CN114913098A - Image processing hyper-parameter optimization method, system, device and storage medium - Google Patents
Image processing hyper-parameter optimization method, system, device and storage medium Download PDFInfo
- Publication number
- CN114913098A CN114913098A CN202210741827.2A CN202210741827A CN114913098A CN 114913098 A CN114913098 A CN 114913098A CN 202210741827 A CN202210741827 A CN 202210741827A CN 114913098 A CN114913098 A CN 114913098A
- Authority
- CN
- China
- Prior art keywords
- hyper
- image
- parameter
- image processing
- model
- 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
- 238000012545 processing Methods 0.000 title claims abstract description 213
- 238000005457 optimization Methods 0.000 title claims abstract description 126
- 238000000034 method Methods 0.000 title claims abstract description 82
- 238000012549 training Methods 0.000 claims abstract description 60
- 230000003068 static effect Effects 0.000 claims abstract description 57
- 238000013441 quality evaluation Methods 0.000 claims description 7
- 230000009467 reduction Effects 0.000 abstract description 21
- 230000006870 function Effects 0.000 description 9
- 230000008569 process Effects 0.000 description 8
- 238000012544 monitoring process Methods 0.000 description 7
- 238000001514 detection method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000012937 correction Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000011478 gradient descent method Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 235000000332 black box Nutrition 0.000 description 2
- 238000005111 flow chemistry technique Methods 0.000 description 2
- 238000012806 monitoring device Methods 0.000 description 2
- 239000013307 optical fiber Substances 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000001303 quality assessment method Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The invention provides a method, a system, equipment and a storage medium for image processing hyper-parameter optimization, wherein the method comprises the following steps: acquiring an original image of a moving object, and acquiring a plurality of hyper-parameter combinations for training; training an image processing model based on the original image of the moving object and the super-parameter combination for training, and enabling the image processing model to be fitted with a preset image processing algorithm; acquiring an original image of a static object, and acquiring a hyper-parameter optimization reference image corresponding to the original image of the static object; and optimizing the hyper-parameter combination adopted by the image processing model based on the original image of the static object and the hyper-parameter optimization reference image. By adopting the invention, the automatic optimization of the 3D noise reduction parameters in the image processing is realized.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to an image processing hyper-parameter optimization method, system, equipment and storage medium.
Background
Commercial-grade imaging systems rely on an Image Signal Processing (ISP) process, which typically consists of several Pixel-level Image Processing modules that contain a large number of hyper-parameters for reconstructing the RAW Image on the sensor into an RGB Image. In the monitoring field, generally, hardware ISP hyper-parameters and reconstructed RGB images have complex interaction. In the conventional method, an experienced ISP engineer usually spends months to optimize the hyper-parameters, which not only consumes a lot of time resources, but also has difficulty in ensuring that the optimized parameters are globally or locally optimized in the incremental iteration process.
In recent years, various methods based on neural Network instead of ISP flow have been developed due to the excellent performance of Convolutional Neural Network (CNN) and Generative Adaptive Network (GAN) in the field of pixel-level image processing. The methods usually need strong computational support and cannot generate diversified output, and the generalization capability of the model in a real scene is difficult to guarantee and cannot meet the actual requirements of a monitoring scene. In order to solve the problems, Mosleh et al proposes an end-to-end optimization method for hardware inner loop, but the method needs to be optimized by combining hardware equipment, has low flexibility and cannot update parameters according to dynamic scene changes. Kosugi et al propose an optimization method based on Reinforcement Learning (Reinforcement Learning), but this method cannot satisfy the requirement of Black-box (Black-box) hardware ISP hyper-parameter optimization. In addition, the method cannot predict ISP hyper-parameters in real time according to dynamic changes of the monitoring scene.
In an actual scene, an ISP engineer usually needs to spend a lot of labor, material and time costs to adjust parameters of the monitoring device, so as to achieve an effect of optimal visual quality. The ISP process inside the monitoring equipment can be divided into a calibration module (such as color correction, automatic white balance, black level and shadow correction) and a parameter adjusting module (such as demosaicing, sharpening, 2D noise reduction, 3D noise reduction and the like). Although the AI algorithm in the past has a good effect on optimizing parameters (such as demosaicing, sharpening and 2D noise reduction) based on images in the parameter adjusting module, the optimization of the parameters for 3D noise reduction still has a problem. The 3D noise reduction parameters are different from image-based parameters, which act on video and have different noise reduction strengths for objects with different motion speeds.
Based on the above analysis, the main problems of the existing image-based AI optimization ISP algorithm are: because the data acquired by the proxy model are still images, the image effect of objects with different movement speeds under different 3D noise reduction parameter settings cannot be reflected.
Disclosure of Invention
Aiming at the problems in the prior art, the invention aims to provide an image processing hyper-parameter optimization method, system, device and storage medium, which can realize automatic optimization of 3D noise reduction parameters in image processing.
The embodiment of the invention provides an image processing hyper-parameter optimization method, which comprises the following steps:
acquiring an original image of a moving object, and acquiring a plurality of hyper-parameter combinations for training;
training an image processing model based on the original image of the moving object and the super-parameter combination for training, and enabling the image processing model to be fitted with a preset image processing algorithm;
acquiring an original image of a static object, and acquiring a hyper-parameter optimization reference image corresponding to the original image of the static object;
and optimizing the hyper-parameter combination adopted by the image processing model based on the original image of the static object and the hyper-parameter optimization reference image.
In the image processing hyper-parameter optimization method, the images of the moving object are acquired when the images used for fitting the training model are acquired, the images are used for reflecting the relation between different 3D noise reduction parameters and the moving object, the image processing algorithm of the image processing model fitting hardware image processing equipment is realized, then training data are further acquired to optimize the hyper-parameter combination based on the image processing model, the full-automatic image processing hyper-parameter optimization without human participation is realized, and the optimized parameters can be suitable for the image processing flow of software or hardware. Compared with the existing optimization method, the method realizes automatic optimization of the 3D noise reduction parameters in image processing.
In some embodiments, the moving object is a turntable rotating at a constant speed.
In some embodiments, training an image processing model based on the original image of the moving object and the plurality of hyper-parameter combinations for training, so that the image processing model fits a preset image processing algorithm, comprises the following steps:
generating a model optimization reference image corresponding to each training hyper-parameter combination based on each training hyper-parameter combination and a preset image processing algorithm for the original image of the moving object;
inputting the original image of the moving object and each super-parameter combination for training into the image processing model;
iteratively training the image processing model based on an output image of the image processing model and the corresponding model optimization reference image.
In some embodiments, obtaining a hyper-parametric optimized reference image corresponding to an original image of the static object includes the following steps:
processing the original image of the static object by adopting a preset image processing algorithm to obtain a first reference image;
acquiring a second reference image of the static object, which is obtained by shooting the same scene corresponding to the original image of the static object;
and aligning the first reference image and the second reference image by using an image registration algorithm to obtain the hyper-parameter optimization reference image.
In some embodiments, the preset image processing algorithm is configured to process a RAW image to obtain an RGB image, the original image is a RAW image, and the model-optimized reference image and the hyper-parameter-optimized reference image are RGB images, respectively.
In some embodiments, optimizing a hyper-parametric combination employed by the image processing model based on the original image of the static object and the hyper-parametric optimization reference image comprises the steps of:
initializing each hyper-parameter to obtain an initialized hyper-parameter combination;
inputting the original image of the static object and the initialized hyper-parameter combination into the image processing model to obtain a model processing image;
and calculating loss based on the model processing image and the hyper-parameter optimization reference image, and reversely optimizing the initialization hyper-parameter combination to obtain an optimized hyper-parameter combination.
In some embodiments, said acquiring raw images of a static object comprises acquiring at least two consecutive raw images of the static object;
and when loss is calculated based on the model processing image and the hyper-parameter optimization reference image, calculating time sequence consistency loss based on at least two model processing images corresponding to the at least two frames of continuous original images, calculating model prediction loss based on the model processing images and the hyper-parameter optimization reference image, and weighting and summing the time sequence consistency loss and the model prediction loss to obtain total loss.
In some embodiments, calculating the loss based on the model processing image and the hyper-parametric optimization reference image, and reversely optimizing the initialized hyper-parametric combination to obtain an optimized hyper-parametric combination, comprises the steps of:
setting a plurality of time sequence consistency loss weights, respectively calculating to obtain total loss based on each time sequence consistency loss weight, and then reversely optimizing the initialized hyper-parameter combinations to obtain a plurality of optimized hyper-parameter combinations corresponding to the time sequence consistency loss weights;
collecting original images for grading, inputting the original images for grading and each optimized hyper-parameter combination into a preset image processing algorithm to obtain a plurality of processed images corresponding to the optimized hyper-parameter combinations;
scoring the processed images respectively by adopting an image quality evaluation method;
and selecting the optimized hyper-parameter combination corresponding to the processed image with the highest score as the final hyper-parameter combination.
The embodiment of the invention also provides an image processing hyper-parameter optimization system, which is applied to the image processing hyper-parameter optimization method, and the system comprises:
the first acquisition module is used for acquiring an original image of a moving object and acquiring a plurality of hyper-parameter combinations for training;
the model fitting module is used for training an image processing model based on the original image of the moving object and the super-parameter combination for training, so that the image processing model fits a preset image processing algorithm;
the second acquisition module is used for acquiring an original image of the static object and acquiring a hyper-parameter optimization reference image corresponding to the original image of the static object;
and the hyper-parameter optimization module is used for optimizing a hyper-parameter combination adopted by the image processing model based on the original image of the static object and the hyper-parameter optimization reference image.
In the image processing hyper-parameter optimization system, the first acquisition module acquires images of moving objects when acquiring images for fitting a training model, the images are used for reflecting the relation between different 3D noise reduction parameters and the moving objects, the model fitting module realizes the image processing algorithm of image processing model fitting hardware image processing equipment, the second acquisition module and the hyper-parameter optimization module further acquire training data to optimize hyper-parameter combination based on the image processing model, full-automatic image processing hyper-parameter optimization without human participation is realized, and the optimized parameters can be suitable for the image processing flow of software or hardware. Compared with the existing optimization method, the method realizes automatic optimization of the 3D noise reduction parameters in image processing.
The embodiment of the invention also provides image processing hyper-parameter optimization equipment, which comprises:
a processor;
a memory having stored therein executable instructions of the processor;
wherein the processor is configured to perform the steps of the image processing hyper-parameter optimization method via execution of the executable instructions.
By adopting the image processing hyper-parameter optimization device provided by the invention, the processor executes the image processing hyper-parameter optimization method when executing the executable instruction, thereby obtaining the beneficial effect of the image processing hyper-parameter optimization method.
An embodiment of the present invention further provides a computer-readable storage medium for storing a program, where the program is executed by a processor to implement the steps of the image processing hyper-parameter optimization method.
By adopting the computer readable storage medium provided by the invention, the stored program realizes the steps of the image processing hyper-parameter optimization method when being executed, thereby the beneficial effects of the image processing hyper-parameter optimization method can be obtained.
Drawings
Other features, objects and advantages of the present invention will become more apparent upon reading of the following detailed description of non-limiting embodiments thereof, with reference to the accompanying drawings.
FIG. 1 is a flow chart of a method for image processing hyper-parameter optimization according to an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating an implementation process of an image processing hyper-parameter optimization method according to an embodiment of the present invention;
FIG. 3 is a flow diagram of training an image processing model according to an embodiment of the invention;
FIG. 4 is a flowchart of acquiring a hyper-parametric optimized reference image corresponding to an original image of the static object according to an embodiment of the present invention;
FIG. 5 is a flow diagram of a hyper-parameter set employed to optimize the image processing model according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of an image processing hyper-parameter optimization system in accordance with an embodiment of the present invention;
FIG. 7 is a schematic structural diagram of an image processing hyper-parameter optimization apparatus according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of a computer storage medium according to an embodiment of the present invention.
Detailed Description
Example embodiments will now be described more fully with reference to the accompanying drawings. Example embodiments may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar structures, and thus their repetitive description will be omitted.
As shown in FIG. 1, in one embodiment, the present invention provides an image processing hyper-parameter optimization method, comprising the following steps:
s100: acquiring an original image of a moving object, and acquiring a plurality of hyper-parameter combinations for training;
the super-parameter combination for training is obtained, namely the super-parameter combination is sampled, and a plurality of super-parameter combinations are obtained by setting the value of each super-parameter and then randomly combining the super-parameters for training an image processing model; the hyper-parameter is the hyper-parameter of the image processing algorithm or the software image processing algorithm applied to the ISP equipment;
s200: training an image processing model based on the original image of the moving object and the super-parameter combination for training, and enabling the image processing model to be fitted with a preset image processing algorithm;
in this embodiment, the preset image processing algorithm is implemented based on a hyper-parameter and is applied to an image processing algorithm or a software image processing algorithm in ISP equipment;
the purpose of this step is mainly to obtain an image processing model fitting the image processing algorithm, so that the image processing model can be used for subsequent hyper-parameter combinatorial optimization;
s300: acquiring an original image of a static object, and acquiring a hyper-parameter optimization reference image corresponding to the original image of the static object;
the static object and the moving object are preferably the same object, that is, the moving object in the moving state of the object is used in step S100, and the static object in the static state of the object is used in step S300; the hyper-parameter optimization reference image herein refers to an image taken by a contrast device (e.g., a digital camera) or a higher quality image obtained by PS;
s400: and optimizing the hyper-parameter combination adopted by the image processing model based on the original image of the static object and the hyper-parameter optimization reference image.
In the image processing hyper-parameter optimization method, the images of the moving object are acquired when the images used for fitting the training model are acquired in the step S100 and are used for reflecting the relation between different 3D noise reduction parameters and the moving object, the image processing algorithm of the image processing model fitting hardware image processing equipment is realized in the step S200, the training data is further acquired in the steps S300 and S400 to optimize the hyper-parameter combination based on the image processing model, the full-automatic image processing hyper-parameter optimization without human participation is realized, and the optimized parameters can be suitable for the image processing flow of software or hardware. Compared with the existing optimization method, the method realizes automatic optimization of the 3D noise reduction parameters in image processing. The method can be well suitable for optimization of the hyper-parameters in the image processing collected by the monitoring equipment in the monitoring scene, and can also be applied to other application scenes, such as face recognition unlocking, target anti-collision detection and the like.
In this embodiment, the original image is a RAW image, an output image obtained through a preset image processing algorithm is an RGB image, and an output image obtained through an image processing model that fits the image processing algorithm is also an RGB image. In the following, an original image is taken as a RAW image, and an image after image processing is taken as an RGB image for example, the image processing algorithm includes an algorithm for converting the RAW image into the RGB image, and may further include an algorithm for enhancing the image, for example, white balance processing, demosaicing processing, denoising processing, color correction processing, and the like. Correspondingly, the hyper-parameters may comprise the hyper-parameters used in each algorithm. However, the present invention is not limited to this, and in other embodiments, the original image and the output image may be images in other formats. The image processing algorithm of the present invention is not limited to the image processing algorithm for processing the image from the RAW image to obtain the RGB image, and may also be other types of image processing procedures, such as only denoising processing, only color correction processing, or enhancement procedure for the resolution of the image, or enhancement procedure for the brightness of the image, etc.
Each over-parameter is preset with a range of values. When the hyper-parameters are sampled, each ISP hyper-parameter is hierarchically sampled within the value range, and then randomly combined to obtain a plurality of hyper-parameter combinations, so that all the values of the hyper-parameters can be uniformly sampled. For example, the hyper-parameters and corresponding value range for demosaicing, sharpening, denoising are as follows in table 1:
TABLE 1 over-parameter value domain Range Table
In this embodiment, the optimized hyper-parameter obtained in step S400 may be used for hardware ISP equipment, or may be used for software image processing algorithm. In the following, the optimized hyper-parameter is used for an image acquisition device as an example, the image acquisition device may acquire a RAW image, obtain an RGB image through an ISP algorithm inside the image acquisition device, and use the image acquisition device as a debugging device. When applied to a monitoring scenario, the commissioning device may be a monitoring device. In other application scenarios, the debugging device may also be other types of image acquisition devices, such as a collision avoidance detection device, an object detection device, and the like. The implementation process of the image processing hyper-parameter optimization method is exemplarily shown in fig. 2. Wherein mainly include: sampling phase, proxy model fitting and parameter optimization. The sampling phase corresponds to the step S100 described above, and the collection of the hyper-parameter combination for training is realized, and the proxy model fitting corresponds to the step S200 described above, that is, the training image processing model is used as a proxy model for replacing the internal ISP algorithm of the debugging device. The parameter optimization corresponds to the above steps S300 and S400, and the hyper-parameters are optimized based on the image processing model. The RGB image obtained by the image processing model and the image processing algorithm can be used in the fields of face recognition, target detection, image quality detection and the like.
In this embodiment, the moving object is a turntable rotating at a constant speed. Because the turntable has different linear velocities from the circle center to the edge, the turntable area of the acquired image reflects the change of the image corresponding to different 3D noise reduction parameters, and therefore the image processing model can well model the 3D noise reduction parameters.
In step S100, acquiring an original image of a moving object, and acquiring a plurality of hyper-parameter combinations for training, including: firstly, arranging a turntable which can rotate at a constant speed continuously in a standard darkroom scene, and automatically acquiring a RAW image I of debugging equipment by using a media control flow snapshot program raw I.e. the original image of the moving object, and randomly setting the ISP hyper-parameter value combination P as the hyper-parameter combination for training.
The invention treats the image processing algorithm as a function f converting an input RAW image to an output RGB image ISP . Since parameterization can be performed by a hyper-parameter P, I rgb ＝f ISP (I raw (ii) a P). Thus, the ISP hyper-parameter optimization problem can be defined as:
wherein, P * Represents the optimal ISP hyper-parameter setting optimized by the task specific evaluation criteria; n represents the target image
The number of (2); p is ISP super parameter value combination; i is i raw Ith original image of moving object, I i gt An ith RGB image representing a snapshot of the contrast device. Equation (1) is the definition of the ISP hyper-parameter optimization for the conventional methodIn the number-optimized definition, L task Refer to different task evaluation criteria.
Image processing algorithms f in general in hardware or software ISP Is not differentiable, which results in the inability to optimize equation (1) using the gradient descent method. In order to solve the problem, the invention provides a differentiable proxy model f based on a convolutional neural network proxy And by training f proxy So that f proxy ≈f ISP . And f ISP Same, f proxy Inputting an input image I by adding a hyper-parameter combination P as input raw Is mapped into
So exist
W represents proxy model parameters. By optimizing the formula (2), the approximation f can be obtained ISP F of (a) proxy Parameter W * And M represents the number of training data. In this embodiment, the trained proxy model is an image processing model.
Representing the ith reference image in the process of training the proxy model, namely a label image used for calculating a loss function together with an output image of the proxy model;
an ith output image representing the proxy model.
As shown in fig. 2 and fig. 3, in this embodiment, in the step S200, training an image processing model based on the original image of the moving object and the combination of the plurality of hyper-parameters for training, so that the image processing model fits a preset image processing algorithm, includes the following steps:
s210: generating a model optimization reference image corresponding to each training hyper-parameter combination based on each training hyper-parameter combination and a preset image processing algorithm for the original image of the moving object;
specifically, under the condition that the hyper-parameter combination P for training is obtained, the original image I of the moving object is acquired through an ISP (internet service provider) in the debugging equipment raw Processed RGB image I rgb I.e. model optimized reference images. Through the acquisition process, a large amount of (I) can be obtained raw ,P,I rgb ) A triplet;
s220: inputting the original image of the moving object and each super-parameter combination for training into the image processing model W;
in this embodiment, the original image of a moving object is represented as I raw The combination of the training hyperparameters is P;
in this embodiment, the image processing model may be constructed by using a convolutional neural network model, and may specifically include a convolutional layer, a max-pooling layer, an upsampling layer, and the like, but the present invention is not limited thereto, and the image processing model may also be selected as another network structure;
s230: iteratively training the image processing model based on an output image of the image processing model and the corresponding model optimization reference image.
In this embodiment, the model-optimized reference image is denoted as I rgb 。
Namely an RGB image output based on an image processing model
And model optimized reference image I rgb Calculating the loss L proxy . After the model converges, a proxy model is obtained that can replace the ISP inside the commissioning device.
After convergence of the image processing model, f can be obtained proxy Model parameter W of * By fixing the model parameters W * Equation (1) can be rewritten as equation (3).
Thus, the corresponding penalty function L for a given task task In steps S300 and S400, a gradient descent method is used for ISP hyper-parameter combination P
And (6) optimizing. pi refers to one of all hyper-parameter combinations. Wherein, different given tasks can correspond to different loss functions L task (ii) a If the given task is higher image visual quality L task As a function of perceptual loss; given that the target detection accuracy rate of the task as an image is higher L task As a function of the detected loss. The loss function may be set according to specific requirements, but is not limited thereto.
In this embodiment, in step S300, acquiring an original image of a static object includes: and stopping rotating the turntable in the standard darkroom scene, wherein the whole darkroom scene is static, and acquiring the RAW image of the debugging equipment, namely the original image of the static object by using a media control flow snapshot program.
In this embodiment, at least two consecutive original images of the static object are acquired while acquiring the original image of the static object. The following description will take the example of acquiring two consecutive frames of original images. In particular, two consecutive frames of RAW images of a commissioning device are acquired using a media control flow snapshot procedure
As shown in fig. 4, in the step S300, acquiring a hyper-parametric optimized reference image corresponding to the original image of the static object includes the following steps:
s310: processing a first frame RAW image in the original image of the static object by adopting a preset image processing algorithm to obtain a first reference image;
in this embodiment, the original image of the static object is a continuous two-frame original image represented as
The original image of the first frame of static object is represented as
The original image of the second frame of static object is represented as
The first reference image is represented as
The first reference image is an RGB image obtained by processing an original image of the static object by an ISP algorithm in the debugging equipment;
s320: acquiring a second reference image of the static object, which is obtained by shooting the same scene corresponding to the original image of the static object;
specifically, the second reference image is an RGB image I captured by using a digital camera or contrast device for capturing the same scene corresponding to the original image of the still image gt 。
S330: and aligning the first reference image and the second reference image by using an image registration algorithm to obtain the hyper-parameter optimization reference image.
Specifically, in this embodiment, the first reference image is represented as
The second reference picture is denoted as I gt The hyper-parametric optimized reference image is represented as
Thereby obtaining a group
Ternary of (2)And (4) grouping.
As shown in fig. 5, in this embodiment, the step S400: optimizing a hyper-parametric combination adopted by the image processing model based on the original image of the static object and the hyper-parametric optimization reference image, comprising the steps of:
s410: initializing each hyper-parameter to obtain an initialized hyper-parameter combination, wherein the initialization can be random initialization to obtain the initialized hyper-parameter combination;
s420: inputting the original image of the static object and the initialized hyper-parameter combination into the image processing model to obtain a model processing image;
in this embodiment, the original image of the static object is represented as
The initialized hyper-parametric composition is denoted as P0 and the model-processed image is denoted as
And
s430: and calculating loss based on the model processing image and the hyper-parameter optimization reference image, and reversely optimizing the initialization hyper-parameter combination to obtain an optimized hyper-parameter combination.
Namely, to
And (4) solving loss by using a triplet, and optimizing P0 by using a gradient descent method to obtain the optimized hyperparameter P.
In this embodiment, in step S430, when calculating the loss based on the model-processed image and the hyper-parametric optimization reference image, a temporal consistency loss is calculated based on at least two model-processed images corresponding to the at least two frames of consecutive original images, a model prediction loss is calculated based on the model-processed image and the hyper-parametric optimization reference image, and the temporal consistency loss and the model prediction loss are weighted and summed to obtain a total loss.
Specifically, the total loss is calculated using the following formula:
and the more the lambda is, the smaller the optimized parameter noise intensity is, the better the time sequence consistency is and the lower the definition is, the smaller the lambda is, the more the optimized parameter noise intensity is, the worse the time sequence consistency is and the higher the definition is.
Generally, the problem of time sequence consistency on a video is solved, continuous multiframes need to be collected to serve as model input, and multiframe information is fused by using mechanisms such as optical flow, variability convolution or attention, so that the difficulty of data collection is increased, a media control flow program is limited by a memory and a bandwidth and cannot collect the continuous frames, the continuous frames also need to be collected when a hyper-parameter optimization reference image is collected, the problem that the positions of a moving object on the image are deviated due to inconsistent snapshot time of a contrast device and a debugging device is difficult to solve, and meanwhile, the requirement for hardware computing power is increased. And the loss of time sequence consistency is used, so that the data is easy to acquire, and a large amount of modification on an image-based AI optimization algorithm is not needed. Therefore, in this embodiment, since the image is processed by the model
And
the time sequence consistency loss can be solved, the optimized hyper-parameter combination P is effectively ensured, the inter-frame consistency and the lower noise intensity are good under the high-gain scene video, and the visual quality of the video is improved. Generally, the larger the timing consistency loss weight is, the smaller the noise jitter but the lower the definition of the AI-optimized 3D noise reduction parameter on the video is, and otherwise, the noise jitter is obvious but the higher the definition is.
As described above, when the timing consistency loss weight λ is selected to have different values, the output image quality obtained by the corresponding method is different, and the corresponding optimized hyper-parameter combinations are also different. In the embodiment, in order to solve the problem that an engineer depends on subjective adjustment and cannot realize automatic optimization, a method for automatically selecting an appropriate time sequence consistency loss weight based on image scoring is provided. Therefore, in this embodiment, the image processing hyper-parameter optimization method further includes a parameter evaluation section that selects an optimal ISP hyper-parameter using an image quality evaluation method.
Specifically, in step S430, calculating a loss based on the model processing image and the hyper-parametric optimization reference image, and performing reverse optimization on the initialized hyper-parametric combination to obtain an optimized hyper-parametric combination, the method includes the following steps:
setting a plurality of the timing consistency loss weights, and respectively adopting L based on each timing consistency loss weight task After the total loss is calculated by the calculation formula, reversely optimizing the initialized hyper-parameter combinations to obtain a plurality of optimized hyper-parameter combinations corresponding to the time sequence consistency loss weight;
specifically, in the parameter optimization stage, different λ ═ { λ is set 1 ,λ 2 … }, each lambda i An optimized hyper-parameter combination P is obtained through optimization i * ；
Collecting original images for grading, inputting the original images for grading and each optimized hyper-parameter combination into a preset image processing algorithm, and obtaining a plurality of processed images corresponding to the optimized hyper-parameter combinations; the original image for scoring can be a RAW image captured under the same scene of the turntable by using debugging equipment;
the plurality of processed images are respectively scored by adopting an image quality evaluation method, wherein the image quality evaluation method can be selected from non-reference-based image quality evaluation methods such as NIQE, BRISQE and the like;
and selecting the optimized hyper-parameter combination corresponding to the processed image with the highest score as the final hyper-parameter combination.
In particular, the hyper-parameter combinations P after a series of optimizations has been obtained i * Then, P is added i * Respectively leading the images into debugging equipment, acquiring original images for grading, and acquiring a plurality of processed images through the debugging equipment
Then, the λ with the highest score is obtained by an Image Quality Assessment method (IQA) i And P i * P is shown in formula (5) i * I.e. the final AI optimized 3D noise reduction parameters.
In another embodiment, the hyper-parameter combinations P are obtained after a series of optimizations i * Thereafter, P may also be added i * Respectively importing the images into the image processing models obtained by training in the step S200, acquiring the processed images output by the image processing models, and then obtaining the lambda with the highest score by an image quality evaluation method i And P i * 。
As shown in fig. 6, an embodiment of the present invention further provides an image processing hyper-parameter optimization system, which is applied to the image processing hyper-parameter optimization method, and the system includes:
the first acquisition module M100 is used for acquiring an original image of a moving object and acquiring a plurality of hyper-parameter combinations for training;
the model fitting module M200 is used for training an image processing model based on the original image of the moving object and the super parameter combination for training so as to enable the image processing model to fit a preset image processing algorithm;
the second acquisition module M300 is used for acquiring an original image of a static object and acquiring a hyper-parameter optimization reference image corresponding to the original image of the static object;
a hyper-parameter optimization module M400, configured to optimize a hyper-parameter combination adopted by the image processing model based on the original image of the static object and the hyper-parameter optimization reference image.
In the image processing hyper-parameter optimization system, the first acquisition module M100 acquires images of moving objects when acquiring images for fitting a training model, and is used for reflecting the relation between different 3D noise reduction parameters and the moving objects, the model fitting module M200 realizes the image processing algorithm of image processing model fitting hardware image processing equipment, the second acquisition module M300 and the hyper-parameter optimization module M400 further acquire training data to optimize hyper-parameter combination based on an image processing model, full-automatic image processing hyper-parameter optimization without human participation is realized, and the optimized parameters can be suitable for the image processing flow of software or hardware. Compared with the existing optimization method, the method realizes automatic optimization of the 3D noise reduction parameters in image processing.
In the image processing hyper-parameter system of the present invention, the functions of each module can be realized by adopting the specific implementation of the corresponding steps in the above method, which are not described again here.
The embodiment of the invention also provides image processing hyper-parameter optimization equipment, which comprises a processor; a memory having stored therein executable instructions of the processor; wherein the processor is configured to perform the steps of the image processing hyper-parameter optimization method via execution of the executable instructions.
As will be appreciated by one skilled in the art, aspects of the present invention may be embodied as a system, method or program product. Thus, various aspects of the invention may be embodied in the form of: an entirely hardware embodiment, an entirely software embodiment (including firmware, microcode, etc.) or an embodiment combining hardware and software aspects that may all generally be referred to herein as a "circuit," module "or" system.
An electronic device 600 according to this embodiment of the invention is described below with reference to fig. 7. The electronic device 600 shown in fig. 7 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present invention.
As shown in fig. 7, the electronic device 600 is embodied in the form of a general purpose computing device. The components of the electronic device 600 may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 that connects the various system components (including the storage unit 620 and the processing unit 610), a display unit 640, and the like.
Wherein the storage unit stores program code executable by the processing unit 610 to cause the processing unit 610 to perform steps according to various exemplary embodiments of the present invention described in the above-mentioned electronic prescription flow processing method section of the present specification. For example, the processing unit 610 may perform the steps as shown in fig. 1.
The storage unit 620 may include readable media in the form of volatile memory units, such as a random access memory unit (RAM)6201 and/or a cache memory unit 6202, and may further include a read-only memory unit (ROM) 6203.
The memory unit 620 may also include a program/utility 6204 having a set (at least one) of program modules 6205, such program modules 6205 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.
The electronic device 600 may also communicate with one or more external devices 700 (e.g., keyboard, pointing device, bluetooth device, etc.), with one or more devices that enable a user to interact with the electronic device 600, and/or with any devices (e.g., router, modem, etc.) that enable the electronic device 600 to communicate with one or more other computing devices. Such communication may occur via an input/output (I/O) interface 650. Also, the electronic device 600 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network such as the Internet) via the network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 via the bus 630. It should be appreciated that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.
By adopting the image processing hyper-parameter optimization device provided by the invention, the processor executes the image processing hyper-parameter optimization method when executing the executable instruction, thereby obtaining the beneficial effect of the image processing hyper-parameter optimization method.
An embodiment of the present invention further provides a computer-readable storage medium for storing a program, where the program is executed by a processor to implement the steps of the image processing hyper-parameter optimization method. In some possible embodiments, aspects of the present invention may also be implemented in the form of a program product comprising program code for causing a terminal device to perform the steps according to various exemplary embodiments of the present invention described in the above-mentioned electronic prescription flow processing method section of this specification, when the program product is run on the terminal device.
Referring to fig. 8, a program product 800 for implementing the above method according to an embodiment of the present invention is described, which may employ a portable compact disc read only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited in this regard and, in the present document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
The computer readable storage medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable storage medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
Program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or cluster. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., through the internet using an internet service provider).
By adopting the computer readable storage medium provided by the invention, the stored program realizes the steps of the image processing hyper-parameter optimization method when being executed, thereby the beneficial effects of the image processing hyper-parameter optimization method can be obtained.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (11)
1. An image processing hyper-parameter optimization method is characterized by comprising the following steps:
acquiring an original image of a moving object, and acquiring a plurality of hyper-parameter combinations for training;
training an image processing model based on the original image of the moving object and the super-parameter combinations for training, and enabling the image processing model to be matched with a preset image processing algorithm;
acquiring an original image of a static object, and acquiring a hyper-parameter optimization reference image corresponding to the original image of the static object;
and optimizing the hyper-parameter combination adopted by the image processing model based on the original image of the static object and the hyper-parameter optimization reference image.
2. The method of claim 1, wherein the moving object is a rotating disk rotating at a constant speed.
3. The method of claim 1, wherein training an image processing model based on a combination of the original image of the moving object and the plurality of training hyperparameters to fit the image processing model to a predetermined image processing algorithm comprises:
generating a model optimization reference image corresponding to each training hyper-parameter combination based on each training hyper-parameter combination and a preset image processing algorithm for the original image of the moving object;
inputting the original image of the moving object and each super-parameter combination for training into the image processing model;
iteratively training the image processing model based on an output image of the image processing model and the corresponding model optimization reference image.
4. The image processing hyper-parameter optimization method according to claim 1, wherein obtaining the hyper-parameter optimized reference image corresponding to the original image of the static object comprises the following steps:
processing the original image of the static object by adopting a preset image processing algorithm to obtain a first reference image;
acquiring a second reference image of the static object, which is obtained by shooting the same scene corresponding to the original image of the static object;
and aligning the first reference image and the second reference image by using an image registration algorithm to obtain the hyper-parameter optimization reference image.
5. The image processing hyper-parameter optimization method according to claim 4, wherein the preset image processing algorithm is used for processing a RAW image to obtain an RGB image, the original image is a RAW image, and the model optimization reference image and the hyper-parameter optimization reference image are respectively an RGB image.
6. The image processing hyper-parameter optimization method according to claim 4, wherein the hyper-parameter combination adopted by the image processing model is optimized based on the original image of the static object and the hyper-parameter optimization reference image, comprising the steps of:
initializing each hyper-parameter to obtain an initialized hyper-parameter combination;
inputting the original image of the static object and the initialized hyper-parameter combination into the image processing model to obtain a model processing image;
and calculating loss based on the model processing image and the hyper-parameter optimization reference image, and reversely optimizing the initialization hyper-parameter combination to obtain an optimized hyper-parameter combination.
7. The image processing hyper-parameter optimization method of claim 6, wherein said acquiring raw images of a static object comprises acquiring at least two consecutive raw images of the static object;
and when loss is calculated based on the model processing image and the hyper-parameter optimization reference image, calculating time sequence consistency loss based on at least two model processing images corresponding to the at least two frames of continuous original images, calculating model prediction loss based on the model processing images and the hyper-parameter optimization reference image, and weighting and summing the time sequence consistency loss and the model prediction loss to obtain total loss.
8. The method of claim 7, wherein calculating the loss based on the model-processed image and the hyper-parametric optimization reference image, and reversely optimizing the initialized hyper-parametric combination to obtain an optimized hyper-parametric combination, comprises the steps of:
setting a plurality of time sequence consistency loss weights, respectively calculating to obtain total loss based on each time sequence consistency loss weight, and then reversely optimizing the initialized hyper-parameter combinations to obtain a plurality of optimized hyper-parameter combinations corresponding to the time sequence consistency loss weights;
collecting original images for grading, inputting the original images for grading and each optimized hyper-parameter combination into a preset image processing algorithm to obtain a plurality of processed images corresponding to the optimized hyper-parameter combinations;
scoring the processed images respectively by adopting an image quality evaluation method;
and selecting the optimized hyper-parameter combination corresponding to the processed image with the highest score as the final hyper-parameter combination.
9. An image processing hyper-parameter optimization system, applied to the image processing hyper-parameter optimization method according to any one of claims 1 to 8, the system comprising:
the first acquisition module is used for acquiring an original image of a moving object and acquiring a plurality of hyper-parameter combinations for training;
the model fitting module is used for training an image processing model based on the original image of the moving object and the super-parameter combination for training, so that the image processing model fits a preset image processing algorithm;
the second acquisition module is used for acquiring an original image of the static object and acquiring a hyper-parameter optimization reference image corresponding to the original image of the static object;
and the hyper-parameter optimization module is used for optimizing a hyper-parameter combination adopted by the image processing model based on the original image of the static object and the hyper-parameter optimization reference image.
10. An image processing hyper-parameter optimization apparatus, comprising:
a processor;
a memory having stored therein executable instructions of the processor;
wherein the processor is configured to perform the steps of the image processing hyper-parameter optimization method of any of claims 1 to 8 via execution of the executable instructions.
11. A computer-readable storage medium storing a program, which when executed by a processor implements the steps of the image processing hyper-parameter optimization method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210741827.2A CN114913098A (en) | 2022-06-28 | 2022-06-28 | Image processing hyper-parameter optimization method, system, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210741827.2A CN114913098A (en) | 2022-06-28 | 2022-06-28 | Image processing hyper-parameter optimization method, system, device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114913098A true CN114913098A (en) | 2022-08-16 |
Family
ID=82772857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210741827.2A Pending CN114913098A (en) | 2022-06-28 | 2022-06-28 | Image processing hyper-parameter optimization method, system, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114913098A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115272423A (en) * | 2022-09-19 | 2022-11-01 | 深圳比特微电子科技有限公司 | Method and device for training optical flow estimation model and readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113610911A (en) * | 2021-07-27 | 2021-11-05 | Oppo广东移动通信有限公司 | Training method and device of depth prediction model, medium and electronic equipment |
| EP3907694A1 (en) * | 2020-05-08 | 2021-11-10 | Research Institute Of Tsinghua University in Shenzhen | Method and system for joint optimization of isp and vision tasks, medium and electronic device |
| CN113688945A (en) * | 2021-09-29 | 2021-11-23 | 苏州科达科技股份有限公司 | Image processing hyper-parameter optimization method, system, device and storage medium |
| CN114092589A (en) * | 2022-01-19 | 2022-02-25 | 苏州瑞派宁科技有限公司 | Image reconstruction method and training method, device, equipment and storage medium |
| CN114169380A (en) * | 2020-08-19 | 2022-03-11 | Oppo广东移动通信有限公司 | Training method and device of image processing model, electronic equipment and storage medium |
-
2022
- 2022-06-28 CN CN202210741827.2A patent/CN114913098A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3907694A1 (en) * | 2020-05-08 | 2021-11-10 | Research Institute Of Tsinghua University in Shenzhen | Method and system for joint optimization of isp and vision tasks, medium and electronic device |
| CN114169380A (en) * | 2020-08-19 | 2022-03-11 | Oppo广东移动通信有限公司 | Training method and device of image processing model, electronic equipment and storage medium |
| CN113610911A (en) * | 2021-07-27 | 2021-11-05 | Oppo广东移动通信有限公司 | Training method and device of depth prediction model, medium and electronic equipment |
| CN113688945A (en) * | 2021-09-29 | 2021-11-23 | 苏州科达科技股份有限公司 | Image processing hyper-parameter optimization method, system, device and storage medium |
| CN114092589A (en) * | 2022-01-19 | 2022-02-25 | 苏州瑞派宁科技有限公司 | Image reconstruction method and training method, device, equipment and storage medium |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115272423A (en) * | 2022-09-19 | 2022-11-01 | 深圳比特微电子科技有限公司 | Method and device for training optical flow estimation model and readable storage medium |
| CN115272423B (en) * | 2022-09-19 | 2022-12-16 | 深圳比特微电子科技有限公司 | Method and device for training optical flow estimation model and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Dai et al. | Curriculum model adaptation with synthetic and real data for semantic foggy scene understanding | |
| JP7206386B2 (en) | Image processing model training method, image processing method, network device, and storage medium | |
| CN110249622B (en) | Real-time semantic aware camera exposure control | |
| CN111246091B (en) | Dynamic automatic exposure control method and device and electronic equipment | |
| WO2022134971A1 (en) | Noise reduction model training method and related apparatus | |
| CN113066017B (en) | Image enhancement method, model training method and equipment | |
| US11030750B2 (en) | Multi-level convolutional LSTM model for the segmentation of MR images | |
| US11392799B2 (en) | Method for improving temporal consistency of deep neural networks | |
| CN112270688A (en) | Foreground extraction method, device, equipment and storage medium | |
| CN112927127A (en) | Video privacy data fuzzification method running on edge device | |
| CN114913098A (en) | Image processing hyper-parameter optimization method, system, device and storage medium | |
| Xin et al. | Video face super-resolution with motion-adaptive feedback cell | |
| CN116486250A (en) | Multi-path image acquisition and processing method and system based on embedded type | |
| Yoshimura et al. | Dynamicisp: dynamically controlled image signal processor for image recognition | |
| WO2024104365A1 (en) | Device temperature measurement method and related device | |
| CN110366034A (en) | A kind of super-resolution image processing method and processing device | |
| CN113596576A (en) | Video super-resolution method and device | |
| CN113792817A (en) | Image processing hyper-parameter prediction method, system, device and storage medium | |
| JP2023126130A (en) | Computer-implemented method, data processing apparatus and computer program for object detection | |
| WO2022227916A1 (en) | Image processing method, image processor, electronic device, and storage medium | |
| CN113688945A (en) | Image processing hyper-parameter optimization method, system, device and storage medium | |
| CN115482588A (en) | Method for predicting posture of three-dimensional model and electronic equipment | |
| KR20230064959A (en) | Surveillance Camera WDR(Wide Dynamic Range) Image Processing Using Object Detection Based on Artificial Intelligence | |
| CN113706390A (en) | Image conversion model training method, image conversion method, device and medium | |
| Hou et al. | HDR-like image generation to mitigate adverse wound illumination using deep bi-directional retinex and exposure fusion |
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 |