CN110020987A - A kind of medical image super resolution ratio reconstruction method based on deep learning - Google Patents
A kind of medical image super resolution ratio reconstruction method based on deep learning Download PDFInfo
- Publication number
- CN110020987A CN110020987A CN201910224792.3A CN201910224792A CN110020987A CN 110020987 A CN110020987 A CN 110020987A CN 201910224792 A CN201910224792 A CN 201910224792A CN 110020987 A CN110020987 A CN 110020987A
- Authority
- CN
- China
- Prior art keywords
- data
- resolution
- medical image
- coordinate
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000013135 deep learning Methods 0.000 title claims abstract description 9
- 238000012549 training Methods 0.000 claims abstract description 32
- 230000003902 lesion Effects 0.000 claims abstract description 14
- 238000005070 sampling Methods 0.000 claims description 8
- 230000003416 augmentation Effects 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 3
- 208000035346 Margins of Excision Diseases 0.000 claims description 2
- 230000006378 damage Effects 0.000 claims description 2
- 238000004088 simulation Methods 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims 1
- 238000003384 imaging method Methods 0.000 abstract description 3
- 230000000694 effects Effects 0.000 description 8
- 238000011156 evaluation Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 3
- 238000013480 data collection Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000003814 drug Substances 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
-
- 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)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radiology & Medical Imaging (AREA)
- Evolutionary Computation (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Artificial Intelligence (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Image Processing (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The present invention provides a kind of method of medical image super-resolution rebuilding based on deep learning, medical image is directly inputted in system, correct super-resolution result can be obtained, network is trained by a large amount of high quality medical image, using the location information for introducing lesion in medical image in the training stage, the location information refer to lesion centre coordinate and size or lesion it is fine edge mark, by the way that these prior informations are added, can solve other methods medical image super-resolution network instruct it is askew caused by medical image after super-resolution the problem of being lost script imaging value.
Description
Technical field
The invention belongs to technical field of computer vision, more particularly to a kind of medical image super-resolution based on deep learning
Rate method for reconstructing.
Background technique
With the development of deep learning, computer vision and the horizontal of image procossing are constantly breaking through in some scenes
The bottleneck of conventional method.Image super-resolution is same to flourish as the important technology in a kind of computer vision.Natural field
Super-resolution method under scape based on deep learning emerges one after another, and has also derived a variety of different network design modes, network
Depth is continuously increased, and feature multiplexing capacity constantly enhances, and the PSNR index under different scale is also rising steadily, and is supervised in video
It has a wide range of applications in the actual scenes such as control, video restoration, digital high-definition and satellite image.
The limitation of dosage when limitation or medical imaging due to medical instrument hardware are shot, super-resolution is in medical image
Also there is very big development space in field, and there are also researchers at present directly migrates to medical image for the method in natural image
On, although directly will be in natural image in PSNR index and visual effect is as also obtaining some promotions, in this
Method moves on more serious medical image, and the way not adjusted accordingly according to the specific features of medical image
Leave a question open.Different with natural image, there is positive region (lesion) and negative areas (normal tissue) in medical image,
Method of the conventional natural image image super-resolution directly by image block training can be caused due to yin-yang data extremely imbalance
Model is askew to going in normal tissue, can cause destruction to the medical image information with imaging value in this way.And
An evaluation index of the PSNR as an image global information signal level only uses PSNR as measurement medical image super-resolution
The evaluation index of rate effect be also it is incomplete, the super-resolution efect that can not evaluate really important region is good on earth not
Good, the signal of focal area may be buried in a large amount of normal tissue regions signal.
Summary of the invention
For the problem to be solved in the present invention as mentioned in background technique, the present invention provides a kind of medicine shadow based on deep learning
As super resolution ratio reconstruction method, main contents are as follows:
(1) medical image features are directed to, the additional markup information of medical data collection is utilized, proposes new data set
The method to solve the problems, such as yin-yang sample imbalance is referred to as PN-sample hereinafter by production and Training strategy
balance。
(2) due to generally acknowledging effective medical image super-resolution rebuilding Indexes of Evaluation Effect, this hair there is no one at present
The bright evaluation index for proposing to use CAD system combination PSNR association evaluation medical image super-resolution rebuilding effect.
The introducing of CAD system, which can make up PSNR, can not measure the focal area whether super-resolution is really paid close attention to doctor
The defect of effective enhancing is carried out.Only super-resolution technique is carried out towards being correctly oriented, and the promotion of the index of PSNR is also
It is meaningful.
Opposite with the prior art, the present invention has following clear superiority:
1, the PN-sample balance method proposed can effectively guarantee model training stability, guarantee yin-yang area
Domain can access the learning effect preferably balanced.
2, the association evaluation index proposed can effectively identify that the promotion of major part negative areas causes when avoiding network instruction askew
Whole PSNR promoted the case where.It can ensure the validity of medical image super-resolution efect.
Detailed description of the invention:
Fig. 1 is flow chart of the present invention
Fig. 2 is the 3D dense network structure that the present invention uses；
Fig. 3 is the network structure of Dense Block and Compressor that the present invention designs；
Fig. 4 validation PSNR performance, left figure are existing method, the method that right figure is proposed for the present invention 1；
Fig. 5 is that the present invention 2 verifies, CAD system effect contrast figure；
Specific embodiment
In order to enable the objectives, technical solutions, and advantages of the present invention are more clearly understood, below in conjunction with attached drawing 1-5 and reality
Example is applied, the present invention will be described in further detail.It should be appreciated that specific embodiment described herein is only to explain this hair
It is bright, it is not intended to limit the present invention.In addition, technical characteristic involved in each embodiment of present invention described below is only
Not constituting a conflict with each other can be combined with each other.
The method and system of the present invention provides a set of medical image super-resolution rebuilding based on deep learning.By medicine
Image, which is directly inputted in system, can be obtained correct super-resolution result.By a large amount of high quality medical image come
Training network neutralizes other methods the difference is that we believe in the position that the training stage introduces lesion in medical image
Breath, but do not needed when test, meet actual demand.Herein location information refer to lesion centre coordinate and size or lesion it is fine
Edge mark.By the way that these prior informations are added, it is askew that we can solve other methods medical image super-resolution network instruction
Medical image after caused super-resolution is lost the problem of script imaging value, our the method are referred to as PN-sample
Balance method.And propose that CAD system joint PSNR jointly really effectively comments the progress of medical image super-resolution performance
Estimate, super-resolution method should be promoted under the premise of ensuring CAD system performance boost PSNR just it is significant, if super-resolution
The PSNR that data afterwards cause the decline of CAD system performance to be promoted is lost meaning, may be to carry out to some extraneous areas
Enhancing, and the shortcomings that our the focal area information of real concern have been destroyed, this is also PSNR, CAD system
Introducing can effectively make up this disadvantage.
In the embodiment of the present invention, step (1) specifically includes that the acquisition and lesions position information of high quality medical image
It introduces and handles.
Specifically, step (1-1), collects the medical images data sets of high quality, and clean to data, by data set
In poor data delete.Since our task is super-resolution rebuilding, so the high-resolution in tri- directions z, y, x is
It is contemplated that primary factor, the spacing that we are typically chosen three directions be respectively less than 1mm be quality data.Spacing
The spatial information being meant that in a voxel is included in medical image actual physics space, the spacing in three directions
Smaller, the object of actual size will be made of more voxels, that is, resolution ratio is higher.And we need to collect data
The prior information that the location information of middle lesion additionally considers when training as us.
Step (1-2), we will start from super-resolution training dataset after gathering data, in order to higher suitable
With property, we are by the way of multiple dimensioned while training, first to quality data collection on the target direction in three directions
The down-sampling operation of corresponding scale is carried out, then original resolution is interpolated by bicubic interpolation method or Cubic Spline Method,
The data and initial data of low resolution as simulation form low resolution (ILR) and high-resolution (IHR) data pair
(data-pairs).To lesions position information, we by Vo xel coords=(World coords-Origin)/
Spacing converts true coordinate information to and the matched voxel coordinate information of 3-D image.Coordinate is according to specific doctor
The dimension for treating image is specifically divided into two-dimensional coordinate (y, x) or three-dimensional coordinate (z, y, x).Vo xel coords is medical image square
Corresponding coordinate in battle array, World coords are the world coordinate system of lesion, and origin is the coordinate origin of filming instrument,
The spatial information that spacing then includes by a voxel mentioned above.We need to record and image used in us
Matched Voxel coords is saved in memory for instructing sampling policy when training.
The training and optimization of the main packet network of step (2).
It is used to train specifically, data are directly cut block before training different from other methods by step (2-1), we adopt
With location information instruct online stripping and slicing strategy, every time obtain training data when according to the yin-yang ratio data of artificial settings come
Generate random number, it is assumed that it is considered that yin-yang is a reasonable ratio than 2: 1, then it can first generate random number 0,1,2, if
0 is generated, then goes to take a coordinate information at random in the positive region coordinate information of preservation in memory from step (1-2), and
The disturbance of 20 voxels, by coordinate centered on the coordinate after disturbance, the institute in step (1-2) are added in each dimension in space
The low resolution high-resolution data stated is to youngster (ILR、IHRData-pairs a biggish data block is intercepted in) respectively, herein
" biggish data block " is larger when referring to than formal training, cuts off extra part after augmentation again, prevents training data from doing
Edge is caused to have the case where padding after complete online data augmentation.By carrying out to training data block: flip horizontal hangs down
Straight overturning, front and back overturning, translation, scaling, margins of excision redundance is to 42*42* after the random augmentation operation such as multi-angle rotary
The size of 42 (being herein 3 dimension datas, 2-D data is then 42*42) is used for network training.If random number is 1 or 2, scheming
Then the random active position sampling as in is carried out at the identical stripping and slicing augmentation operation of positive region, for training.
The batch of data block for meeting yin-yang ratio that sampling in step (1-2) obtains is sent into subsequent net by step (2-2)
It goes to be trained in network.High-definition picture (the I of reconstruction after obtaining super-resolutionSR), the I that will be obtainedSRWith as goldstandard
IHRCarry out MSE lossIt calculates
Loss is rebuild, then network parameter is optimized by back-propagation algorithm (back propagation).I in formula, j, k are respectively three
The coordinate serial number in direction.It is subsequent use that suitable super-resolution network model is chosen after training is steady.
The mode of the outstanding densenet of our service performances of network intensively connected carrys out planned network herein, to low layer height
Layer feature carries out multiplexing fusion, with reach it is better as a result, as shown in Fig. 2, and the parameter amount that is used under same performance it is smaller,
Can the characterization ability to limited parameter maximumlly developed, be illustrated in figure 3 we dense block and
The basic structure of compressor, dense block are dense connection type, complete main feature extraction work,
Compressor is mainly in the effect for wherein playing Data Dimensionality Reduction, because 3D data are very easy to cause the deficiency of video memory.
So far our first inventive point statement finishes, when without reference to our second inventive point, i.e. oversubscription net
Whether network correctly enhances medical image, only in PSNR index we to have also exceeded traditional training method very much,
Fig. 4 is our existing Comparative results.Though not considering that yin-yang sample imbalance problem present in medical image also has larger
It is promoted, but limited capacity.And the present invention 2 in it can be found that existing oversubscription method there are the problem of.
Step 3, existing super-resolution method is verified in the present invention 2, that verifies super-resolution method has conscientiously
Effect property.The validity of the present invention 1 can be verified simultaneously.Note that the present invention 2 is verified in real data, that is, test
Used in high-resolution data be the high-resolution data really acquired, and it is inartificial by sampling obtained false low point
Resolution data.Wherein:
Step (3-1) first carries out the training and test of CAD system in the data operated without any oversubscription, as
This batch data attainable performance baseline.
True high-resolution data is carried out oversubscription operation using the model of training in our steps (2) by step (3-2)
Enhanced to obtain ISR, and use ISRInstead of original high-resolution data, according to identical parameter training CAD system algorithm,
Obtain the performance indicator of the present invention 1.
Step (3-3) uses other configurations all the same in step (2) in addition to the equilibrium between yin and yang method in the present invention 1, into
Row training, represents as existing methods.
Step (3-4) compares experiment three times, it is found that existing oversubscription method will lead to the appearance of a large amount of false positives,
Also mean that the medical information in influencing has been destroyed.And the method in the present invention 1 is then it is possible to prevente effectively from this problem, and
Performance is promoted, as shown in Figure 5.
Step (3-5) to sum up illustrates the problem of present invention 2 can find out the study of mistake existing for existing oversubscription method, can
More fully evaluate the performance of medical image super-resolution.And the present invention 1 then can effectively solve the above problems, correctly to doctor
It learns image feature to be learnt, to achieve the effect that correctly to promote medical image quality.
Above embodiments are only exemplary embodiment of the present invention, are not used in the limitation present invention, protection scope of the present invention
It is defined by the claims.Those skilled in the art can within the spirit and scope of the present invention make respectively the present invention
Kind modification or equivalent replacement, this modification or equivalent replacement also should be regarded as being within the scope of the present invention.
Claims (1)
1. a kind of medical image super resolution ratio reconstruction method based on deep learning characterized by comprising
The acquisition of step 1, high quality medical image
Step (1-1), the medical images data sets for collecting high quality, and data are cleaned；
Step (1-2) makes super-resolution training dataset, by the way of multiple dimensioned while training, first to high quality number
The down-sampling operation of corresponding scale is carried out on the target direction in three directions according to collection, then by bicubic interpolation method or
Cubic Spline Method is interpolated into original resolution, and the data and initial data of the low resolution as simulation form low resolution
(ILR) and high-resolution (IHR) data pair (data-pairs) passes through Voxel coords=to lesions position information
(World coords-Origin)/Spacing converts true coordinate information to and sits with the matched voxel of 3-D image
Mark information.Coordinate is specifically divided into two-dimensional coordinate (y, x) or three-dimensional coordinate (z, y, x) according to the dimension of specific medical image,
Voxel coords is corresponding coordinate in medical image matrix, and World coords is the world coordinate system of lesion, origin
The spatial information for then including by a voxel mentioned above for the coordinate origin of filming instrument, spacing；
The introducing and processing of step 2, lesions position information
Data are cut block and are used to train by step (2-1) before training, the online stripping and slicing strategy instructed using location information,
Random number is generated according to the yin-yang ratio data of artificial settings when obtaining training data every time, it is assumed that yin-yang ratio is 2:1, then may be used
First to generate random number 0,1,2, if generating 0, go to save positive region coordinate information in memory from step (1-2)
In take a coordinate information at random, and in each dimension in space be added 20 voxels disturbance, using the coordinate after disturbance as
Centre coordinate, the low resolution high-resolution data described in step (1-2) is to youngster (ILR、IHRData-pairs in) respectively
Intercept a biggish data block；By carrying out to training data block: flip horizontal, flip vertical, front and back overturn, translate, put
The size of margins of excision redundance to 42*42*42 are used for network training after contracting, the random augmentation operation of multi-angle rotary；If with
Machine number is 1 or 2, then random active position sampling in the picture, is then carried out at the identical stripping and slicing augmentation operation of positive region,
For training；
Step (2-2), sampling obtains in step (1-2) the data block for meeting yin-yang ratio is sent into subsequent network go into
Row training, the high-definition picture (I of the reconstruction after obtaining super-resolutionSR), the I that will be obtainedSRWith the I as goldstandardHRIt carries out It calculates and rebuilds damage
It loses, then network parameter is optimized by back-propagation algorithm (back propagation), wherein i, j, k are respectively three directions
Coordinate serial number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910224792.3A CN110020987B (en) | 2019-03-24 | 2019-03-24 | Medical image super-resolution reconstruction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910224792.3A CN110020987B (en) | 2019-03-24 | 2019-03-24 | Medical image super-resolution reconstruction method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110020987A true CN110020987A (en) | 2019-07-16 |
| CN110020987B CN110020987B (en) | 2023-06-30 |
Family
ID=67189877
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910224792.3A Active CN110020987B (en) | 2019-03-24 | 2019-03-24 | Medical image super-resolution reconstruction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110020987B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110610459A (en) * | 2019-09-17 | 2019-12-24 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Image processing method and device |
| CN111192255A (en) * | 2019-12-30 | 2020-05-22 | 上海联影智能医疗科技有限公司 | Index detection method, computer device, and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105654425A (en) * | 2015-12-07 | 2016-06-08 | 天津大学 | Single-image super-resolution reconstruction method applied to medical X-ray image |
| US20180075581A1 (en) * | 2016-09-15 | 2018-03-15 | Twitter, Inc. | Super resolution using a generative adversarial network |
| CN108615236A (en) * | 2018-05-08 | 2018-10-02 | 上海商汤智能科技有限公司 | A kind of image processing method and electronic equipment |
| CN109360152A (en) * | 2018-10-15 | 2019-02-19 | 天津大学 | 3 d medical images super resolution ratio reconstruction method based on dense convolutional neural networks |
-
2019
- 2019-03-24 CN CN201910224792.3A patent/CN110020987B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105654425A (en) * | 2015-12-07 | 2016-06-08 | 天津大学 | Single-image super-resolution reconstruction method applied to medical X-ray image |
| US20180075581A1 (en) * | 2016-09-15 | 2018-03-15 | Twitter, Inc. | Super resolution using a generative adversarial network |
| CN108615236A (en) * | 2018-05-08 | 2018-10-02 | 上海商汤智能科技有限公司 | A kind of image processing method and electronic equipment |
| CN109360152A (en) * | 2018-10-15 | 2019-02-19 | 天津大学 | 3 d medical images super resolution ratio reconstruction method based on dense convolutional neural networks |
Non-Patent Citations (1)
| Title |
|---|
| 刘蓬博: "基于深度学习的医学图像增强算法的研究" * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110610459A (en) * | 2019-09-17 | 2019-12-24 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Image processing method and device |
| CN111192255A (en) * | 2019-12-30 | 2020-05-22 | 上海联影智能医疗科技有限公司 | Index detection method, computer device, and storage medium |
| CN111192255B (en) * | 2019-12-30 | 2024-04-26 | 上海联影智能医疗科技有限公司 | Index detection method, computer device, and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110020987B (en) | 2023-06-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9317970B2 (en) | Coupled reconstruction of hair and skin | |
| CN109934832A (en) | Liver neoplasm dividing method and device based on deep learning | |
| CN105453139A (en) | Sparse GPU voxelization for 3D surface reconstruction | |
| CN107464216A (en) | A kind of medical image ultra-resolution ratio reconstructing method based on multilayer convolutional neural networks | |
| Zhang et al. | Data-driven synthetic modeling of trees | |
| Zhang et al. | Adaptive and quality 3d meshing from imaging data | |
| Luo et al. | Details preserved unsupervised depth estimation by fusing traditional stereo knowledge from laparoscopic images | |
| CN109285225A (en) | A kind of method for building up of the virtual reality auxiliary operation based on medical image | |
| CN107341846A (en) | A kind of method and apparatus of the extensive three-dimensional reconstruction scene of real-time display | |
| CN112785692B (en) | Single-view-angle multi-person human body reconstruction method based on depth UV prior | |
| CN110020987A (en) | A kind of medical image super resolution ratio reconstruction method based on deep learning | |
| Muqeet et al. | Hybrid residual attention network for single image super resolution | |
| Zhang et al. | Scenewiz3d: Towards text-guided 3d scene composition | |
| Son et al. | Singraf: Learning a 3d generative radiance field for a single scene | |
| CN110322548B (en) | Three-dimensional grid model generation method based on geometric image parameterization | |
| Schirmacher et al. | Efficient Free Form Light Field Rendering. | |
| CN116801115A (en) | Sparse array camera deployment method | |
| CN112002019B (en) | Method for simulating character shadow based on MR mixed reality | |
| Jäger et al. | A comparative Neural Radiance Field (NeRF) 3D analysis of camera poses from HoloLens trajectories and Structure from Motion | |
| Wang et al. | GScream: Learning 3D Geometry and Feature Consistent Gaussian Splatting for Object Removal | |
| Zeng et al. | 3D plants reconstruction based on point cloud | |
| WO2024156209A1 (en) | Security boundary generation method and apparatus, device, storage medium and program product | |
| Tanay et al. | Global Latent Neural Rendering | |
| Hesse et al. | A Quantitative Analysis of Culling Techniques for Real-time Rendering of Digital Elevation Models | |
| Westerteiger | Virtual Reality Methods for Research in the Geosciences |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |