The present application is concerned with sample array coding such as picture or video coding.
Parallelization of encoder and decoder is very important due to the increased processing requirements by the HEVC standard as well as by the expected increase of video resolution. Multi-core architectures are becoming available in a wide range of modern electronic devices. Consequently, efficient methods to enable the use of multiple-core architectures are necessitated.
Encoding or decoding of LCUs occurs in raster scan, by which the CABAC probability is adapted to the specificities of each image. Spatial dependencies exist between adjacent LCUs. Each LCU (largest coding unit) depends on its left, above, above-left and above-right neighbor LCUs, because of different components, for instance, motion-vector, prediction, intra-prediction and others. Due to enable parallelization in decoding, these dependencies typically need to be interrupted or are interrupted in state-of-the-art applications.
Some concepts of parallelization, namely wavefront processing have been proposed. The motivator for the further study is to perform techniques, which lower the coding efficiency loss and thus reduce the burden on the bitstream for parallelization approaches in encoder and decoder. Furthermore, low-delay processing was not possible with the available techniques.
Thus, it is the object of the present invention to provide a coding concept for sample arrays allowing lower delay at comparatively less penalties in coding efficiency.
According to an embodiment, a decoder for reconstructing a sample array from an entropy encoded data stream may be configured to: entropy decode a plurality of entropy slices within the entropy encoder data stream so as to reconstruct different portions of the sample array associated with the entropy slices, respectively, with performing, for each entropy slice, the entropy decoding along a respective entropy coding path using respective probability estimations, adapting the respective probability estimations along the respective entropy coding path using a previously decoded part of the respective entropy slice, starting the entropy decoding the plurality of entropy slices sequentially using an entropy slice order, and performing, in entropy decoding a predetermined entropy slice, entropy decoding a current part of the predetermined entropy slice based on the respective probability estimations of the predetermined entropy slice as adapted using the previously decoded part of the predetermined entropy slice, and probability estimations as used in the entropy decoding of a spatially neighboring, in entropy slice order preceding entropy slice at a neighboring part of the spatially neighboring entropy slice.
According to another embodiment, a decoder configured to reconstruct a sequence of sample arrays from an entropy encoded data stream may be configured to: entropy decode a current frame of the entropy encoded data stream so as to reconstruct a current sample array of the sequence of sample arrays, perform the entropy decoding along an entropy coding path and using probability estimations and adapt the probability estimations along the entropy coding path using a previously decoded part of the current frame, wherein the decoder is configured to initialize or determine the probability estimations for the current frame based on probability estimations used in decoding a previously decoded frame of the entropy encoded data stream.
According to another embodiment, an encoder for encoding a sample array into an entropy encoded data stream may be configured to: entropy encode a plurality of entropy slices into the entropy encoder data stream each entropy slice being associated with a different portion of the sample array, respectively, with performing, for each entropy slice, the entropy encoding along a respective entropy coding path using respective probability estimations, adapting the respective probability estimations along the respective entropy coding path using a previously decoded part of the respective entropy slice, starting the entropy encoding the plurality of entropy slices sequentially using an entropy slice order, and performing, in entropy encoding a predetermined entropy slice, entropy encoding a current part of the predetermined entropy slice based on the respective probability estimations of the predetermined entropy slice as adapted using the previously encoded part of the predetermined entropy slice, and probability estimations as used in the entropy encoding of a spatially neighboring, in entropy slice order preceding entropy slice at a neighboring part of the spatially neighboring entropy slice.
According to still another embodiment, an encoder configured to encode a sequence of sample arrays into an entropy encoded data stream may be configured to: entropy encode a current frame of the entropy encoded data stream so as to reconstruct a current sample array of the sequence of sample arrays, perform the entropy encoding along an entropy coding path and using probability estimations and adapt the probability estimations along the entropy coding path using a previously encoded part of the current frame, wherein the encoder is configured to initialize or determine the probability estimations for the current frame based on probability estimations used in encoding a previously encoded frame of the entropy encoded data stream.
According to another embodiment, a method for reconstructing a sample array from an entropy encoded data stream may have the steps of: entropy decoding a plurality of entropy slices within the entropy encoder data stream so as to reconstruct different portions of the sample array associated with the entropy slices, respectively, with performing, for each entropy slice, the entropy decoding along a respective entropy coding path using respective probability estimations, adapting the respective probability estimations along the respective entropy coding path using a previously decoded part of the respective entropy slice, starting the entropy decoding the plurality of entropy slices sequentially using an entropy slice order, and performing, in entropy decoding a predetermined entropy slice, entropy decoding a current part of the predetermined entropy slice based on the respective probability estimations of the predetermined entropy slice as adapted using the previously decoded part of the predetermined entropy slice, and probability estimations as used in the entropy decoding of a spatially neighboring, in entropy slice order preceding entropy slice at a neighboring part of the spatially neighboring entropy slice.
According to another embodiment, a method configured to reconstruct a sequence of sample arrays from an entropy encoded data stream may have the steps of: entropy decoding a current frame of the entropy encoded data stream so as to reconstruct a current sample array of the sequence of sample arrays, performing the entropy decoding along an entropy coding path and using probability estimations and adapting the probability estimations along the entropy coding path using a previously decoded part of the current frame, and wherein the method comprises initialize or determine the probability estimations for the current frame based on probability estimations used in decoding a previously decoded frame of the entropy encoded data stream.
According to still another embodiment, a method for encoding a sample array into an entropy encoded data stream may have the steps of: entropy encoding a plurality of entropy slices into the entropy encoder data stream each entropy slice being associated with a different portion of the sample array, respectively, with performing, for each entropy slice, the entropy encoding along a respective entropy coding path using respective probability estimations, adapting the respective probability estimations along the respective entropy coding path using a previously decoded part of the respective entropy slice, starting the entropy encoding the plurality of entropy slices sequentially using an entropy slice order, and performing, in entropy encoding a predetermined entropy slice, entropy encoding a current part of the predetermined entropy slice based on the respective probability estimations of the predetermined entropy slice as adapted using the previously encoded part of the predetermined entropy slice, and probability estimations as used in the entropy encoding of a spatially neighboring, in entropy slice order preceding entropy slice at a neighboring part of the spatially neighboring entropy slice.
According to another embodiment, a method for encoding a sequence of sample arrays into an entropy encoded data stream may have the steps of: entropy encoding a current frame of the entropy encoded data stream so as to reconstruct a current sample array of the sequence of sample arrays, performing the entropy encoding along an entropy coding path and using probability estimations and adapting the probability estimations along the entropy coding path using a previously encoded part of the current frame, wherein the method has initializing or determining the probability estimations for the current frame based on probability estimations used in encoding a previously encoded frame of the entropy encoded data stream.
Another embodiment may have a computer program having a program code configured to perform, when running on a computer, the methods as mentioned above.
If the entropy coding of a current part of a predetermined entropy slice is based on, not only, the respective probability estimations of the predetermined entropy slice as adapted using the previously coded part of the predetermined entropy slice, but also probability estimations as used in the entropy coding of a spatially neighboring, in entropy slice order preceding entropy slice at a neighboring part thereof, the probability estimations used in entropy coding are adapted to the actual symbol statistics more closely, thereby lowering the coding efficiency decrease normally caused by lower-delay concepts. Temporal interrelationships may be exploited additionally or alternatively.
For example, the dependency on probability estimations as used in the entropy coding of a spatially neighboring, in entropy slice order preceding entropy slice may involve the initialization of the probability estimations at the beginning of entropy coding the predetermined entropy slice. Usually, probability estimations are initialized to values adapted to symbol statistics of a representative blend of sample array material. In order to avoid the transmission of the initialization values of the probability estimations, they are known to encoder and decoder by convention. However, such pre-defined initialization values are, naturally, merely a compromise between side information bitrate on the one hand and coding efficiency on the other hand as such initialization values naturally—more or less—deviate from the actual sample statistics of the currently coded sample array material. The probability adaptation during the course of coding an entropy slice adapts the probability estimations to the actual symbol statistics. This process is accelerated by initializing the probability estimations at the beginning of entropy coding the current/predetermined entropy slice using already adapted probability estimations of the just-mentioned spatially neighboring, in entropy slice order preceding entropy slice as the latter values have already been, to some extent, adapted to the actual symbol statistics of the sample array currently at hand. Low-delay coding may nevertheless be enabled by using, in initializing the probability estimations for the predetermined/current entropy slices, the probability estimation used at the neighboring part thereof, rather than manifesting themselves at the end of entropy coding the preceding entropy slice. By this measure, wavefront processing is still possible.
Further, the above-mentioned dependency on the probability estimations as used in the entropy coding of the spatially neighboring, in entropy slice order preceding entropy slice may involve the adaptation process of adapting the probability estimations used in entropy coding the current/predetermined entropy slice itself. Probability estimation adaptation involves the use of the just-coded part, i.e., the just-coded symbol(s), in order to adapt the current state of the probability estimations to the actual symbol statistics. By this measure, initialized probability estimations are adapted at some adaptation rate to the actual symbol statistic. This adaptation rate is increased by performing the just-mentioned probability estimation adaptation not only based on the currently coded symbol of the current/predetermined entropy slice, but also dependent on probability estimations as manifesting themselves at a neighboring part of the spatial neighboring, in entropy slice order preceding entropy slice. Again, by choosing the spatially neighborhood of the current part of the current entropy slice and the neighboring part of the preceding entropy slice appropriately, wavefront processing is still possible. The benefit from coupling the own probability estimation adaption along the current entropy slice, with the probability adaptation of the preceding entropy slice is the increased rate at which the adaptation to the actual symbol statistics takes place as the number of symbols traversed in the current and previous entropy slices contributes to the adaptation, rather than merely the symbols of the current entropy slice.
Embodiments of the present invention will be described subsequently referring to the appended drawings, in which:
In order to ease the understanding of the below-outlined measures for improving the achievement of low-delay at less penalties in terms of coding efficiency, the encoder of
The encoder of
In describing the sample array 10 using the syntax elements, the precoder 42 may subdivide the sample array 10 into coding units 50. The term “coding unit” may, for reasons set out in more detail below, alternatively be called “coding tree units” (CTU)s. One possibility of how precoder 42 may subdivide the sample array 10 into coding units 50 is exemplarily shown in
The precoder 42 may exploit spatial interrelationships between content of neighboring coding units 50 of sample array 10. For example, precoder 42 may predict syntax elements for a certain coding unit 50 from syntax elements determined for previously coded coding units 50 which are spatially adjacent to the currently coded coding unit 50. In
That is, precoder 42 may describe the content of sample array 10 coding unit wise and may, to this end, use spatial prediction. The spatial prediction is restricted for each coding unit 50 to spatially neighboring coding units of the same sample array 10 such that when following a coding order 66 among the coding units 50 of sample array 10, the neighboring coding units serving as prediction reference for the spatial prediction, have generally been traversed by the coding orders 66 prior to the current coding unit 50. As illustrated in
Precoder 42 forwards the syntax elements to the entropy coding stage 44. As just outlined, some of these syntax elements have predictively been coded, i.e. represent prediction residuals. Precoder 42 may, thus, be regarded as a predictive coder. Beyond that, precoder 42 may be a transform coder configured to transform code residuals of the prediction of the content of the coding units 50.
An exemplary internal structure of the entropy coding stage 44 is also shown in
Owing to the properties of entropy coding, the coding efficiency in terms of compression rate, increases with an improvement of the probability estimation: the better the probability estimation matches the actual symbol statistics, the better the compression rate.
In the example of
Manager 76 comprises a storage which stores, for each available context an associated probability estimation. For example, the symbol alphabet may be a binary alphabet so that merely one probability value may have to be stored for each available context.
The initializer 74 may, intermittently, initialize or re-initialize the probability estimations stored within manager 76 for the available contexts. Possible time instants at which such initialization may be performed, are discussed further below.
The adapter 78 has access to the pairs of symbols si and corresponding probability estimations pi and adapts the probability estimations within manager 76 accordingly. That is, each time a probability estimation is applied by entropy coding core 80 so as to entropy code the respective symbol si, into data stream 20, adapter 78 may vary this probability estimation in accordance with the value of this current symbol si, so that this probability estimation pi is better adapted to the actual symbol statistics when encoding the next symbol which is associated with that probability estimation (by way of its context). That is, adaptor 78 receives the probability estimation for the selected context from manager 76 along with the corresponding symbol si and adapts the probability estimation pi accordingly so that, for the next symbol si of the same context ci, the adaptive probability estimation is used.
The entropy coding core 80 is, for example, configured to operate in accordance with an arithmetic coding scheme or in a probability-interval-partitioning-entropy coding scheme. In arithmetic coding, entropy coding core 80 would, for example, continuously update its state when coding the sequence of symbols, with the state being defined by a probability interval defined by a probability interval width value and probability interval offset value, for example. When operating in the pipe concept, the entropy coding core 80 would, for example, subdivide the domain of possible values of the probability estimations into different intervals with performing fixed-probability entropy coding with respect to each of these intervals, thereby obtaining a substream for each of the sub-intervals the coding efficiency of which is respectively tailored to the associated probability interval. In case of entropy coding, the data stream 20 output would be an arithmetically coded data stream signaling to the decoding side information enabling emulating or redoing the interval subdivision process.
Naturally, it would be possible for entropy coding state 44 to entropy code all information, i.e. all syntax elements/symbols si, relating to sample array 10 with initializing the probability estimations merely once at the beginning thereof, and then continuously updating the probability estimations by adaptor 78. This would, however, result in a data stream 20 which would have to be sequentially decoded at the decoding side. In other words, there would not be a possibility for any decoder to subdivide resulting data stream into several sub-portions and decoding the sub-portions in parallel. This would, in turn, hamper any low-delay efforts.
Accordingly, as will be outlined in more detail below, it is favorable to subdivide the amount of data describing the sample array 10 into so-called entropy slices. Each of these entropy slices would, accordingly, cover a different set of the syntax elements relating to the sample array 10. If the entropy coding stage 44 would, however, entropy encode each entropy slice completely independent of each other by firstly initializing the probability estimation once with then continuously updating the probability estimation for each entropy slice individually, then the coding efficiency would be decreased due to the increased percentage of the data relating to, and describing, the sample array 10 for which the probability estimations used are (yet) less accurately adapted to the actual symbol statistics.
In order to overcome the just-mentioned problems in accommodating the wish for low coding delay on the one hand and high coding efficiency on the other hand, the following encoding scheme may be used, which is now described with respect to
Firstly, the data describing the sample array 10 is subdivided into portions called “entropy slices” in the following. The subdivision 80 needs not to be overlap-free. On the other hand, this subdivision may at least partially correspond to a spatial subdivision of the sample array 10 into different portions. That is, according to the subdivision 80, syntax elements describing sample array 10 may be distributed onto different entropy slices depending on the location of the coding unit 50 which the respective syntax element relates to. See, for example.
By subdividing sample array 10 in that way, an entropy slice order 16 is defined among portions 12 along which portions 12 follow each other along the coding order 66. Moreover, for each entropy slice, a respective entropy coding path 14 is defined, namely the fragment of coding path 66 running to the respective portion 12. In the example of
It should be noted that it would be possible to restrict spatial predictions performed by precoder 42 and context derivations performed by context selector 72 so as to not cross slice boundaries, i.e. so that the spatial predictions and context selections do not depend on data corresponding to another entropy slice. This way, the “entropy slices” would correspond the usual definition of “slices” in H.264, for example, which are completely decodable independently from each other, except for the below-outlined probability initialization/adaptation dependency. However, it would also be feasible to allow spatial predictions and context selections, i.e. generally speaking dependencies, to cross slice boundaries in order to exploit local/spatial inter-dependencies as WPP processing is still feasible even as far as the reversal of the precoding, i.e. the reconstruction based on the syntax elements, and the entropy context selection is concerned. In so far, the entropy slices would somehow correspond to “dependent slices”.
Subdivision 80 may, for example, be performed by entropy coding stage 44. The subdivision may be fixed or may vary among the array of sequence 30. The subdivision may be fixed per convention or may be signaled within data stream 20.
Based on the entropy slices, the actual entropy coding may take place, i.e. 82. For each entropy slice, the entropy coding may be structured into a starting phase 84 and a continuation phase 86. The staring phase 84 involves, for example, the initialization of the probability estimations as well as the triggering of the actual entropy coding process for the respective entropy slice. The actual entropy coding is then performed during the continuation phase 86. The entropy coding during phase 86 is performed along the respective entropy coding path 14. The starting phase 84 for each entropy slice is controlled such that the entropy coding of the plurality of entropy slices starts sequentially using the entropy slice order 16.
Now, in order to avoid the above-outlined penalty which would result from entropy coding each entropy slice completely independent of each other, the entropy coding process 82 is controlled such that a current part, e.g. a current coding unit of a current entropy slice is entropy coded based on the respective probability estimations of the current entropy slice as adapted using the previously encoded part of the current entropy slice, i.e. the part of the current entropy slice to the left of the current coding unit 50 in case of
In order to describe the aforementioned dependency more clearly, reference is made to
As described above, the probability estimations used in entropy coding the entropy slices 90 are continuously updated during the continuation phase 86 along the entropy coding path 14 so that the probability estimations increasingly better adapt the actual symbol statistics of the respective entropy slice 90—i.e. the probability estimations are associated with the respective entropy slice. While the probability estimations 94 used to entropy code the entropy slice 90 during the continuation phase 86, are continuously updated, in
Now, in order to achieve the above-outlined dependency, the initial state 96 for entropy coding the n-th entropy slice 90 is set dependent on any intermediate state of the probability estimations 94 manifesting itself during encoding the preceding entropy slice n−1. “Intermediate state” shall denote any state of the probability estimations 94, excluding the initial state 96 and the final state manifesting itself after entropy coding the complete entropy slice n−1. By doing so, entropy coding the sequence of entropy slices 90 along the entropy slice order 16 may be parallelized with a degree of parallelization being determined by the ratio of the number of segments 92 preceding the state used for initialization of the probability estimations 94 for entropy coding the next entropy slice, i.e. a, and a number of segments 92 succeeding this stage, i.e. b. In particular, in
By this measure, the entropy coding of any segment 92 following state 100 in entropy coding path order 14 would depend on the probability estimation 94 as adapted during the continuation phase 86 based on the preceding segments of the same entropy slice as well as the probability estimation as used in entropy coding the third segment 92 of the preceding entropy slice 90.
Accordingly, the entropy coding of the entropy slices 90 could be performed in parallel in a pipelined scheduling. The only restrictions imposed onto the time scheduling would be that the entropy coding of some entropy slice may begin merely after finishing the entropy coding of the a-th segment 92 of the preceding entropy slice. Entropy slices 90 immediately following each other in entropy slice order 16, are not subject to any other restrictions regarding time-alignment of the entropy coding procedure during the continuation phase 86.
In accordance with another embodiment, however, a stronger coupling is used, additionally and/or alternatively. In particular, as illustrated in
More specifically, let p(n→{i,j}, with i,j denoting the position of any coding unit (with (0,0) being the left, top and (I,J) the right, bottom position), i∈{1 . . . . I} and j.epsilon. {1 . . . J}, I being the number or columns, J being the number of rows and p( ) defining the path order 66,
P{i,j} the probability estimation used in entropy coding coding unit {i,j}; and
T(P{i,j}) the result of the probability adaptation 106 of based on coding unit {i,j};
Then, the probability estimations 106 of consecutive entropy slices 90 may be combined to replace the usual, entropy slice internal adaptation according to Pp(n+1)=T(Pp(n)), by
where N may be 1 or more than 1 and {i,j}1 . . . N is/are selected from (lie within) any previous (in entropy slice order 16) entropy slice 90 and its associated portion 12, respectively. The function “average” may be one of a weighted sum, a median function, etc. p(n)={i,j} is the current coding unit and p(n+1) follows in accordance with the coding order 14 and 66, respectively. In the presented embodiments p(n+1)={i+1,j}.
Advantageously, {i,j}1 . . . N fulfill, for each k∈{1 . . . N}, {i,j}1 . . . N={ik,j,jk} and ik<j+3 and jk<j with p(n)={i,j} being the current coding unit (i.e. any of the second or following coding units of the current entropy slice), i.e. they do not lie beyond the wavefront.
In the latter alternative, the entropy coding time scheduling when entropy coding the entropy slices 90 in parallel, is more closely coupled to each other. That is, during the continuation phase 86, the next segment 92 in line of a current entropy slice, may merely be commenced upon finishing the corresponding segment of the preceding entropy slice which is a ranking position further in entropy coding path order 14.
In other words, the above discussion showed an example where the decoder 40, and in particular, the entropy coding stage 44, is configured to perform, for an entropy slice 90 such as the n-th, an initialization of its probability estimations 94 before decoding the first segment 92 corresponding to the first coding unit/first block 50 of the portion 12 corresponding to the n-th entropy slice along the respective encoding path 14 with probability estimations manifesting themselves after having entropy decoded the second coding unit/block 50 of the portion 12 corresponding to the, in entropy slice order 16, preceding entropy slice along the respective encoding path 14. Additionally, or alternatively, the decoder 40 and, in particular, entropy coding stage 44, may be configured to perform, for each entropy slice 90, the entropy decoding and probability estimation adaptation such that, after a current part/block/coding unit 50 of the current entropy slice has been entropy decoded based on the respective probability estimations 94 of the current entropy slice 90, the respective probability estimations 94 of the current entropy slice are adapted depending on this current part of the current entropy slice and the probability estimations as manifesting themselves in entropy decoding a neighboring part/block/coding unit 50 of the spatially neighboring, preceding entropy slice, such as the one in the above neighboring row in the second column to the right of the current part/block/coding unit of the current entropy slice.
As became clear from the above discussion, probability estimation has to be adapted/managed for each entropy slice 50 separately. This may be done by sequentially processing the entropy slices and storing probability estimation states as those exemplarily shown and mentioned with respect to
Before proceeding with
The decoder of
The entropy slices 90 are then, in the entropy decoding process 232, entropy decoded in a manner, mirroring the encoding process 82, namely with performing for each entropy slice 90, a starting phase 234 and a continuation phase 236 with probability estimation initialization and adaptation in the same manner and at the same instances as in the encoding procedure.
The same parallelization as described above with respect to the encoding, is possible at the decoding side. The entropy decoding stage instantiations 130 shown in
After having described embodiments of the present application, the concepts discussed so far are described in the following, again using a different wording. Thereinafter, several further aspects of the present application are described. In the following, the aforementioned coding units 50 are called LCU (largest coding unit), thereby adapting the wording to the upcoming HEVC standard.
Firstly, the above-discussed probability adaptation 106 is briefly discussed again with respect to
A current LCU uses the probability estimations such as CABAC probabilities, available after coding the preceding left LCU. For example, the LCU of
As described above, in entropy decoding and entropy encoding the entropy slices, new probability estimations are computed before encoding or decoding of each LCU, with the help of already existing probabilities (probability estimations) from other LCUs. To be more precise, the probability adaptation is not only performed from any preceding LCUs of the current entropy slice, but also from LCUs of entropy slices preceding an entropy slice order. This approach is represented in
where al, . . . ak are the weighting factors of the LCUs.
It has been tested which weighting of probabilities provide the best results. In this experiment only the neighboring LCUs have been used. The study reveals to use the weighting: 75% from the left LCU and 25% from the above-right LCU. In
However, in the probability estimation adaptation, not only adjacent blocks can be used. Each nearest LCU has peculiarly own neighbors, whose use for probability optimization may be significant. In other words, not only the LCUs from the upper nearest row can be applied. In
It should be admitted that some complexity is introduced by the above outlined probability recalculation or probability estimation adaptation. New computation of probability estimation occurs, for example, in three steps: at first, probability estimations of each candidate have to be gained from each context state. This is done by storing in storage 120 and 240, respectively, or by steering the parallel entropy slice n/decoding processes such that these states are available concurrently. Secondly, by using the equation (1), optimized probability (pnew) will be created. That is, some averaging, for example, may be used in order to combine adapted probability estimations of different entropy slices. As a last step, a new context state is converted from pnew and replaces the old one. That is, probability estimation manager 76 adopts the new probability estimations thus obtained. This performance for each syntax element, particularly using multiplication operations, can increase complexity enormously. The only way to reduce this penalty is to try to avoid these three steps. If the number of candidates and their weightings are determined, a pre-calculated table for each situation can be approximated. Thus, only one simple access to the table dates with the help of candidates' indices (context states) is needed.
It is asserted that this technique can provide good results for both applications using and not using entropy slices. The first application uses only one slice per frame, thus probability adaptation is optimized without any other changes. In case of entropy slices, the adaptation of the probabilities occurs within each slice independent from other slices. This allows for fast learning of the probabilities of the current LCU.
In the above description, the usage of the second LCU of the upper line was also presented, i.e. the usage of the second LCU for probability estimation initialization. Parallelization of encoding and decoding is possible, if some conditions of the bitstream that have been mentioned above (entropy slices) are accomplished. The CABAC probabilities dependency between LCUs must be broken. By the Wavefront Parallel Processing is important to make the first LCU of each line independent from the last LCU of the previous line. It can be achieved if, for instance, the CABAC probabilities are re-initialized at the beginning of each line of LCUs. However, this method is not optimal, because each re-initialization loses the achieved CABAC probabilities, which are adapted to the specificities of the image. This penalty can be reduced, if the initialization of the CABAC probabilities of the first LCU of each line is occurred with the probabilities obtained after the second LCU of the previous line.
As has been described above, a gain in probability adaptation speed is achievable by coupling the probability adaptations of spatially neighboring entropy slices. In particular, in other words, above discussion also anticipates a decoder such as the one of
The different portions may be rows of blocks (e.g. LCUs or macroblocks) of the sample array. The latter may be a picture of a video. The entropy coding path may extend row-wise. The entropy coding, and thus, the probability adaptation as well, may be context adaptive. The entropy slice order may generally be chosen such that, along the entropy slice order, the different portions follow each other in a direction (16) angled relative to the entropy coding paths (14) of the entropy slices, which, in turn, extend substantially in parallel to each other. By this measure, a “wavefront” of currently decoded parts (such as p1, p4, p5+1 and pk in the figure) of the entropy slices may generally be arranged along a line forming an angle to the entropy slice paths smaller than the sequence direction of the portions. The wavefront may have to have a slope of 1y per 2× block positions in order for the upper left reference to be there, for all threads processing slices in parallel.
The decoder may be configured to perform, for each entropy slice, the entropy decoding along the respective entropy coding path in units of parts of the sample array portion of the respective entropy slice so that the entropy slices consist of the same number of parts, respectively, and the sequence of parts of the portions along the entropy slice paths are aligned to each other in a direction lateral to the entropy slice paths. The current part of the portion of the predetermined entropy slice belongs to the resulting grid of parts (e.g. LCUs or macroblocks). In performing, for each entropy slice, the entropy decoding along a respective entropy coding path, the decoder may preserve offsets/shifts between starting times between immediately consecutive entropy slices in entropy slice order so that the wavefront of currently decoded parts of the entropy slices forms a diagonal such as up to a line with slope of 0.5× block positions with respect to entropy slice paths and entropy slice order direction. The offset/shift may correspond to two parts for all pairs of immediately consecutive entropy slices. Alternatively, the decoder may merely prevent the distance of currently decoded parts of immediately consecutive (and immediately neighboring as far as their sample array portions 12 are concerned) entropy slices to become lower than two parts. See, the figure above: as soon as the part/block p4 has been decoded, the part/block to the right thereof is decoded according to path order 16, and concurrently, if any, X is decoded or any of the parts/blocks preceding the same). By this measure, the decoder may use the probability estimates as already adapted based on the content of part/block p4, i.e. the part within the spatially neighboring portion, being aligned to the part of the portion 12 of the predetermined entropy slice succeeding the current part X in path order 16, in order to determine the probability estimates to be used in decoding X. In case of a constant offset in decoding the immediately entropy slices of two parts, the decoder is able to use the probability estimates as already adapted based on the content of part/block p4 concurrently for entropy decoding the subsequent part (i.e. the part to the right of p4) of the spatially neighboring entropy slice.
As described above, a weighted sum of the already adapted probability estimations may be used in order to determine the probability estimates to be used for decoding X.
As also described above, the entropy slice order may also go over frame boundaries.
It should be noted that the just-outlined probability adoption from the predecessor entropy slices may be performed for every part of the current/predetermined entropy slice for which such neighboring parts in the predecessor entropy slices are available. That is this is also true for the first part along the path direction 16, and for this very first part/block (the leftmost in each entropy slice in the figure) the adoption equals the initialization described above.
In order to better adapt, also in this case, the 2 above-mentioned methods can be combined together. The results of this process with 1 and 4 Threads, i.e. the parallel used processing units, are illustrated in
In order to better understand the context of the above embodiments and, especially, the further embodiments described below, in particular the use of LCUs, one may first have a look at the structure of H.264/AVC.
A coded video sequence in H.264/AVC consists of series of access units that are collected in the NAL unit stream and they use only one sequence parameter set. Each video sequence can be decoded independently. A coded sequence consists of a sequence of coded pictures. A coded frame can be an entire frame or a single field. Each picture is partitioned into fixed-size macroblocks (in HEVC: LCUs). Several macroblocks or LCUs can be merged together into one slice. A picture is therefore a collection of one or more slices. The goal of this data separation is to allow independent decoding of the samples in the area of the picture, which is represented by the slice, without the use of data from other slices.
A technique that has also often referred to as “entropy slices” is a splitting of the traditional slice into additional sub-slices. Specifically, it means slicing of entropy coded data of a single slice. The arrangement of entropy slices in a slice may have different varieties. The simplest one is to use each row of LCUs/macroblocks in a frame as one entropy slice Alternative, columns or separate regions can be utilized as entropy slices, which even can be interrupted and toggled with each other, e.g. slice 1 in
An obvious aim of the entropy slice concept is to enable the use of parallel CPU/GPU and multi-core architectures in order to improve the time of the decoding process, i.e. to speed-up the process. The current slice can be divided into partitions that can be parsed and reconstructed without reference to other slice data. Although a couple of advantages can be achieved with the entropy slice approach, thereby emerging some penalties.
First of all, a main aim is to create a bitstream that is suitable for a parallel encoding and decoding process. One must take into account, that a LCU can be encoded only if the neighboring LCUs (left, top, top-right) are already available as coded version, in order to use spatial and motion information for prediction. To enable a parallelism by slicing, a shift between processing of slices must be realized (for example shift of 2 LCUs, as typical for the wavefront approach). Due to CABAC probabilities adaptation, an LCU use the probabilities that are available from the previous encoded LCU. In regard to raster scan order, the problem, which occurs by slicing the picture, is the prevention of parallelism, since the first LCU of each line depends on the last LCU of the previous line. On impact of this is that the CABAC probabilities dependencies between slices must be broken, so that several slices can be started simultaneously. One way to do this is the usual CABAC re-initialization whereby, however, all the adopted data will be lost. As a result, bitrate may increase.
Secondly, each slice generates its own sub-bitstream, which can be put in series into the main stream. But it is necessitated to pass decoder special information, so that these slices and their positions in the main stream can be correctly identified. Two scenarios for signaling are supported. The location information can be stored in the picture header (slices length information) or in each slice header (points as start code). Byte alignment at the end of each entropy slice and location information raise loss.
In order to reduce the penalty introduced by the signaling for entropy slices, it is essential to use a good coding technique for the signaling. A noticeable penalty for entropy slice signaling is introduced in a frame, if start codes are used for each slice, i.e. too much additional bytes (e.g. min. 4 bytes per slice) are added to the bitstream. Certainly, insertion of entropy slices 90 using start codes has the advantage in low delay scenarios where the encoder shall be able to instantaneously output entropy slices. In such cases, there is no up-front signaling of entry points possible.
In less low-delay strict scenarios, however, the opportunity to store slice lengths (offsets) seems more suitable. One well known method to code such information is Variable Length Code (VLC) or Exponential-Golomb coding. The main feature of VLC is adding empty information (zeros) before true information. With the help of these zeros, a code that stores information about offset length can be determined. We propose another technique to realize this, which schema is shown in
That is, in accordance with the aspect of
In particular, the VLC bit sequence may have a prefix of variable length y, indicating that the difference x falls into the y-th of a sequence of a number z of intervals [0,2a−1]. [2a,2b+2a−1], [2b+2a, 2c+2b+2a−1] as so forth, and a PCM coded suffix of the y-th of the sequence of lengths a, b, c . . . If a, b, c, . . . are chosen to be a power of two, if the corresponding y is added, i.e. so that a+1, b+2, c+3 and so forth are all powers of two, then byte alignment may be preserved. The number z is not restricted to three as exemplarily chosen in
The encoder, such as the one in
A further advantage may be achieved by segmentation of entropy slices, namely for low-delay transmission and decoding speed-up.
In video-streaming, enabling of higher resolutions (Full-HD, QUAD-HD etc.) leads to higher amount of data that has to be transmitted. For time-sensitive scenarios, so called Low-Delay use-case (<145 ms), the transmission time becomes a critical factor Consider the up-load link of ADSL for a video conferencing application. Here, so called random access points of stream, usually these refer to I-frames, will be the candidates to cause a bottleneck during transmission.
To solve that problem and to minimize the delay of transmission and decoding time. i.e. the end-to-end delay, a new technique for an interleaved entropy slice scheme for parallel transmission and processing may be applied.
HEVC will allow for so called Wavefront-processing at the decoder side. That is going to be enabled by use of entropy slices. In normal case the data of a whole slice will be delivered at one time. The decoding process will start as soon as encoded data is arrived at Wavefront-decoder-engines. To reduce the time when decoder can start and finish a frame, a segmentation of entropy slices into small chunks using an interleaving approach is used in accordance with the present embodiment. Hence the encoder can deliver data, correspondent to particular entropy slice, to transport layer earlier than in normal case. It results then in faster transmission and earlier start of a parallel decoding process at the client.
Creating chunks of a slice can be further achieved by sub-dividing an entropy slice in further slices keeping all dependencies (dependent slices). If done so in each Largest Coding Unit (LCU)/Coding tree unit (CTU), these chunks can be interleaved by additionally using system layer techniques which allow for transmitting the chunks in an interleaved form and recovering or at least providing the knowledge of the original decoding order of the chunks via additional signaling. Such signaling may be a decoding order number (DON), such as defined in the IETF RTP Payload Format for H.264/AVC (RFC 3984). Another system method may be to apply the chunks of a wavefront substream to a different transportstream, as in MEPG-2 Systems, by assigning a different PID to each of them, further multiplexing and by that interleaving them on the transport channel.
The approach may be also applied over frame boundaries, in case, if the following frame's slice(s) or entropy slice(s) can be already decoded, e.g. in wavefront manner, based on the knowledge that the necessitated information for decoding an entropy slice of a following frame due to the availability of inter-frame references. Those already decodable data of a frame succeeding in decoding order may be derived from the maximum allowed/signaled motion vector length or additional information in the stream indicating the dependencies of data parts to the preceding frame(s)), or a fixed referencing scheme, indicating the position signaled in a sequence-fixed position such as a parameter set. This will further be outlined below.
Picture may be encoded with one entropy slice per largest coding unit (LCU)-row(s) as depicted in
There are two general solutions possible. First one is a generation of one-byte segments (usually the bitstream representation of slice is byte-aligned) and controlling of consuming of bytes by each decoder-engine, i.e. the decoder finds out the completion of an entropy slice.
The second is using of finalizing code at the end of slice. In the last case, variable length segments are possible, but it can also lead to higher amount of data. Another method is the signaling of the entropy slice length. One way of such an alternative is described below
The size of segment and interleaving mode can be signaled either in one SEI-Message or in SPS.
The transmission scheme is shown in
Thus, in accordance with the aspect of
As denoted above and described with respect to the other aspects, the slices may be entropy slices 90 which, in turn, may be proper subsets of slices of frames, and accordingly, the encoded data stream may be an entropy encoded data stream 20.
The interleaving 312 of the slices 90 enables lower delay since the decoder responsible for decoding any of the slices 90 does not need to wait for a time duration consumed by previous slices of other decoders (according to slice order 16). Rather, all available decoders are able to begin decoding their associated slices as soon as the first chunk 310 thereof is available and potential inter-slice dependencies are resolved, cf. wavefront approach.
The different portions 12 may be entropy encoded into the entropy slices 90 using probability estimations which are set independently among the entropy slices, or using inter-entropy-slice adoption of probability estimates as described above such as, for example, namely in that the different portions 12 are encoded into the entropy slices 90 along respective entropy coding paths 14, and subordinate entropy slices have their corresponding portions 12 encoded therein using probabilities estimates derived, inter alias, from probability estimates used in higher ranking entropy slices at spatially neighboring parts of the corresponding portion, as described previously.
The entropy encoded data stream 20 may additionally comprise a header as is shown as an option in
For example, the entropy slices 90 may be arranged after the header 300 as illustrated in
The chunks 310 may be of equal length, at least as far as a beginning contiguous portion of the sequence of chunks 310 from a first one of the chunks on along an order in which the chunks are arranged within the entropy encoded data stream 20, is concerned. Subsequent chunks may vary in length. The subsequent chunks may be equal to the length of the beginning contiguous portion of the sequence or smaller than this length. The length of the subsequent chunks may be derived from the aforementioned information within the header 300 revealing the length of the entropy slices 90 or the start position.
The chunks 310 may be arranged within the entropy encoded data stream 20 in a cyclic manner in accordance with an order defined among the entropy slices. For entropy slices, the chunks of which are completely within the previous cycles, may be skipped in the current and subsequent cycles.
Other information signals then a sample array sequence such as a video signal could by conveyed via data stream 20 as well. The different portions 12 need, thus, be no portions of a predetermined sample array such as a picture/frame.
As described above, the entropy slices 90 may have the different portions 12 of the sample array 10 encoded thereinto via predictive coding using inter-entropy-slice prediction and/or inter-frame prediction an entropy encoding of a prediction residual of the inter-entropy-slice prediction and/or inter-frame prediction. That is, as described above, the different portions may be spatially distinct portions of the frame 10 or multiple frames 30. Latter case applies, if the following frame's(s′) slice(s) or entropy slice(s) can be already decoded, e.g. in wavefront manner, based on the knowledge that the necessitated information for decoding an entropy slice of a following frame due to the available inter-frame references. Those already decodable data of a frame succeeding in decoding order may be derived from the maximum allowed/signaled motion vector length or additional information in the stream indicating the dependencies of data parts to the preceding frame(s)), and inter-entropy-slice prediction may involve intra prediction, while inter-frame prediction may involve motion-compensated prediction. An example is outlined below.
The aforementioned independency of the probability estimation setting among the entropy slices, may relate to both probability adaptation as well as context modeling. That is, the context chosen within an entropy slice, may be chosen independently from other entropy slices, and the probability estimation of a context may be initialized and adapted independent from any other entropy slice as well.
An corresponding entropy decoder may be constructed as follows.
An entropy decoder configured to decode an entropy encoded data stream 20 which comprises entropy slices 90 into which different portions of a frame are entropy encoded, wherein the entropy slices are sub-divided into chunks 310 which are arranged within the entropy encoded data stream 20 in an interleaved manner, may be configured as shown in
In particular, as illustrated in
In other words, the entropy slices may be sub-divided into chunks, which, in turn, may be interleaved, and the decoder may comprise a de-interleaver to de-interleave the chunks and may start operating on the entropy slices in parallel along the paths 16 even before the reception of any of the entropy slices as a whole. It should be recalled that the length of the chunks may be measured in the entropy encoded domain rather than in the syntax domain so as to correspond, for example, to a number of certain spatial parts/blocks in the picture or the like although the latter option would be available as well.
Next, possible usage of temporal dependencies is described. Same may be used in addition to, or alternatively to the probability estimation enhancement embodiments described so far.
Wavefront Processing pattern may, as it is described now, be extended to the entropy coding with the new probabilities calculation for each LCU to use also the temporal dependencies between frames.
As well-known, the probabilities will be re-initialized at the beginning of each frame (the first LCU). Thereby the probabilities, which have already been acquired in the previous frame, are lost. In order to decrease the lose of coding efficiency, one may pass the end state of picture (cp. 320) or, in the case of entropy slice usage, end state of slice (cp. 322) of the probabilities from the reference frame 324 to the first LCU 50 of the current frame 10 or entropy slice 12 respectively (
Using above notation, one may thus, for initializing in starting phase 84 and 234, respectively, set P{i,j} with {i,j} denoting the first CU 50 in the k-th entropy slice of the current sample array equal to, or at least depending on, any T(P{i,j}′) with {i,j}′ denoting a CU within a preceding (in sample array coding order which may be equal to the presentation order) sample array or a combination of several T(P{i,j}′)'s. This may be done for k=0 only, or for each entropy slice k∈{1 . . . . K} with K denoting the number of entropy slices in the current frame. The temporal initialization may be done additionally or alternatively to the spatial initialization described above. That is, P{i,j} with {i,j} denoting the first CU 50 in the k-th entropy slice may be set equal to some combination (such as sme average) of T(P{i,j}′) and T(P{i,j}spatial) with {i,j}′ denoting a CU within a preceding (previously de/coded) sample array or a combination of several T(P{i,j}′)'s and {i,j}spatial denoting a CU within a preceding entropy slice of the current sample array. As far as the location of {i,j}′ is concerned, P{i,j} with {i,j} denoting the first (in entropy coding order 14) CU 50 in the k-th (in entropy coding order 14) entropy slice of the current sample array may be set equal to T(P{i,j}′) with {i,j}′ denoting the last (in entropy coding order 14) CU within the k-th (in entropy slice order) entropy slice in the preceding (in sample array coding order) sample array or the last CU within the last (in entropy slice order) entropy slice in the preceding (in sample array coding order) sample array. Again, this temporal initialization may be performed for the first entropy slice in the sample array only.
The parsing process of the end state of the reference frame was tested with the method of probabilities adaptation, which results are illustrated on
Another opportunity to use the dates from other frames is to exchange the obtained probabilities between collocated LCUs. The main idea is based on assertion that properties of reference frame differ from current frame not greatly. In order to quicken the learning of the probabilities along LCUs in a frame, one can try to pass the end state of each LCU to the appropriate LCU in current frame. This proposal is illustrated on
Under reference frame, different opportunities can be understood. As for instance, a frame, which is the last coded can be used as a reference frame. Otherwise the last coded frame only from the same temporal layer can be appropriated for reference.
Moreover, this approach can be merged with already above proposed methods, as usage of last (slice) information from reference frame, probabilities adaptation and usage of the second LCU of the upper line.
The above spatial adaptation process may be modified to be
where N may be 1 or more than 1 and {i,j}1 . . . N is/are selected from (lie within) any previous (in entropy slice order 16) entropy slice 90 in the current sample array 10 and its associated portion 12, respectively, and M may be 1 or more than 1 and {i,j}′1 . . . M is/are lie within the preceding sample array 350. It may be that (the) at least one of CU's 50 {i,j}′1 . . . M is co-located to p(n). With regard to possible selections of CU's 50 {i,j}1 . . . N reference is made to the above description. The function “average” may be one of a weighted sum, a median function, etc.
The above spatial adaptation process may be replaced by
where M may be 1 or more than 1 and {i,j}′1 . . . m is/are lie within the preceding sample array. It may be that (the) at least of {i,j}′1 . . . M is co-located to p(n). With regard to possible selections of {i,j1 . . . N reference is made to the above description. The function “average” may be one of a weighted sum, a median function, etc. It may be that (the) at least of {i,j}′1 . . . M is co-located to p(n).
As a specific extension of usage of collocated information, an approach to utilize the obtained dates from other blocks from one or even more reference frames can be applied.
Earlier mentioned techniques use only the obtained information from the direct neighbors in current frame or in reference frames. However it does not mean that the gained probabilities in this case are the best ones. The adjacent LCUs, according to the picture partitions (residuals), do not have the best probabilities models. It is supposed that the best results can be achieved with the help of blocks, from which prediction will be done And thus, this appropriate block can be used as a reference for the current LCU.
Thus, in the above adaptation example, {i,j}1 . . . N and/or {i,j}′1 . . . M may be selected depending on CU's serving as providers of predictors for p(n).
The temporal probability adaptation/initialization schemes presented can also be used without entropy slices or a single entropy slice per frame.
In accordance with the latter aspect, a gain in probability adaptation speed is achieved by coupling the probability adaptations of temporally neighboring/related frames. What it is described there is, a decoder such as the one of
That is, for example, the probability estimations for the current frame are initialized based on probability estimations resulting after having finished decoding the previously decoded frame of the entropy encoded data stream. Buffer requirements are, therefore low, since merely the end state of the probability estimates have to be buffered till the start of the decoding of the current frame. Of course, this aspect is combinable with the aspect of
Further, for example, the probability estimations used for coding the parts/blocks of the previously decoded frame are buffered all, not only the end state, and the decoder would, in entropy decoding the predetermined entropy slice (with reference to above description of spatially couple probability derivation), perform entropy decoding the current part (X) of the predetermined entropy slice based on the respective probability estimations of the predetermined entropy slice as adapted using the previously decoded part of the predetermined entropy slice (including p1, for example), and probability estimations as used in the entropy decoding of a spatially corresponding part of an entropy slice of the previously decoded frame with, optionally, additionally using probability estimations as used in the entropy decoding of a spatially neighboring, in entropy slice order preceding entropy slice (the slice comprising X, for example) at a neighboring part (such as p4) of the spatially neighboring entropy slice, as described above. As has also been described above, the spatial correspondence among parts, and the identification of an appropriate one for probability adoption for the current frame among the previously decoded frame may be defined by help of motion information such as, motion indices, motion vectors and the like, of the current part/block.
Up to now, the wavefront extending during wavefront processing has primarily been described as extending obliquely through one sample array 10 with the coding/decoding being performed one sample array after the other. However, this is not a must. Reference is made to
Please note that the just-mentioned wavefront aspect does also work in combination with the above-outlined probability estimation couplings across entropy slice boarders.
Further, with respect to the above outlined chunk aspect, it should also be noted that subdividing entropy slices into smaller pieces, i.e. chunks, is not restricted to being performed in the entropy coded domain, i.e. in the entropy compressed domain. Think of the above discussion: entropy slices as described above have the advantage of reducing the coding efficiency loss despite the enablement of wavefront processing due to the derivation of probability estimations from previously coded/decoded entropy slices of the same or a previously coded/decoded frame, i.e. an initialization and/or adaptation of the probability estimations based on probability estimations of such previous entropy slices. Each of these entropy slices is supposed to be entropy coded/decoded by one thread in case of wavefront processing. That is, when subdividing the entropy slices, it is not necessary to render the chunks codable/decodable in parallel. Rather, the encoder shall merely be provided with the opportunity to output subparts of its entropy slice's bitstream prior to finalizing the entropy encoding, and the decoder shall be provided with the opportunity to operate on these subparts, i.e. chunks, prior to the reception of the remaining chunks of the same entropy slice. Moreover, the interleaving shall be enabled at the reception side. In order to enable the latter de-interleaving, however, it is not necessary to perform the subdivision in the entropy coded domain. In particular, it is possible to perform the above presented subdivision of entropy slices into smaller chunks without severe coding efficiency loss, by intermittently merely resetting the internal state of the probability interval, i.e. the probability interval width value, and the offset value, respectively, of the entropy coding/decoding core. The probability estimations, however, are not reset. Rather, they are continuously updated/adapted from the beginning to the end of the entropy slices, respectively. By this measure, it is possible to subdivide the entropy slices into individual chunks with the subdivision being performed in the syntax element domain rather than the compressed bitstream domain. The sub-division may follow a spatial sub-division as outlined below in order to ease a signaling of the chunk interfaces to the decoder. Each of the chunks could be provided with its own chunk header revealing, for example, its starting position in the sample array, measured, for example, with respect to the coding order 14 relative to the respective entropy slice's start position along with an index to its entropy slice, or relative to a prominent location of the sample array 10 such as the upper left corner.
In order to more clearly describe the subdivision of entropy slices into chunks in accordance with the latter embodiment, reference is made to
Concurrently, another thread operates on the second entropy slice or portion 12 in entropy slice order 16. Upon finishing a first sub-portion of this second entropy slice/portion 12, a chunk 2a is output, whereupon entropy coding the remainder of the second entropy slice is commenced while maintaining, however, the probability estimation as valid at the end of chunk 2a.
With a time axis 394,
The decoder has to reassign the chunks to its sub-portions 12a, 12b and so forth. To this end, each chunk 390 may have a header section 398 revealing the location of the beginning of its associated sub-portion 12a or 12b, i.e. the sub-portion the syntax elements describing the same are entropy coded in the respective chunk. By using this information, the decoder is able to associate each chunk 390 with its entropy slice and with its sub-portion within the portion 12 of that entropy slice.
For illustration purposes,
As became clear from the above discussion, almost no coding efficiency loss resulted from the subdivision of the entropy slices into chunks. Merely the entropy finishing processes and the packetizing may involve some coding efficiency loss, but on the other hand low delay gains are enormous.
Again, please note that the just-mentioned spatial sub-division chunk aspect does also work in combination with the above-outlined probability estimation couplings across entropy slice boarders, spatially and temporally.
The decoder such as decoder of
The above description reveals different methods, which can be useful for parallel encoding and decoding as well as helpful for optimization of already existing processes in the emerging HEVC video coding standard. A short overview of entropy slices has been presented. It has been shown how they can be formed, which advantages can be achieved by slicing and what penalties can occur from those techniques. A number of methods have been proposed, which are supposed to improve the learning process of probabilities along LCUs (largest coding unit) in frame, by better exploiting the local dependences between LCUs, and also the temporal dependences between LCUs of different frames. It is asserted that different combinations provide improvements for both concepts with and without parallelization of encoding and decoding.
The performance improvement in High Efficiency, for instance, by the best mix of proposed approaches, is −0.4% in Intra, −0.78% in Low Delay and −0.63% in Random Access in comparison with HM3.0 without use of entropy slices or −0.7% in Intra, −1.95% in Low Delay and −1.5% in Random Access in comparison with entropy slice approach with usual re-initialization.
In particular, inter alias, the following techniques have been presented above. [0166] to use not only local but also temporal dependencies of LCUs, to optimize adaptation of the CABAC probabilities before coding each LCU, see
All the methods, which were mentioned above, have been integrated and tested in HM3.0.
The obtained results, where reference point is HM3.0 without any entropy slice implementation, are presented in tables 1 and 2 (where 2LCU—usage of the second LCU of the upper line; 2LCU+Prob.Adap-2LCU merged with the method of probabilities adaptation; Temporal—usage of temporal dependences (end state of a reference frame) with the probabilities adaptation for each LCU).
However, it is interesting to know how the proposed approaches affect the wavefront processing with the re-initialization of probabilities at the beginning of each line of LCUs. These results are illustrated on tables 3 and 4 (where orig_neilnit is comparison HM3.0 without use of entropy slices with usage of entropy slices with re-initialization).
The above results show that considerably more use of dependences within and between frames and rational application of already obtained information prevent average loss.
An approach for Wavefront Processing for HEVC video encoding and decoding merges the possibility to use dependences between adjacent LCUs as well as temporal frame dependences with the concept of Wavefront Parallel Processing. In this way the loss can be reduced and the performance advance can be achieved.
A gain in probability adaptation speed has been achieved by computing the probability adaptations of spatially neighboring entropy slices.
As already mentioned above, all of the above aspects may be combined with each other, and thus, the mentioning of certain implementation possibilities with regard to a certain aspect shall, of course, also apply to the other aspects.
Although some aspects have been described in the context of an apparatus, it is clear that these aspects also represent a description of the corresponding method, where a block or device corresponds to a method step or a feature of a method step. Analogously, aspects described in the context of a method step also represent a description of a corresponding block or item or feature of a corresponding apparatus. Some or all of the method steps may be executed by (or using) a hardware apparatus, like for example, a microprocessor, a programmable computer or an electronic circuit. In some embodiments, some one or more of the most important method steps may be executed by such an apparatus.
The inventive encoded signals mentioned above can be stored on a digital storage medium or can be transmitted on a transmission medium such as a wireless transmission medium or a wired transmission medium such as the Internet.
Depending on certain implementation requirements, embodiments of the invention can be implemented in hardware or in software. The implementation can be performed using a digital storage medium, for example a floppy disk, a DVD, a Blu-Ray, a CD, a ROM, a PROM, an EPROM, an EEPROM or a FLASH memory, having electronically readable control signals stored thereon, which cooperate (or are capable of cooperating) with a programmable computer system such that the respective method is performed. Therefore, the digital storage medium may be computer readable.
Some embodiments according to the invention comprise a data carrier having electronically readable control signals, which are capable of cooperating with a programmable computer system, such that one of the methods described herein is performed.
Generally, embodiments of the present invention can be implemented as a computer program product with a program code, the program code being operative for performing one of the methods when the computer program product runs on a computer. The program code may for example be stored on a machine readable carrier.
Other embodiments comprise the computer program for performing one of the methods described herein, stored on a machine readable carrier.
In other words, an embodiment of the inventive method is, therefore, a computer program having a program code for performing one of the methods described herein, when the computer program runs on a computer.
A further embodiment of the inventive methods is, therefore, a data carrier (or a digital storage medium, or a computer-readable medium) comprising, recorded thereon, the computer program for performing one of the methods described herein. The data carrier, the digital storage medium or the recorded medium are typically tangible and/or non-transitionary.
A further embodiment of the inventive method is, therefore, a data stream or a sequence of signals representing the computer program for performing one of the methods described herein. The data stream or the sequence of signals may for example be configured to be transferred via a data communication connection, for example via the Internet.
A further embodiment comprises a processing means, for example a computer, or a programmable logic device, configured to or adapted to perform one of the methods described herein.
A further embodiment comprises a computer having installed thereon the computer program for performing one of the methods described herein.
A further embodiment according to the invention comprises an apparatus or a system configured to transfer (for example, electronically or optically) a computer program for performing one of the methods described herein to a receiver. The receiver may, for example, be a computer, a mobile device, a memory device or the like. The apparatus or system may, for example, comprise a file server for transferring the computer program to the receiver.
In some embodiments, a programmable logic device (for example a field programmable gate array) may be used to perform some or all of the functionalities of the methods described herein. In some embodiments, a field programmable gate array may cooperate with a microprocessor in order to perform one of the methods described herein. Generally, the methods may be performed by any hardware apparatus.
While this invention has been described in terms of several embodiments, there are alterations, permutations, and equivalents which will be apparent to others skilled in the art and which fall within the scope of this invention. It should also be noted that there are many alternative ways of implementing the methods and compositions of the present invention. It is therefore intended that the following appended claims be interpreted as including all such alterations, permutations, and equivalents as fall within the true spirit and scope of the present invention.
The present application is a continuation of U.S. patent application Ser. No. 16/840,043 filed Apr. 3, 2020, which is a continuation of U.S. patent application Ser. No. 16/578,716 filed Sep. 23, 2019, now U.S. Pat. No. 10,652,564, which is a continuation of U.S. patent application Ser. No. 16/109,874, filed Aug. 23, 2018, now U.S. Pat. No. 10,659,798, which is a continuation of U.S. patent application Ser. No. 15/824,196, filed Nov. 28, 2017, now U.S. Pat. No. 10,085,035, which is a continuation of U.S. patent application Ser. No. 15/639,312, filed Jun. 30, 2017, now U.S. Pat. No. 9,860,547, which is a continuation of U.S. patent application Ser. No. 15/414,356, filed Jan. 24, 2017, now U.S. Pat. No. 9,729,891, which is a continuation of U.S. patent application Ser. No. 14/141,374, filed Dec. 26, 2013, now U.S. Pat. No. 9,596,469, which is a continuation of International Application No. PCT/EP2012/063929, filed Jul. 16, 2012, which claims priority from U.S. Provisional Application No. 61/508,477, filed Jul. 15, 2011, all of which are incorporated herein by reference in their entireties.
Number | Date | Country | |
---|---|---|---|
61508477 | Jul 2011 | US |
Number | Date | Country | |
---|---|---|---|
Parent | 17210026 | Mar 2021 | US |
Child | 18160689 | US |
Number | Date | Country | |
---|---|---|---|
Parent | 18160689 | Jan 2023 | US |
Child | 18597342 | US | |
Parent | 16987955 | Aug 2020 | US |
Child | 17210026 | US | |
Parent | 16840043 | Apr 2020 | US |
Child | 16987955 | US | |
Parent | 16578716 | Sep 2019 | US |
Child | 16840043 | US | |
Parent | 16109874 | Aug 2018 | US |
Child | 16578716 | US | |
Parent | 15824196 | Nov 2017 | US |
Child | 16109874 | US | |
Parent | 15639312 | Jun 2017 | US |
Child | 15824196 | US | |
Parent | 15414356 | Jan 2017 | US |
Child | 15639312 | US | |
Parent | 14141374 | Dec 2013 | US |
Child | 15414356 | US | |
Parent | PCT/EP2012/063929 | Jul 2012 | WO |
Child | 14141374 | US |