CN109089115A - The adaptive QP compensation and CU high-speed decision of 360 degree of Video codings - Google Patents
The adaptive QP compensation and CU high-speed decision of 360 degree of Video codings Download PDFInfo
- Publication number
- CN109089115A CN109089115A CN201810567705.XA CN201810567705A CN109089115A CN 109089115 A CN109089115 A CN 109089115A CN 201810567705 A CN201810567705 A CN 201810567705A CN 109089115 A CN109089115 A CN 109089115A
- Authority
- CN
- China
- Prior art keywords
- ctu
- depth
- weight
- column
- numerical value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/119—Adaptive subdivision aspects, e.g. subdivision of a picture into rectangular or non-rectangular coding blocks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/103—Selection of coding mode or of prediction mode
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/42—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by implementation details or hardware specially adapted for video compression or decompression, e.g. dedicated software implementation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
- H04N19/61—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding in combination with predictive coding
Abstract
A method of 360 degree of videos are encoded in efficient video coding (HEVC), the method includes CU high-speed decision and QP adaptive equalization are respectively used to CU in frame quickly to divide and Video compression.Wherein, the CU high-speed decision includes following operation: dividing type to CTU according to CTU mean depth numerical value and classifies, obtain the three CTU classified types adjacent with current CTU, it puts it into initialization list and is sorted from small to large, take list median as current CTU depth prediction type according to spatial correlation;The QP adaptive equalization includes following operation: based on the weight of WSPSNR, using each CTU in a column as basic unit, calculating the weighted mean of each CTU and their summation, the QP offset of every a line CTU is finally calculated according to formula.
Description
Joint study
The application obtains following fund assistance by North China University of Tech and the joint study of Beijing Jiaotong University's information institute:
State natural sciences fund (No.61103113, No.60903066), Beijing institution of higher education talent directly under the jurisdiction of a municipal government teach by force in-depth plan item
Mesh (PHR201008187);Jiangsu Province's Natural Science Fund In The Light (BK2011455), Beijing's Natural Science Fund In The Light
(No.4172020), Ministry of Education beginning teacher fund (No. 20090009120006);973 plan (2012CB316400) of country,
Central colleges and universities' basic research fund (No.2011JBM214).
Technical field
It is the present invention relates to a kind of video coding processing method, in particular to a kind of for the adaptive of 360 degree of Video codings
QP compensation and CU high-speed decision algorithm.
Background technique
In VR system, multiple cameras capture continually changing 360 degree of real-world scenes, then use image rendering
To restore scene.Collected spherical surface information by rendering after, people using head-mounted display can be on the spot in person it is glad
Appreciate video.Therefore 360 videos need to have high frame per second and high-resolution, this brings memory capacity and transmission bandwidth very big
Challenge, so 360 videos need more efficient compression algorithm.At present about the compression algorithm of 360 Video codings mainly by
It is formed in terms of CU Fast Segmentation and PU model selection two.In terms of CU Fast Segmentation, mainly from CU recursive subdivision in advance in
Only optimized with the prediction of depth bounds;In PU model selection, mainly from the mode of 35 kinds of prediction modes and MPM this
It is optimized in the two.
Summary of the invention
It is compensated and CU high-speed decision algorithm, institute according to the adaptive QP on one side, providing a kind of 360 degree of Video codings
The method of stating includes that adaptive QP backoff algorithm and CU division stop in advance, wherein algorithm steps are as follows:
The weight of WS-PSNR is suitable for ERP format, therefore based on the weight of WS-PSNR, in one column of frame lining
Each CTU be basic unit, calculate the average of weight sum contained by each CTU in a column, be denoted as WCtu(n), wherein n is a column
Then the weight of all CTU and average in one column are added up, obtain each CTU average sum in a column by the number of middle CTU
Summation is denoted as WAlways, the weighted value of every a line CTU is finally revised as WCtu(n)*n/WAlways;
The type that CTU is divided is analyzed, is classified according to mean depth numerical value;
The mean depth for obtaining 3 adjacent CTU (Left, LeftTop, Above) of current CTU, according to the classification
Formula obtains its corresponding predetermined depth range, obtains corresponding DR numerical value according to table;
Three DR numerical value are put into the sequence in the array that one is initialized to { 5,5,5,1,5 } from small to large, are fetched
Predetermined depth type of the DR numerical value in group middle position as current CTU, obtains its predetermined depth range according to table;
According to the characteristic of ERP format and the characteristic distributions of its weight, i.e. equator weight is 1 and successively decreases to two-stage until being
0, the judgment condition of modification CTU predetermined depth type appropriate relaxes the judgment condition in two-stage region, the judgement of remaining region
Condition remains unchanged, and further improves compression ratio.
The invention proposes the Video Codecs using the above method or device according to another aspect,.
According on the other hand, the invention proposes a kind of computer program product, it includes instruction, described instruction when by
When processor executes, the above method is executed.
Detailed description of the invention
Fig. 1 shows algorithm overview flow chart.
Fig. 2 shows CTU to stop algorithm determination step in advance.
Fig. 3 shows the division type according to CTU mean depth.
Specific embodiment
Various schemes are described referring now to the drawings.In the following description, it in order to explain, elaborates multiple specific thin
Section is in order to provide the thorough understanding to one or more schemes.It may be evident, however, that also can without these specific details
Enough realize these schemes.
As used in this specification, term " component ", " module ", " system " etc. are intended to refer to related to computer
Entity, such as, but not limited to, hardware, firmware, the combination of hardware and software, software or software in execution.For example,
Component can be but not limited to: process, processor, object, the executable (executable), execution run on a processor
Thread, program, and/or computer.For example, the application program and the calculating equipment run on the computing device is ok
It is component.One or more components can be located in executive process and/or execution thread, and component can be located at a meter
On calculation machine and/or it is distributed on two or more computers.In addition, these components can from have be stored thereon it is each
The various computer-readable mediums of kind data structure execute.Component can be communicated by means of locally and/or remotely process, example
If basis has the signal of one or more data groupings, for example, from by means of signal and local system, distributed system
In another component interaction and/or with interacted by means of signal with other systems on the network of such as internet etc one
The data of a component.
Present invention is primarily concerned with the adaptive QP compensation and CU high-speed decision for 360 degree of video high efficient codings.
1. adaptive QP compensation
The characteristic of virtual reality video being projected in whole image with Non uniform sampling.Rectangles are waited for example, having
(ERP) the virtual reality video projected has the sampling of more crypto set in two-stage region, this makes encoder need to spend more
Bit go to describe.It is adjusted herein according to the sampling density situation of virtual reality video projection, is guaranteeing that distortion is not rapid
Compression efficiency is improved in the case where change.Because of the characteristic of ERP projective transformation, poles region has higher relative to equatorial zone
Sampling density, adaptive QP coding method by rougher QP be applied to closer to poles region block.
However, in order to control the video quality in two-stage region, adaptive QP cannot be excessively thick for the optimization of poles region
It is rough.Herein by analysis WS-PSNR weight w, research show that WS-PSNR weight w is suitable for ERP projection scheme.With existing method
It compares, area can reduce the QP numerical value of this paper under the line, and will increase in polar region region, that is to say, that the method can be more smart
The more concerned image of equatorial zone and the rough image for dividing poles region and seriously being stretched really are divided, rather than simply
Quality is sacrificed by increasing QP to reduce bit.It non-uniform may be distributed on virtual reality video in view of distortion, such as box
Shown in 502, we correct the QP value for coding by updating following formula:
QPnew=QP-3 × log2(w)
The weight used for ERP, WS-PSNR is as follows:
Wherein N is the height of CTU, and wherein j indicates the height (range from 0 to picture altitude) of location of pixels.Herein with WS-
The weight w of PSNR is that core is handled, and weight calculation is carried out as unit of CTU, steps are as follows for calculating:
The number that Num_Of_Ctu_Height is CTU in a column is defined, weight sum in each CTU in one column of calculating
Average value:
Wherein Index is the serial number of CTU in a column, and value is [0, Num_Of_Ctu_Height], and N is the height of CTU;
The summation of all CTU weighted means in a column in the first step is calculated, is defined as:
Weight is as follows after modifying herein:
QPnew=QP-3 × log2(w)
2.CU high-speed decision
All there is very strong correlation, and each CTU grades of coding list in video between the pixel of each frame image
There is also very strong correlations between member.Therefore being averaged for adjacent C TU (upper left CTU, upper CTU, left side CTU) can be used
Depth predicts that current CTU depth bounds, flow chart are as shown in Figure 2 by classification processing.
Firstly, we classify to the dividing condition of CTU.As shown in figure 3, the value above CTU after each division is
Its mean depth, the section on the left side every row CTU include all depth that row CTU is divided, such as if it is current divide after CTU
Mean depth belongs to the third line, then we determine that the depth contained by it is 1,2,3.When the mean depth of CTU is less than or equal to 1
When, predetermined depth range of CTU is [0,1].When the mean depth of CTU is equal to 1.25,1.5 or 1.75, the pre- depth measurement of CTU
Spending range is [1,2].When the mean depth of CTU is greater than or equal to 1.3125 and is less than or equal to 2.4375, wherein not including
Depth 1.5 and 1.75, predetermined depth range of CTU are [1,3], when the mean depth of CTU is greater than 2.0625, the prediction of CTU
Depth bounds are [2,3].For example, if the mean depth of current CTU is greater than 2.0625, it may be only comprising deep in CTU division
Degree 2 and 3.
2 DR value of table and range
DR type | DR numerical value | Predetermined depth range [Min, Max] |
T1 | 0 | [0,1] |
- | 1 | [0,2] |
T2 | 2 | [1,2] |
T3 | 3 | [1,3] |
T4 | 4 | [2,3] |
- | 5 | [0,3] |
According to table 2, predetermined depth range that CTU is divided can be divided into four classes, respectively T1, T2, T3 and T4.Each point
Class all includes that it corresponds to all depth within the scope of predetermined depth.For example, prediction should if CTU depth type belongs to T2
CTU only includes depth 1 and 2 after dividing.When predetermined depth range of CTU is [0,1], depth type belongs to T1.Work as CTU
Predetermined depth range when being [1,2], depth type belongs to T2.When predetermined depth range of CTU is [1,3], depth
Type belongs to T3.When predetermined depth range of CTU is [2,3], depth type belongs to T4.Wherein, when CTU depth type is
When T3 and T4, their predetermined depth section is overlapping.In order to accurately divide, we define the average depth of CTU after dividing
When degree section is [2.0625,2.4375], we conclude its depth type to type T3;CTU mean depth after division
Type T4 is concluded when greater than 2.4375, as shown in formula (7).
In view of the non-uniform characteristic of ERP Projection Sampling, and its projection is formed by two-dimensional surface close to two-stage region
Weight close to 0, therefore can by modification decision condition appropriate, in the case where guaranteeing rate distortion costs further plus
It is fast to determine speed.Pass through experimental test herein, discovery is worked as in the region of weight w < 0.4 (w is exactly the weight in WS-PSNR),
Decision condition [1.3125,2.437] can be changed to [1.3125,2], while Depth > 2.4375 is changed to Depth > 2, i.e.,
The decision condition in the mean depth section [2,2.4375] of CTU is become T4 from T3 before, as shown in formula (8).Purpose is
Reducing depth is 1 traversal.Then the Rule of judgment of Depth=2 is added in predetermined depth range [1,2], it is therefore an objective to reduce deep
The traversal of degree 3, but can also introduce mistake.It is contemplated that the weight coefficient in two-stage region is lower, and here only for weight w <
0.4 region advanced optimizes.It is counted by test data, can guarantee that rate distortion costs are changed less and further speeded up
The judgement of CU depth.
This paper high-speed decision algorithm: according to the CTU dividing condition concluded, by the mean depth of 3 adjacent C TU according to
Table 2 and Fig. 3 obtain corresponding type, then determine DR numerical value according to mapping relations and table 2, then by DR numerical value be put into list from
It is small to longer spread and to take DR value among list, correspond to predetermined depth range of predetermined depth range as current CTU.But it should
Algorithm has a problem that: probably due to narrow section and cause to judge by accident.Such as: the depth bounds by obtaining adjacent C TU obtain
DR list is [0,0,2], and taking intermediate DR numerical value at this time is 0, is determined as T1 type by table 2 and mapping relations.But full traversal
Its DR type is really T2 after current CTU, this is the erroneous judgement as caused by narrow section, therefore the algorithm needs to add two DR values (1
With 5), corresponding forecast interval is as shown in table 2.List at this time then becomes [0,0,1,2,5] after arrangement, takes in-between
DR numerical value 1, the in this way algorithm determine that its estimation range is [0,2], contain [1,2] this section.So addition 1 energy of DR value
Reduce False Rate to a certain extent.When in list there are three numerical value be 0 when, show that current CTU is in flat site mostly,
Therefore a possibility that being type T1, is very big.Similarly, increasing DR=5 is the processing higher region of complexity, reduces the mistake of T4 type
Sentence.Therefore, DR number list is initialized as [1,5,5,5,5] herein, is repaired by traversing the depth intervals of adjacent CTU
Just, estimation range of the corresponding forecast interval of intermediate value DR of list as current CTU depth is taken.
The above embodiment of the present invention can all be realized as the encoder based on HEVC.In the encoder based on HEVC
Portion's structure can be as shown in Figure 1.It should be appreciated by those skilled in the art that the decoder can be implemented as software, hardware and/or consolidate
Part.
When implemented in hardware, video encoder can use general processor, digital signal processor (DSP), dedicated collection
At circuit (ASIC), field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device
Part, discrete hardware components are designed as executing any combination thereof of function described herein, to realize or execute.General processor
It can be microprocessor, but alternatively, which is also possible to any conventional processor, controller, microcontroller
Or state machine.Processor also can be implemented as calculating the combination of equipment, for example, the combination of DSP and microprocessor, multiple micro- places
Manage the combination of device, the combination or any other such structure of one or more microprocessors and DSP core.In addition, at least one
A processor may include that can operate to execute one or more modules of above-mentioned one or more steps and/or operation.
When with hardware circuits such as ASIC, FPGA to realize video encoder, may include be configured as executing it is various
The various circuit blocks of function.Those skilled in the art can be according to the various constraint conditions applied over the whole system come with various
These circuits of mode design and implementation, to realize various functions disclosed in this invention.
Although aforementioned open file discusses exemplary arrangement and/or embodiment, it should be noted that being wanted without departing substantially from by right
In the case where the scheme for the description for asking book to define and/or the range of embodiment, many change and modification can be made herein.And
And although in the singular describe or require the scheme and/or embodiment element, it is also contemplated that plural number feelings
Condition is limited to odd number unless expressly stated.In addition, all or part of any scheme and/or embodiment can with it is any its
Its scheme and/or all or part of of embodiment are used in combination, unless showing different.
Claims (7)
1. a kind of method for being used to encode 360 degree of videos at efficient video coding (HEVC), it is fast that the method includes CU
Quick decision plan and QP adaptive compensation algorithm, wherein
The QP adaptive equalization, for solving the problems, such as that two-stage sampling point distributions are uneven in ERP format, using CTU as unit, with
Based on the weight of WS-PSNR, adaptive compensation is done by QP parameter of the formula to a line CTU every in a frame, improves compression
Rate;
The CU high-speed decision is divided for reducing unnecessary CU unit depth in cataloged procedure, according to mean depth numerical value
Classify to CTU classified types, obtains the three CTU classified types adjacent with current CTU, put it into initialization list simultaneously
It is sorted from small to large, according to spatial correlation, takes list median as the type of prediction of current CTU, closed according to corresponding
System obtains predetermined depth section of current CTU, improves and divides processing speed in frame.
2. algorithm according to claim 1, characterized in that described that CU high-speed decision algorithm is asked to specifically include:
The case where CTU is divided first is classified according to the numerical value of mean depth, obtains 3 adjacent CTU of current CTU
The mean depth of (Left, LeftTop, Above) simultaneously obtains affiliated type according to classification, obtains it further according to affiliated type correspondence
Resulting three DR numerical value is put into the array that one is initialized as { 5,5,5,1,5 } and arranges from small to large, arranges by DR numerical value
DR value after column among access group, according to spatial correlation, depth type corresponding to the DR value is the pre- depth measurement of current CTU
Degree.
3. algorithm according to claim 1, characterized in that the QP adaptive compensation algorithm asked specifically includes:
The weight used for ERP, WS-PSNR is as follows:
Wherein N is the height of CTU, and j indicates the height (range from 0 to picture altitude) of location of pixels, herein with the power of WS-PSNR
Weight w is that core is handled, and weight calculation is carried out as unit of CTU, steps are as follows for calculating:
The number that Num_Of_Ctu_Height is CTU in a column is defined, being averaged for weight sum in each CTU is calculated in a column
Value:
Wherein Index is the serial number of CTU in a column, and value is [0, Num_Of_Ctu_Heigtt], and N is the height of CTU;
The summation of all CTU weighted means in a column in the first step is calculated, is defined as:
Weight is as follows after modifying herein:
QPnew=QP-3 × log2(wnew)
4. a kind of Video Codec for realizing method or apparatus of any of claims 1-3.
5. a kind of computer program product for requiring method described in any one of 1-3 for perform claim.
6. a kind of encoded video stream for the device code that the method according to any one of perform claim requirement 1-3 generates.
7. a kind of Video Decoder is generated or is filled for decoding the method according to any one of perform claim requirement 1-3
Set the encoded video stream of coding.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810567705.XA CN109089115B (en) | 2018-05-28 | 2018-05-28 | Method for encoding 360-degree video in HEVC |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810567705.XA CN109089115B (en) | 2018-05-28 | 2018-05-28 | Method for encoding 360-degree video in HEVC |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109089115A true CN109089115A (en) | 2018-12-25 |
CN109089115B CN109089115B (en) | 2022-07-01 |
Family
ID=64839360
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810567705.XA Active CN109089115B (en) | 2018-05-28 | 2018-05-28 | Method for encoding 360-degree video in HEVC |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109089115B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109889829A (en) * | 2019-01-23 | 2019-06-14 | 北方工业大学 | The point self-adapted compensation of the quick sample product of 360 degree of videos |
CN112055211A (en) * | 2019-08-15 | 2020-12-08 | 华为技术有限公司 | Video encoder and QP setting method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140254670A1 (en) * | 2013-03-07 | 2014-09-11 | Samsung Electronics Co., Ltd. | Method and apparatus for roi coding using variable block size coding information |
US20140334555A1 (en) * | 2011-12-15 | 2014-11-13 | Thomson Licensing | Method and apparatus for video quality measurement |
GB201710568D0 (en) * | 2017-06-30 | 2017-08-16 | Canon Kk | 360-Degree video encoding with block-based extension of the boundary of projected parts |
CN107094251A (en) * | 2017-03-31 | 2017-08-25 | 浙江大学 | A kind of video, image coding/decoding method and device adjusted based on locus adaptive quality |
-
2018
- 2018-05-28 CN CN201810567705.XA patent/CN109089115B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140334555A1 (en) * | 2011-12-15 | 2014-11-13 | Thomson Licensing | Method and apparatus for video quality measurement |
US20140254670A1 (en) * | 2013-03-07 | 2014-09-11 | Samsung Electronics Co., Ltd. | Method and apparatus for roi coding using variable block size coding information |
CN107094251A (en) * | 2017-03-31 | 2017-08-25 | 浙江大学 | A kind of video, image coding/decoding method and device adjusted based on locus adaptive quality |
GB201710568D0 (en) * | 2017-06-30 | 2017-08-16 | Canon Kk | 360-Degree video encoding with block-based extension of the boundary of projected parts |
Non-Patent Citations (2)
Title |
---|
HENDRY等: "AHG8: Adaptive QP for 360º video ERP projection", 《JOINT VIDEO EXPLORATION TEAM (JVET) OF ITU-T SG 16 WP 3 AND ISO/IEC JTC 1/SC 29/WG 11 6TH MEETING》 * |
YULE SUN等: "AHG8: Stretching ratio based adaptive quantization for 360 video", 《JOINT VIDEO EXPLORATION TEAM (JVET) OF ITU-T SG 16 WP 3 AND ISO/IEC JTC 1/SC 29/WG 11 6TH MEETING》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109889829A (en) * | 2019-01-23 | 2019-06-14 | 北方工业大学 | The point self-adapted compensation of the quick sample product of 360 degree of videos |
CN109889829B (en) * | 2019-01-23 | 2022-08-09 | 北方工业大学 | Fast sample adaptive compensation for 360 degree video |
CN112055211A (en) * | 2019-08-15 | 2020-12-08 | 华为技术有限公司 | Video encoder and QP setting method |
Also Published As
Publication number | Publication date |
---|---|
CN109089115B (en) | 2022-07-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111882002B (en) | MSF-AM-based low-illumination target detection method | |
Cao et al. | Total variation regularized tensor RPCA for background subtraction from compressive measurements | |
CN110363716B (en) | High-quality reconstruction method for generating confrontation network composite degraded image based on conditions | |
DE102018108324A1 (en) | System and method for estimating an optical flow | |
CN103188493B (en) | Image encoding apparatus and image encoding method | |
CN111739082B (en) | Stereo vision unsupervised depth estimation method based on convolutional neural network | |
CN104899921B (en) | Single-view videos human body attitude restoration methods based on multi-modal own coding model | |
CN109903315B (en) | Method, apparatus, device and readable storage medium for optical flow prediction | |
Luan et al. | Fast single image dehazing based on a regression model | |
CN109191498B (en) | Target detection method and system based on dynamic memory and motion perception | |
CN104685538A (en) | Systems and methods for reducing noise in video streams | |
CN113870335A (en) | Monocular depth estimation method based on multi-scale feature fusion | |
CN103927734B (en) | A kind of based on the broad image quality evaluating method without reference | |
CN110738663A (en) | Double-domain adaptive module pyramid network and unsupervised domain adaptive image segmentation method | |
CN111931603B (en) | Human body action recognition system and method of double-flow convolution network based on competitive network | |
CN103313047A (en) | Video coding method and apparatus | |
CN113095277B (en) | Unmanned aerial vehicle aerial photography vehicle detection method based on target space distribution characteristics | |
CN106664404B (en) | Block partitioning scheme processing method and relevant apparatus in Video coding | |
CN116012950A (en) | Skeleton action recognition method based on multi-heart space-time attention pattern convolution network | |
CN109089115A (en) | The adaptive QP compensation and CU high-speed decision of 360 degree of Video codings | |
CN110033483A (en) | Based on DCNN depth drawing generating method and system | |
CN109871790B (en) | Video decoloring method based on hybrid neural network model | |
CN116205962A (en) | Monocular depth estimation method and system based on complete context information | |
CN112200752B (en) | Multi-frame image deblurring system and method based on ER network | |
Zhang et al. | Spatial-information Guided Adaptive Context-aware Network for Efficient RGB-D Semantic Segmentation |
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 |