CN107071424A - A kind of load-balancing method based on scramble time forecast model - Google Patents
A kind of load-balancing method based on scramble time forecast model Download PDFInfo
- Publication number
- CN107071424A CN107071424A CN201710159343.6A CN201710159343A CN107071424A CN 107071424 A CN107071424 A CN 107071424A CN 201710159343 A CN201710159343 A CN 201710159343A CN 107071424 A CN107071424 A CN 107071424A
- Authority
- CN
- China
- Prior art keywords
- frame
- mrow
- load
- time
- band
- 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
- H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
- H04N19/156—Availability of hardware or computational resources, e.g. encoding based on power-saving criteria
-
- 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
- H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
- H04N19/593—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving spatial prediction techniques
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Computing Systems (AREA)
- Theoretical Computer Science (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a kind of load-balancing method based on scramble time forecast model, belong to multimedia coding field, this method to video flowing before parallel encoding is carried out, the scramble time of current video frame each coding unit is predicted according to scramble time forecast model first, the equalization operation then loaded using the predictive coding time of present frame each coding unit for foundation;Load balancing automatically determines the average load of each core according to the configuration of current device first primarily with respect to the distribution of load, and certain coding unit number is then distributed for each core according to the scramble time of prediction;The load balancing of HEVC parallel encoders is realized finally by the adjustable strategies of load.The load-balancing method of HEVC parallel encoders proposed by the present invention based on scramble time forecast model can significantly improve coding rate on the premise of coding quality is ensured.
Description
Technical field
The invention belongs to multimedia coding field, and in particular to a kind of load balancing side based on scramble time forecast model
Method.
Background technology
With video encoding standard HEVC of new generation (High Efficiency Video Coding) issue, video
Code efficiency is further lifted.Compared to former video encoding standard H.264/AVC, in the case of identical coding quality,
HEVC can save the code check of nearly half, but this is to sacrifice encoder complexity as cost.How code stream knot is not being changed
On the premise of structure and the original code check of holding, algorithm complex is farthest reduced, coding rate is significantly improved, is the big rule of HEVC
Mould promotes the key point with application.
In order to reduce HEVC computation complexity, and it is applied to real-time application, many scholars are devoted to this side
To.The optimization method of reduction computation complexity is broadly divided into two kinds:Serial optimization and parallel optimization.
Serial optimization is broadly divided into two classes, a class is by coding unit primarily directed in the optimization of coding flow
Quick to divide the quick selection with coding mode to reduce computation complexity, a class is dropped by reducing the time of estimation
Low computation complexity.The effect of this two class optimization is it will be evident that by the optimization of these coding flows, can about reduce HEVC
Close to the computation complexity of half.But, with the development of the society, people have higher requirement for the definition of video.
Only the computation complexity of reduction half still can not meet demand of the people for ultra high-definition video.Then occurred as soon as on
The research of parallel optimization.
Parallel optimization is also classified into many classes, from small to large main parallel, frame level parallel, band including image sets level or
Tile grades parallel, and code tree cell level are parallel.Image sets level is parallel due to needing caching is substantial amounts of to regard with frame level
Frequency evidence, time delay is big, is not suitable for application in real time;Correlation between the parallel tree unit due to adjacent encoder of code tree cell level
Relatively strong, there is very big limitation in parallel encoding, and parallel efficiency is not high.Band and tile are the collection of the code tree unit in a frame
Close, without time delay and restriction, and correlation limitation is little.But the division of band is more flexible compared with tile, and with more preferable
Parallel performance.Therefore, this patent is other parallel primarily directed to slice level.
The method about band parallel optimization mainly has two kinds at present, and a kind of is the scramble time according to adjacent encoded frame
To predict the scramble time of present frame, a kind of is to set up a kind of normalized computation complexity to determine that the coding of present frame is complicated
Degree.The first scheme only only used the information of adjacent encoded frame, and have ignored separation time coding structure, second scheme
The normalized computation complexity set up then due to the texture features and kinetic characteristic of image are dissimilar and precision of prediction is poor, no
Adapt in all video sequences.Therefore we, which need badly, finds one kind and can use separation time coding structure and adjacent simultaneously
The method of encoded frame, preferably predicts the computation complexity of the code tree unit of each in present frame.
The content of the invention
In order to find the load-balancing method that a kind of significantly more efficient slice level is parallel, the present invention proposes a kind of based on volume
The load-balancing method of code time prediction model, it is reasonable in design, the deficiencies in the prior art are overcome, with good effect.
To achieve these goals, the present invention is adopted the following technical scheme that:
A kind of load-balancing method based on scramble time forecast model, comprises the following steps:
Step 1:Into a frame video flowing;
Step 2:Averagely divide the code tree number of unit that each band has in present frame;
Step 3:Whether judge present frame is intracoded frame;
If:Judged result is that present frame is intracoded frame, then performs step 10;
Or judged result is that present frame is not intracoded frame, then step 4 is performed;
Step 4:Judge whether present frame has the encoded non-intracoded frame of same time horizon;
If:Judged result is the encoded non-intracoded frame that present frame has same time horizon, then performs step 5;
Or judged result is that present frame does not have the encoded non-intracoded frame of same time horizon, then step 7 is performed;
Step 5:The scramble time of the adjacent encoded frame in position is synchronized to and same a period of time by scramble time forecast model
In the adjacent encoded frame identical level of interbed;
Step 6:The scramble time of present frame is predicted by the scramble time of two frames after synchronization jointly, step is then performed
Rapid 8;
Step 7:The scramble time of present frame is predicted according to the scramble time of the adjacent encoded frame in position;
Step 8:The code tree unit that each band is possessed is distributed on the basis of the predictive coding time of present frame
Number;
Step 9:It is finely adjusted according to the allocation result of band, each band is obtained identical computational load;
Step 10:Each band of parallel encoding present frame, terminates a frame.
Preferably, in steps of 5, following steps are specifically included:
Step 5.1:Adjacent encoded frame on position is done into the adjustment based on its position in the group of images, it is synchronized to
With same time horizon in adjacent encoded frame identical image sets position level;
Step 5.2:The scramble time of adjacent encoded frame on position is done into the adjustment based on quantization parameter size, makes it
It is synchronized to same time horizon in adjacent encoded frame identical quantization parameter level;
Step 5.3:The scramble time of adjacent encoded frame on position is done based on the adjustment of reference frame number purpose, makes its same
Walk with same time horizon in adjacent encoded frame identical reference frame level.
Preferably, in step 8, following steps are specifically included:
Step 8.1:Calculate total predictive coding time of present frame;
Step 8.2:The number of band is distributed according to the CPU core calculation of each equipment;
Step 8.3:The average load of each band is determined by the result of step 8.1 and step 8.2;
Step 8.4:It is the code tree unit that each band distributes certain amount according to the average load in step 8.3.
Preferably, in step 9, following steps are specifically included:Step 9.1:Into a frame video flowing;
Step 9.2:Predict the total encoding time of present frame and the scramble time of each code tree unit;
Step 9.3:Calculate the average load of each band and divide the code tree number of unit of each band;
Step 9.4:The load and sequence of each band after computation partition,
Step 9.5:Max calculation load is recorded, and is assigned to a variable a;
Step 9.6:Judge whether a load value is more than the computational load before or after it;
If:Judged result is that a load value is more than the computational load before or after it, then performs step 9.7;
Or judged result is that a load value is less than the computational load before or after it, then performs step 9.11;
Step 9.7:A code tree number of unit is subtracted one, the code tree number of unit of smaller band adds one;
Step 9.8:Calculate the maximum load after change;
Step 9.9:Judge the max calculation after change loads the maximum load before whether being less than;
If:Judged result loads the maximum load before being less than for the max calculation after change, then performs step 9.10;
Or judged result loads the maximum load before being more than for the max calculation after change, then step 9.11 is performed;
Step 9.10:Resequence the computational load of each band, then perform step 9.5;
Step 9.11:Whether judge a is minimum load;
If:Judged result is that a is minimum load, then terminates;
Or judged result is that a is not minimum load, then step 9.12 is performed;
Step 9.12:A is entered as secondary small computational load, step 9.6 is then performed;
Preferably, scramble time forecast model is in step 5:
Tpred=a*Ttemp+b*β*Tpos*
In formula, TpredIt is final predicted time, TposIt is the scramble time of position consecutive frame, qpposIt is position consecutive frame
Quantization parameter, refposIt is the reference frame number of position consecutive frame, a is the weight of position consecutive frame;TtempIt is that time horizon is adjacent
The scramble time of frame, qptempIt is the quantization parameter of time horizon consecutive frame, reftempIt is the reference frame number of time horizon consecutive frame, b
It is the weight of time horizon consecutive frame;α is that we calculate obtained changed factor from statistics, and β is to be based on compiling among image sets
The position of code frame calculates obtained ratio.
The advantageous effects that the present invention is brought:The present invention is configured from coding configuration by studying three kinds of codings
Influence degree of the parameter for the scramble time, it is proposed that three kinds of Adjusted Options based on coding parameter, and integrated as one
Plant scramble time forecast model.
The present invention is according to above-mentioned scramble time forecast model, by the adjacent encoded frame on position and same time horizon
Adjacent encoded frame combines consideration, predicts the scramble time of present frame jointly using both encoded frames, establishes one
The load-balancing method based on scramble time forecast model is planted, this method can be on the premise of coding quality be ensured, well
High definition, the Real Time Compression application of ultra high-definition video are adapted to, coding rate is substantially increased.
Brief description of the drawings
Fig. 1 is the flow chart of the load-balancing method of the parallel HEVC encoders of band based on scramble time forecast model.
Fig. 2 is the flow chart of the scramble time Forecasting Methodology based on coding parameter.
Fig. 3 is the particular flow sheet of load-balancing method.
Embodiment
The present invention is further illustrated with reference to the accompanying drawings and detailed description.
In order to meet high definition, the Real Time Compression application demand of ultra high-definition video image, the volume of HEVC video encoders is improved
Code speed, the present invention proposes a kind of load-balancing method of the parallel HEVC encoders of band based on coding forecast model.Should
Method mainly includes two parts:Scramble time method of adjustment based on coding parameter and the load based on scramble time forecast model
Equalization methods, the scramble time method of adjustment based on coding parameter is related to step (5) and step (6) in the content of the invention, is based on
The load-balancing method of scramble time forecast model is related to step (8) and step (9) in the content of the invention.
The main flow of the invention as shown in figure 1, first determine whether current video stream whether intracoded frame, if having
The adjacent encoded frame of same time horizon, is not intracoded frame when meeting, and the adjacent encoded frame with same time horizon
When, according to the configuration parameter of the adjacent encoded frame (referred to as time horizon consecutive frame) of same time horizon, by position
Adjacent encoded frame (referred to as position consecutive frame) is done based on it in image sets position, quantization parameter size and reference frame number respectively
Purpose is adjusted;Then according to scramble time forecast model, the position consecutive frame with time horizon consecutive frame and after being adjusted is common
Predict scramble time and the total encoding time of each code tree unit of present frame;For determining band is configured further according to computer
Number, and determine therefrom that the average load of each band;Finally according to the predicted time and each band of code tree unit
Average load determines the code tree number of unit of each band, and is finely adjusted on this basis, so that each
Band obtains identical computational load, to reach load balancing that band is parallel.
It developed below and illustrates:
1st, the scramble time method of adjustment based on coding parameter
Fixed coding configuration template is occurred in that in video encoding standard HEVC of new generation, it is main to include three kinds:In full frame
Coded frame configuration (AI), low time delay configuration (LP) and arbitrary access configuration (RA).There is similar volume under each configuration file
Code configuration, but coding configuration in there is number of values to change, these change configuration parameters often to coding when
Between and coding efficiency produce influence.These different configuration parameters, which have also had a strong impact on, simultaneously joins two distinct types of frame
It is used for the degree of accuracy of actual prediction altogether.Therefore, in order to simultaneously using two kinds of different encoded frames, it would be desirable to study this
Influence of a little coding parameters for the scramble time.
1) position in image sets
Found by experiment statisticses, remaining frame can be significantly greater than in the presence of a scramble time in each image sets
Frame, after the size of image sets is changed, is as a result still so, this frame is exactly last in each image sets
Frame.Found by studying, the reason for scramble time of this frame is seriously more than the scramble time of remaining frame in this image sets be, this
The time horizon ID of one frame is that 0, λ is smaller.In addition, research shows in same video sequence, the last frame in image sets
Scramble time is similar in each image sets to the ratio of the scramble time of remaining frame.Therefore, we define one
Individual ratio β, shown in calculation formula such as following formula (1).
β=Tother÷TGOPsize (1)
In formula, TGOPsizeIt is the scramble time of last frame in image sets, TOtherWhen being the coding of other frames in image sets
Between.By using this ratio, the coded frame of diverse location level in image sets can be synchronized in a position level.
2) quantization parameter size
In order to study influence level of the quantization parameter size for the scramble time, six are have selected in HEVC cycle tests
Individual cycle tests is as training set, in being respectively the ParkScene and Cactus, C class testing sequence in B class testing sequences
BasketballPass and RaceHorses in BQMall and PartyScene and D class testing sequences.In reference frame number
For 1, gop size is 4, and under conditions of remaining configuration parameter is default configuration, by changing quantization parameter, (scope is 20-
The quantization parameter scope commonly used in 48, that is, Video coding) size record the scramble time of above-mentioned six cycle tests.
Each situation is all tested 3 times, is used as the scramble time of each frame finally by averaging.
Experimental data shows:There is the rule as shown in following formula (2) between scramble time and quantization parameter size.
In formula,It is the scramble time for needing to predict,It is to have known and by the time for prediction,
QPpredIt is the quantization parameter for the frame for needing prediction, QPreferenceIt is the quantization parameter of the frame for prediction.By using this rule
Rule, can be by a kind of frame synchronization with different quantization parameter sizes to quantization parameter level.
3) reference frame number
Estimation is operation most time-consuming in HEVC inter predictions, when a reference frame is increased, accordingly
The estimation of a frame will be done more.In order to study influence of the reference frame number to the scramble time, during 2) we still select
6 used cycle tests are every by only changing under the conditions of all configuration parameters of holding are all immovable as training set
The number of the reference frame of one frame tests each above-mentioned cycle tests, and records the scramble time of each frame, wherein, reference frame
Number be each set to 1,2,3 and 4.After test 4 times, average as last test result.
Statistics shows that the relation between scramble time and reference frame number can be described with following formula (3).
In formula, i, j is the number of reference frame,It is the scramble time for needing to predict, its number for possessing reference frame is j,It it is the scramble time for prediction, it possesses reference frame number for i, and α is that the changed factor drawn is analyzed from data.
By above-mentioned achievement, we have proposed a kind of scramble time forecast model based on coding parameter, its formula is as follows
Shown in formula (4).
In formula, TpredIt is final predicted time, TposIt it is the scramble time of position consecutive frame, its quantization parameter is qppos,
Reference frame number is refpos, weight is a;TtempIt it is the scramble time of time horizon consecutive frame, quantization parameter is qptemp, reference frame
Number is reftemp, weight is b;α be 3) in changed factor, β be 1) in ratio.
The flow of scramble time method of adjustment based on coding parameter is as shown in Figure 2.To two kinds of volumes accessed by us
The scramble time of code frame, both are synchronized in the position level of same image sets first.Then to the position phase after adjustment
The scramble time of adjacent frame carries out the adjustment based on quantization parameter and the adjustment based on reference frame respectively.So as to which we can obtain two
The individual frame in Unified coding level.The process description of adjustment is as follows:
(1) to present frame, we can obtain two kinds of encoded frames, i.e. time horizon consecutive frame and position consecutive frame.First
Both frames are judged whether in the level of same image sets position, if need not adjust;Otherwise, it is position is adjacent
Frame carries out the adjustment as shown in formula (1).
(2) if both frames are not in same quantization parameter level, we carry out position consecutive frame as shown in formula (2)
Adjustment;Otherwise, it is not necessary to adjust.
(3) if both frames are not in same reference frame number level, we carry out position consecutive frame as shown in formula (3)
Adjustment;Otherwise, it is not necessary to adjust.
By above-mentioned adjustment, we can obtain two different classes of encoded frames in same coding parameter level.
The code time information for the encoded frame being synchronized to by using both in same coding parameter level, we can be pre-
Measure the scramble time of present frame.
2. the load-balancing method based on scramble time forecast model
Load balancing be it is a kind of we can obtain the perfect condition of highest speed-up ratio.But in actual coding, due to
The randomness of natural image and the select permeability of optimal coding mode, the computation complexity distribution of a two field picture are simultaneously uneven.Cause
This, when we evenly distribute coding unit number to each band, it may appear that load imbalance.Load imbalance is reduction
One critically important factor of parallel processor system performance.In order to realize load balancing that band level is parallel, it would be desirable to
Find a kind of method that can effectively predict each code tree unit computation complexity.So, we just can averagely divide
The computational load of each band so that each coding core is completed simultaneously.By doing so it is possible, parallel processing system (PPS) would not
There is idle waiting phenomenon, we will also obtain highest speed-up ratio.
The present invention is obtained each in present frame using the scramble time Forecasting Methodology based on coding parameter above and compiled
The predictive coding time of code unit, and then realize using these information the parallel load balancing of band level.Load distribution and
The flow of adjustment is as shown in figure 3, embodiment is as follows:
(1) total encoding time and each coding for obtaining present frame according to above-mentioned scramble time Forecasting Methodology first are single
The scramble time of member.
(2) number of band is automatically determined according to the configuration of device therefor.
(3) average load of each band is gone out according to the number of band and total predictive coding Time Calculation.
(4) according to the average load of each band and predict come each coding unit predictive coding when
Between, to determine the number for the coding unit that each band should have.
(5) computational load of each band after the number of the coding unit divided according to each band, computation partition
And sort, maximum load is recorded, a variable a is assigned to.
(6) by variable a compared with the load of band adjacent before and after it, and a code tree is shifted to smaller band
Unit;
(7) if the max calculation before the maximum load after change is less than is loaded, or the minimum load after change is more than it
Preceding minimum load, it is determined that this new division, and the sequence of band computational load is updated, maximum load is assigned to variable a;
Perform step (6);Otherwise former divide is kept;
(8) if load a is not minimum, time big computational load is entered as, step (6) is continued executing with;Otherwise start
Coding.
Certainly, described above is not limitation of the present invention, and the present invention is also not limited to the example above, this technology neck
The variations, modifications, additions or substitutions that the technical staff in domain is made in the essential scope of the present invention, should also belong to the present invention's
Protection domain.
Claims (5)
1. a kind of load-balancing method based on scramble time forecast model, it is characterised in that comprise the following steps:
Step 1:Into a frame video flowing;
Step 2:Averagely divide the code tree number of unit that each band has in present frame;
Step 3:Whether judge present frame is intracoded frame;
If:Judged result is that present frame is intracoded frame, then performs step 10;
Or judged result is that present frame is not intracoded frame, then step 4 is performed;
Step 4:Judge whether present frame has the encoded non-intracoded frame of same time horizon;
If:Judged result is the encoded non-intracoded frame that present frame has same time horizon, then performs step 5;
Or judged result is that present frame does not have the encoded non-intracoded frame of same time horizon, then step 7 is performed;
Step 5:The scramble time of the adjacent encoded frame in position is synchronized to and same time horizon by scramble time forecast model
Adjacent encoded frame identical level on;
Step 6:The scramble time of present frame is predicted by the scramble time of two frames after synchronization jointly, step 8 is then performed;
Step 7:The scramble time of present frame is predicted according to the scramble time of the adjacent encoded frame in position;
Step 8:The code tree number of unit that each band is possessed is distributed on the basis of the predictive coding time of present frame;
Step 9:It is finely adjusted according to the allocation result of band, each band is obtained identical computational load;
Step 10:Each band of parallel encoding present frame, terminates a frame.
2. the load-balancing method according to claim 1 based on scramble time forecast model, it is characterised in that in step
In 5, following steps are specifically included:
Step 5.1:Adjacent encoded frame on position is done into the adjustment based on its position in the group of images, make its be synchronized to together
In the adjacent encoded frame identical image sets position level of one time horizon;
Step 5.2:The scramble time of adjacent encoded frame on position is done into the adjustment based on quantization parameter size, makes its synchronization
To with same time horizon in adjacent encoded frame identical quantization parameter level;
Step 5.3:The scramble time of adjacent encoded frame on position is done based on the adjustment of reference frame number purpose, it is synchronized to
With same time horizon in adjacent encoded frame identical reference frame level.
3. the load-balancing method according to claim 1 based on scramble time forecast model, it is characterised in that in step
In 8, following steps are specifically included:
Step 8.1:Calculate total predictive coding time of present frame;
Step 8.2:The number of band is distributed according to the CPU core calculation of each equipment;
Step 8.3:The average load of each band is determined by the result of step 8.1 and step 8.2;
Step 8.4:It is the code tree unit that each band distributes certain amount according to the average load in step 8.3.
4. the load-balancing method according to claim 1 based on scramble time forecast model, it is characterised in that in step
In 9, following steps are specifically included:
Step 9.1:Into a frame video flowing;
Step 9.2:Predict the total encoding time of present frame and the scramble time of each code tree unit;
Step 9.3:Calculate the average load of each band and divide the code tree number of unit of each band;
Step 9.4:The load and sequence of each band after computation partition;
Step 9.5:Max calculation load is recorded, and is assigned to a variable a;
Step 9.6:Judge whether a load value is more than the computational load before or after it;
If:Judged result is that a load value is more than the computational load before or after it, then performs step 9.7;
Or judged result is that a load value is less than the computational load before or after it, then performs step 9.11;
Step 9.7:A code tree number of unit is subtracted one, the code tree number of unit of smaller band adds one;
Step 9.8:Calculate the max calculation load after change;
Step 9.9:Judge that the max calculation after change loads the max calculation load before whether being less than;
If:Max calculation before judged result is less than for the max calculation load after change is loaded, then performs step 9.10;
Or judged result loads the max calculation load before being more than for the max calculation after change, then performs step 9.11;
Step 9.10:Resequence the computational load of each band, then perform step 9.5;
Step 9.11:Whether judge a is minimum load;
If:Judged result is that a is minimum load, then terminates;
Or judged result is that a is not minimum load, then step 9.12 is performed;
Step 9.12:A is entered as secondary small computational load, step 9.6 is then performed.
5. the load-balancing method according to claim 1 based on scramble time forecast model, it is characterised in that step 5
Middle scramble time forecast model is:
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>T</mi>
<mrow>
<mi>p</mi>
<mi>r</mi>
<mi>e</mi>
<mi>d</mi>
</mrow>
</msub>
<mo>=</mo>
<mi>a</mi>
<mo>*</mo>
<msub>
<mi>T</mi>
<mrow>
<mi>t</mi>
<mi>e</mi>
<mi>m</mi>
<mi>p</mi>
</mrow>
</msub>
<mo>+</mo>
<mi>b</mi>
<mo>*</mo>
<mi>&beta;</mi>
<mo>*</mo>
<msub>
<mi>T</mi>
<mrow>
<mi>p</mi>
<mi>o</mi>
<mi>s</mi>
</mrow>
</msub>
<mo>*</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msup>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>+</mo>
<mi>&alpha;</mi>
<mo>)</mo>
</mrow>
<mrow>
<msub>
<mi>ref</mi>
<mrow>
<mi>t</mi>
<mi>e</mi>
<mi>m</mi>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>ref</mi>
<mrow>
<mi>p</mi>
<mi>o</mi>
<mi>s</mi>
</mrow>
</msub>
</mrow>
</msup>
<mo>*</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mfrac>
<mrow>
<msub>
<mi>qp</mi>
<mrow>
<mi>t</mi>
<mi>e</mi>
<mi>m</mi>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>pq</mi>
<mrow>
<mi>p</mi>
<mi>o</mi>
<mi>s</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mi>pq</mi>
<mrow>
<mi>p</mi>
<mi>o</mi>
<mi>s</mi>
</mrow>
</msub>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>;</mo>
</mrow>
In formula, TpredIt is final predicted time, TposIt is the scramble time of position consecutive frame, qpposIt is the quantization of position consecutive frame
Parameter, refposIt is the reference frame number of position consecutive frame, a is the weight of position consecutive frame;TtempIt is the volume of time horizon consecutive frame
Code time, qptempIt is the quantization parameter of time horizon consecutive frame, reftempIt is the reference frame number of time horizon consecutive frame, b is the time
The weight of layer consecutive frame;α is that we calculate obtained changed factor from statistics, and β is based on coded frame among image sets
Position calculates obtained ratio.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710159343.6A CN107071424B (en) | 2017-03-17 | 2017-03-17 | A kind of load-balancing method based on scramble time prediction model |
PCT/CN2018/079378 WO2018166535A1 (en) | 2017-03-17 | 2018-03-17 | Method for load balancing based on encoding time prediction model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710159343.6A CN107071424B (en) | 2017-03-17 | 2017-03-17 | A kind of load-balancing method based on scramble time prediction model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107071424A true CN107071424A (en) | 2017-08-18 |
CN107071424B CN107071424B (en) | 2018-09-25 |
Family
ID=59618376
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710159343.6A Expired - Fee Related CN107071424B (en) | 2017-03-17 | 2017-03-17 | A kind of load-balancing method based on scramble time prediction model |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN107071424B (en) |
WO (1) | WO2018166535A1 (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018166535A1 (en) * | 2017-03-17 | 2018-09-20 | 山东科技大学 | Method for load balancing based on encoding time prediction model |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118120234A (en) * | 2021-08-05 | 2024-05-31 | 抖音视界有限公司 | Method, apparatus and medium for video processing |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120140816A1 (en) * | 2010-12-01 | 2012-06-07 | Jean-Francois Franche | Method and system for parallel encoding of a video |
CN102833551A (en) * | 2012-09-25 | 2012-12-19 | 中南大学 | Slice level coding/decoding end combined time minimization method |
CN102868888A (en) * | 2012-04-27 | 2013-01-09 | 北京航空航天大学 | Dynamic slice control method oriented to parallel video encoding |
CN103227919A (en) * | 2013-03-29 | 2013-07-31 | 苏州皓泰视频技术有限公司 | Scalable video coding (SVC) method based on multi-core processor Tilera |
CN104980764A (en) * | 2014-04-14 | 2015-10-14 | 深圳中兴力维技术有限公司 | Parallel coding/decoding method, device and system based on complexity balance |
CN106231320A (en) * | 2016-08-31 | 2016-12-14 | 上海交通大学 | A kind of unicode rate control method supporting multi-host parallel to encode and system |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150016503A1 (en) * | 2013-07-15 | 2015-01-15 | Qualcomm Incorporated | Tiles and wavefront processing in multi-layer context |
CN103702126B (en) * | 2013-12-10 | 2017-01-18 | 清华大学深圳研究生院 | Parallel encoding optimization method based on standard video HEVC (High Efficiency Video Coding) |
CN104539972A (en) * | 2014-12-08 | 2015-04-22 | 中安消技术有限公司 | Method and device for controlling video parallel decoding in multi-core processor |
CN105245896A (en) * | 2015-10-09 | 2016-01-13 | 传线网络科技(上海)有限公司 | HEVC (High Efficiency Video Coding) parallel motion compensation method and device |
CN106454354B (en) * | 2016-09-07 | 2019-10-18 | 中山大学 | A kind of AVS2 parallel encoding processing system and method |
CN107071424B (en) * | 2017-03-17 | 2018-09-25 | 山东科技大学 | A kind of load-balancing method based on scramble time prediction model |
-
2017
- 2017-03-17 CN CN201710159343.6A patent/CN107071424B/en not_active Expired - Fee Related
-
2018
- 2018-03-17 WO PCT/CN2018/079378 patent/WO2018166535A1/en active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120140816A1 (en) * | 2010-12-01 | 2012-06-07 | Jean-Francois Franche | Method and system for parallel encoding of a video |
CN102868888A (en) * | 2012-04-27 | 2013-01-09 | 北京航空航天大学 | Dynamic slice control method oriented to parallel video encoding |
CN102833551A (en) * | 2012-09-25 | 2012-12-19 | 中南大学 | Slice level coding/decoding end combined time minimization method |
CN103227919A (en) * | 2013-03-29 | 2013-07-31 | 苏州皓泰视频技术有限公司 | Scalable video coding (SVC) method based on multi-core processor Tilera |
CN104980764A (en) * | 2014-04-14 | 2015-10-14 | 深圳中兴力维技术有限公司 | Parallel coding/decoding method, device and system based on complexity balance |
CN106231320A (en) * | 2016-08-31 | 2016-12-14 | 上海交通大学 | A kind of unicode rate control method supporting multi-host parallel to encode and system |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018166535A1 (en) * | 2017-03-17 | 2018-09-20 | 山东科技大学 | Method for load balancing based on encoding time prediction model |
Also Published As
Publication number | Publication date |
---|---|
CN107071424B (en) | 2018-09-25 |
WO2018166535A1 (en) | 2018-09-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102843559B (en) | Method and device for quickly selecting HEVC intra prediction mode on basis of texture characteristics | |
CN101815218B (en) | Method for coding quick movement estimation video based on macro block characteristics | |
CN103517069B (en) | A kind of HEVC intra-frame prediction quick mode selection method based on texture analysis | |
CN102763411B (en) | Method and apparatus to Video coding and the method and apparatus to video decode | |
CN106162167A (en) | Efficient video coding method based on study | |
CN104796694B (en) | Optimization intraframe video coding method based on video texture information | |
CN106131546B (en) | A method of determining that HEVC merges and skip coding mode in advance | |
CN104718753B (en) | Compressed using the bounded rate of chipping rate control | |
CN109660800A (en) | Method for estimating, device, electronic equipment and computer readable storage medium | |
CN106028047B (en) | A method of the fast mode decision and PU of the intra prediction based on Hadamard transformation divide | |
CN102300094B (en) | Video coding method | |
CN101621694B (en) | Motion estimation method, motion estimation system and display terminal | |
CN109040764A (en) | Fast coding algorithm in a kind of HEVC screen content frame based on decision tree | |
CN107071424A (en) | A kind of load-balancing method based on scramble time forecast model | |
CN103702126B (en) | Parallel encoding optimization method based on standard video HEVC (High Efficiency Video Coding) | |
CN101335891B (en) | Video rate control method and video rate controller | |
CN107690069A (en) | A kind of cascade method for video coding of data-driven | |
CN110419214A (en) | Intra prediction mode searching method and device, method for video coding and device and recording medium | |
CN107547902B (en) | Adaptive rate distortion optimization method for surveillance video coding | |
CN102075784A (en) | Video quality assessment method under condition of comprehensively considering compression and packet loss impairment | |
CN110381312A (en) | A kind of method and apparatus that predetermined depth based on HEVC divides range | |
CN101179729A (en) | Interframe mode statistical classification based H.264 macroblock mode selecting method | |
CN100474932C (en) | Video frequency frame image fast coding method based on optimal prediction mode probability | |
CN106658024B (en) | A kind of quick method for video coding | |
CN107734331A (en) | A kind of video transcoding method based on HEVC standard |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180925 |
|
CF01 | Termination of patent right due to non-payment of annual fee |