picture according to the multi-label classification labeling task and the classification algorithm model to obtain a plurality of multi-label pictures, where each multi-label picture corresponds to a plurality of label types, and the multi-level classification labeling includes orderly arranged multi-level labels.
The merging unit 5013 is further configured to merge multi-labeled pictures of the same multi-level label type into the same picture data set to obtain different types of picture data sets.
In the embodiment of the present invention, the image labeling device can implement each step of the image labeling method provided in the above embodiment, and can achieve the same effect, and is not described herein again to avoid repetition.
As shown in fig. 9, fig. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, where the electronic device 90 includes: the image annotation method comprises a memory 902, a processor 901, a network interface 903 and a computer program which is stored on the memory 902 and can run on the processor 901, wherein the processor 901 implements the steps in the image annotation method provided by the above embodiments when executing the computer program.
Specifically, the processor 901 is configured to perform the following steps:
classifying and labeling the obtained pictures according to a classification algorithm model to obtain different types of picture data sets;
training a model of the same type as the picture data set by using the picture data set to obtain a model threshold value of the model;
and if the model threshold value of the model does not reach the preset threshold value, creating a task interface corresponding to the model, selecting a new picture data set for retraining, and stopping training the model reaching the preset threshold value when the model threshold value of the model reaches the preset threshold value.
Optionally, the step of training the picture data set to a model of the same type as the picture data set by the processor 901 to obtain the model threshold of the model specifically includes:
selecting a newly built or selected data channel corresponding to the picture data set;
and adding the picture data set, the pre-configured model and the function in the data channel into a training frame corresponding to the data channel for training, and calculating the model threshold value of the model corresponding to the picture data set in the data channel.
Optionally, the step, executed by the processor 901, of creating a task interface corresponding to the model and selecting a new picture data set for retraining if the model threshold of the model does not reach the preset threshold specifically includes:
judging whether the model threshold of the model corresponding to the picture data set reaches a preset threshold or not;
if the number of the models does not reach the preset threshold value, triggering digital light processing to establish a task interface corresponding to the models, selecting a new picture data set, and establishing a new annotation task for the models which do not reach the preset threshold value;
and classifying and labeling the new labeling tasks until the model threshold value corresponding to the model which does not reach the preset threshold value in the new labeling tasks reaches the preset threshold value, and stopping training the model reaching the preset threshold value.
Optionally, the step of determining whether the model threshold of the model corresponding to the image data set reaches the preset threshold executed by the processor 901 specifically includes:
judging whether the model threshold values of the models corresponding to the plurality of picture data sets reach preset threshold values or not;
and if only part of the plurality of model thresholds reach the preset threshold, suspending continuous labeling of the model reaching the preset threshold, and retraining the model corresponding to the model threshold which does not reach the preset threshold based on a new labeling task.
Optionally, the processor 901 performs a task of obtaining a single label classification label;
performing single-label classification labeling on a plurality of acquired pictures according to a single-label classification labeling task and a classification algorithm model to obtain a plurality of single-label pictures, wherein each single-label picture corresponds to one label type;
and merging the single-label pictures of the same label type into the same picture data set to obtain different types of picture data sets.
Optionally, the processor 901 performs a task of obtaining multi-label classification labels;
performing multi-level classification labeling on each picture according to a multi-label classification labeling task and a classification algorithm model to obtain a plurality of multi-label pictures, wherein each multi-label picture corresponds to a plurality of label types, and the multi-level classification labeling comprises orderly arranged multi-level labels;
and merging the multi-label pictures with the same multi-level label types into the same picture data set to obtain different types of picture data sets.
The electronic device 90 provided in the embodiment of the present invention can implement each implementation manner in the embodiment of the method for labeling a picture and corresponding beneficial effects, and for avoiding repetition, details are not repeated here.
It is noted that only 901 and 903 having components are shown, but it is understood that not all of the shown components are required and that more or fewer components may alternatively be implemented. As will be understood by those skilled in the art, the electronic device 90 is a device capable of automatically performing numerical calculation and/or information processing according to a preset or stored instruction, and the hardware includes, but is not limited to, a microprocessor, an Application Specific Integrated Circuit (ASIC), a Programmable gate array (FPGA), a Digital Signal Processor (DSP), an embedded device, and the like.
The electronic device 90 may be a desktop computer, a notebook, a palm computer, or other computing devices. The electronic device 90 may interact with a user via a keyboard, mouse, remote control, touch pad, or voice control device.
The memory 902 includes at least one type of readable storage medium including a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the storage 902 may be an internal storage unit of the electronic device 90, such as a hard disk or a memory of the electronic device 90. In other embodiments, the memory 902 may also be an external storage device of the electronic device 90, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), or the like, provided on the electronic device 90. Of course, the memory 902 may also include both internal and external memory units of the electronic device 90. In this embodiment, the memory 902 is generally used for storing an operating system installed in the electronic device 90 and various application software, such as program codes of the labeling method of pictures. In addition, the memory 902 may also be used to temporarily store various types of data that have been output or are to be output.
The network interface 903 may comprise a wireless network interface or a wired network interface, and the network interface 903 is generally used to establish a communication connection between the electronic device 90 and other electronic devices.
An embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when being executed by the processor 901, the computer program implements each process in the method for labeling pictures provided in the foregoing embodiment, and can achieve the same technical effect, and is not described here again to avoid repetition.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware related to instructions of a computer program, and the program can be stored in a computer readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.
The above disclosure is only for the purpose of illustrating the preferred embodiments of the present invention, and it is therefore to be understood that the invention is not limited by the scope of the appended claims.
Claims (10)
1. A method for labeling pictures is characterized by comprising the following steps:
classifying and labeling the obtained pictures according to a classification algorithm model to obtain different types of picture data sets;
training a model of the same type as the picture data set by using the picture data set to obtain a model threshold value of the model;
and if the model threshold value of the model does not reach the preset threshold value, creating a task interface corresponding to the model, selecting a new picture data set for retraining, and stopping training the model reaching the preset threshold value when the model threshold value of the model reaches the preset threshold value.
2. The method for labeling pictures according to claim 1, wherein the step of training the picture data set on a model of the same type as the picture data set to obtain the model threshold of the model specifically comprises:
selecting a newly built or selected data channel corresponding to the picture data set;
and adding the picture data set, the pre-configured model and the function in the data channel into a training frame corresponding to the data channel for training, and calculating the model threshold value of the model corresponding to the picture data set in the data channel.
3. The method for labeling pictures according to claim 1, wherein the step of creating a task interface corresponding to the model to select a new picture data set for retraining if the model threshold of the model does not reach a preset threshold specifically comprises:
judging whether the model threshold of the model corresponding to the picture data set reaches a preset threshold or not;
if the current image data set does not reach the preset threshold value, triggering digital light processing to establish a task interface corresponding to the model, and selecting the new image data set to establish a new labeling task for the model which does not reach the preset threshold value;
and classifying and labeling the new labeling tasks until the model threshold value corresponding to the model which does not reach the preset threshold value in the new labeling tasks reaches the preset threshold value, and stopping training the model reaching the preset threshold value.
4. The method for labeling a picture according to claim 3, wherein the step of determining whether the model threshold of the model corresponding to the picture data set reaches a preset threshold specifically comprises:
judging whether the model threshold values of the models corresponding to the plurality of picture data sets reach the preset threshold value or not;
and if only part of the plurality of model thresholds reach the preset threshold, suspending continuous labeling of the model reaching the preset threshold, and retraining the model corresponding to the model threshold which does not reach the preset threshold based on the new labeling task.
5. The method for labeling pictures according to claim 1, wherein the step of classifying and labeling the obtained pictures according to the classification algorithm model to obtain different types of picture data sets comprises:
acquiring a single label classification labeling task;
performing single-label classification labeling on the obtained pictures according to the single-label classification labeling task and the classification algorithm model to obtain multiple single-label pictures, wherein each single-label picture corresponds to one label type;
and merging the single label pictures of the same label type into the same picture data set to obtain different types of picture data sets.
6. The method for labeling pictures according to claim 1, wherein the step of classifying and labeling the obtained pictures according to the classification algorithm model to obtain different types of picture data sets comprises:
acquiring a multi-label classification labeling task;
performing multi-level classification labeling on each picture according to the multi-label classification labeling task and the classification algorithm model to obtain a plurality of multi-label pictures, wherein each multi-label picture corresponds to a plurality of label types, and the multi-level classification labeling comprises orderly arranged multi-level labels;
merging the multi-label pictures of the same multi-level label type into the same picture data set to obtain different types of picture data sets.
7. A picture labeling device is characterized by comprising:
the labeling module is used for classifying and labeling the acquired pictures according to the classification algorithm model to obtain different types of picture data sets;
the training module is used for training the image data set to a model with the same type as the image data set to obtain a model threshold value of the model;
and the creating module is used for creating a task interface corresponding to the model to select a new picture data set for retraining if the model threshold of the model does not reach the preset threshold, and stopping the training of the model reaching the preset threshold until the model threshold of the model reaches the preset threshold.
8. The apparatus for labeling pictures as claimed in claim 7, wherein said training module comprises:
the selection unit is used for selecting a newly-built or selected data channel corresponding to the picture data set;
and the training unit is used for adding the picture data set, the pre-configured model and the function in the data channel into a training frame corresponding to the data channel for training, and calculating the model threshold value of the model corresponding to the picture data set in the data channel.
9. An electronic device, comprising: memory, processor and computer program stored on the memory and executable on the processor, the processor implementing the steps in the method of labeling of pictures according to any of claims 1 to 6 when executing the computer program.
10. A computer-readable storage medium, characterized in that a computer program is stored thereon, which, when being executed by a processor, carries out the steps in the method of annotation of pictures according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911141366.XA CN112825144A (en) | 2019-11-20 | 2019-11-20 | Picture labeling method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911141366.XA CN112825144A (en) | 2019-11-20 | 2019-11-20 | Picture labeling method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112825144A true CN112825144A (en) | 2021-05-21 |
Family
ID=75906569
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911141366.XA Pending CN112825144A (en) | 2019-11-20 | 2019-11-20 | Picture labeling method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112825144A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114519404A (en) * | 2022-04-20 | 2022-05-20 | 四川万网鑫成信息科技有限公司 | Image sample classification labeling method, device, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764372A (en) * | 2018-06-08 | 2018-11-06 | Oppo广东移动通信有限公司 | Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set |
| CN109582793A (en) * | 2018-11-23 | 2019-04-05 | 深圳前海微众银行股份有限公司 | Model training method, customer service system and data labeling system, readable storage medium storing program for executing |
| WO2019100724A1 (en) * | 2017-11-24 | 2019-05-31 | 华为技术有限公司 | Method and device for training multi-label classification model |
| CN110196908A (en) * | 2019-04-17 | 2019-09-03 | 深圳壹账通智能科技有限公司 | Data classification method, device, computer installation and storage medium |
| US20190325259A1 (en) * | 2018-04-12 | 2019-10-24 | Discovery Communications, Llc | Feature extraction and machine learning for automated metadata analysis |
-
2019
- 2019-11-20 CN CN201911141366.XA patent/CN112825144A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019100724A1 (en) * | 2017-11-24 | 2019-05-31 | 华为技术有限公司 | Method and device for training multi-label classification model |
| US20190325259A1 (en) * | 2018-04-12 | 2019-10-24 | Discovery Communications, Llc | Feature extraction and machine learning for automated metadata analysis |
| CN108764372A (en) * | 2018-06-08 | 2018-11-06 | Oppo广东移动通信有限公司 | Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set |
| CN109582793A (en) * | 2018-11-23 | 2019-04-05 | 深圳前海微众银行股份有限公司 | Model training method, customer service system and data labeling system, readable storage medium storing program for executing |
| CN110196908A (en) * | 2019-04-17 | 2019-09-03 | 深圳壹账通智能科技有限公司 | Data classification method, device, computer installation and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| 张洛阳;毛嘉莉;刘斌;吴涛;: "基于贝叶斯模型的多标签分类算法", 计算机应用, no. 01 * |
| 汪鹏;张奥帆;王利琴;董永峰;: "基于迁移学习与多标签平滑策略的图像自动标注", 计算机应用, no. 11 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114519404A (en) * | 2022-04-20 | 2022-05-20 | 四川万网鑫成信息科技有限公司 | Image sample classification labeling method, device, equipment and storage medium |
| CN114519404B (en) * | 2022-04-20 | 2022-07-12 | 四川万网鑫成信息科技有限公司 | Image sample classification labeling method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109165249B (en) | Data processing model construction method and device, server and user side | |
| CN108629043A (en) | Extracting method, device and the storage medium of webpage target information | |
| RU2018119149A (en) | IDENTIFICATION OF TASKS IN MESSAGES | |
| US20210256326A1 (en) | Systems, techniques, and interfaces for obtaining and annotating training instances | |
| CN111340054A (en) | Data labeling method and device and data processing equipment | |
| CN109598307B (en) | Data screening method and device, server and storage medium | |
| US11551171B2 (en) | Utilizing natural language processing and machine learning to automatically generate proposed workflows | |
| CN109492222A (en) | Intension recognizing method, device and computer equipment based on conceptional tree | |
| US10489637B2 (en) | Method and device for obtaining similar face images and face image information | |
| US20170185913A1 (en) | System and method for comparing training data with test data | |
| US10162879B2 (en) | Label filters for large scale multi-label classification | |
| CN110827236A (en) | Neural network-based brain tissue layering method and device, and computer equipment | |
| CN113657483A (en) | Model training method, target detection method, device, equipment and storage medium | |
| US20170109680A1 (en) | System for standardization of goal setting in performance appraisal process | |
| CN112765403A (en) | Video classification method and device, electronic equipment and storage medium | |
| CN115905528A (en) | Event multi-label classification method and device with time sequence characteristics and electronic equipment | |
| CN110909768A (en) | Method and device for acquiring marked data | |
| CN112825144A (en) | Picture labeling method and device, electronic equipment and storage medium | |
| US10832161B2 (en) | Method and system of processing data for training a target domain classifier | |
| CN110532448B (en) | Document classification method, device, equipment and storage medium based on neural network | |
| CN116661936A (en) | Page data processing method and device, computer equipment and storage medium | |
| CN113554062B (en) | Training method, device and storage medium for multi-classification model | |
| CN111143568A (en) | Method, device and equipment for buffering during paper classification and storage medium | |
| CN113378958A (en) | Automatic labeling method, device, equipment, storage medium and computer program product | |
| CN112597012A (en) | Traversal method and device of application program, electronic equipment and storage medium |
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 |