A Fast Algorithm for VVC Intra Coding Based on the Most Probable Partition Pattern List

Round 1

Reviewer 1 Report

The paper proposes a fast partition method based on the Most Probable Partition Pattern List (MPPPL) to reduce the computational complexity for VVC intra coding. Experimental results demonstrate the effectiveness of the proposed method. Some concerns are as follows.

1.Please pay attention to grammatical mistakes, such as “on the basis of MPPPL, use the sub-block pixel mean difference ”, “.when employing an all-intra configuration”, “a neural network model is trained use 88 cross-entropy and use for early termination of the partition process. ”...

2.It should give the full name of “MPPPL” at the first time in the main body of the manuscript. 

3.In section III, it claims that “The partition mode with the largest average texture complexity of the sub-CU is recorded as Modemax ”. What is the meaning of the texture complexity? Please explain it. Besides, Modemax should be written in the form of variable.

4.At line 293, "MPPPM" should be modified to "MPPPL."

5.Fig. 6 should be enlarged to make it more readable.

6. Missing references:

“Motion-homogeneous-based fast transcoding method from H.264/AVC to HEVC,” IEEE Transactions on Multimedia, vol.19, no. 7, pp. 1416-1430, Jul. 2017.

NA

Author Response

1. Please pay attention to grammatical mistakes, such as “on the basis of MPPPL, use the sub-block pixel mean difference ”, “.when employing an all-intra configuration”, “a neural network model is trained use 88 cross-entropy and use for early termination of the partition process. ”...

Thank you for your correction. We have made detailed modifications to the article

2. It should give the full name of “MPPPL” at the first time in the main body of the manuscript. 

Thank you for your correction. We have made modifications to this

3. In section III, it claims that “The partition mode with the largest average texture complexity of the sub-CU is recorded as Modemax ”. What is the meaning of the texture complexity? Please explain it. Besides, Modemax should be written in the form of variable.

Thank you for your correction. We have made modifications to the errors you pointed out and provided the concept of texture complexity in the article

4. At line 293, "MPPPM" should be modified to "MPPPL."

Thank you for your correction. We have made modifications to this

5. 6 should be enlarged to make it more readable.

 

Thank you for your correction. We have redrawn a clearer flowchart and replaced the previously unclear one 

 

6. Missing references:

“Motion-homogeneous-based fast transcoding method from H.264/AVC to HEVC,” IEEE Transactions on Multimedia, vol.19, no. 7, pp. 1416-1430, Jul. 2017.

Thank you for your correction. We have updated the references .

Reviewer 2 Report

This paper introduces a heuristic algorithm to reduce the complexity of intra coding in VVC standard based on video content. The contributions of this paper contain: (i) fast decision-making process of partitioning mode and (ii) skip the reference line 3 in intra prediction process.

1   1)     The proposed method is totally based on video content. However, to generate the final partition map, the encoder must also consider the input parameters such as QP. Please let me know if the input parameters have any effect.

2  2)     VVC has some exceptions (as presented in Section 3.2.4 and 3.2.5 in [1]). Please present how the proposed algorithm will handle these exceptions?

3  3)     Figure 1a and Figure 8 need to be modified. Figure 1a is quite low quality, while there is an error with Figure 8. It should be noted that Fig.8 is important to know how Threshold is determined.

4   4)     Letters in Fig.6 are too small to read.

5   5)     This paper proposes two algorithms: fast partition mode decision and skip reference line 3 decision. Please indicate the contribution of each algorithm to the overall result

6  6)     Please provide some figures to compare the quality of the reconstructed frame

[1]: J. Chen et al. “Algorithm description for Versatile Video Coding and Test Model 13”, JVET Document, JVET-V2002-v1, 30th Meeting, Antalya, TR, 21-28 April 2023

A carefull proofreading should also be done to improve the quality of presentation. 

Author Response

1)     The proposed method is totally based on video content. However, to generate the final partition map, the encoder must also consider the input parameters such as QP. Please let me know if the input parameters have any effect.

 

 

In the decision-making process of partitioning patterns, the smaller the qp parameter, the smaller the resulting CU, which will affect the termination partitioning strategy of CU. However, during the termination process, the blocks corresponding to Ta and Te will also be smaller, so this step was not further processed

 

 2)     VVC has some exceptions (as presented in Section 3.2.4 and 3.2.5 in [1]). Please present how the proposed algorithm will handle these exceptions?

Thank you for your correction. After obtaining the MPPPL, modifications were made to the MPPPL according to the partitioning rules. We explained this step in the article

 

3)     Figure 1a and Figure 8 need to be modified. Figure 1a is quite low quality, while there is an error with Figure 8. It should be noted that Fig.8 is important to know how Threshold is determined.

Thank you for your correction. We have replaced the figures in Figure 1 and modified Figure 8 to make them look clearer

4)     Letters in Fig.6 are too small to read.

Thank you for your correction. We have redrawn Figure 6 to make it easier to read

 5)     This paper proposes two algorithms: fast partition mode decision and skip reference line 3 decision. Please indicate the contribution of each algorithm to the overall result

Thank you for your correction. We have added data comparison in the article to further illustrate their performance

• Please provide some figures to compare the quality of the reconstructed frame

Thank you for your correction. We have provided the bitrate changes of the reconstructed frames in the article to demonstrate the quality of the reconstructed frames

Reviewer 3 Report

Authors propose a fast VVC Intra coding algorithm based on MPPL. After a statistical analysis that it indicates the relationship between partitioning and texture complexity.  Based on these results propose an algorithm to construct the MPPPL and a fast decision algorithm by pruning the choices of the lest, when possible. The paper has adequate related work which verifies the contribution of this work. Moreover, authors provide a comprehensive and full analysis of the way whey ended up with the proposed algorithm. 

However, authors should pay more attention to some details as for example how equations are presented. 

1. In Eq. 1 and 2 letter p is used both for denoting pixel and coordinate.   
2. The dominator is the same to all eq. 8 -12. It would be better to use an Equation to express the dominator and use the equivalent instead.
3. In Eq. 262 and 266 min is not properly typed.

The main problem with the proposed algorithm in the manuscript is flow chart in Fig.6

1. In many cases after Mode-a (or b - e)  a symbol with C is present. Please remove it.
2. The last part of the chart would be better to change. It is practical a for loop with early breaking condition. You should not unfold the loop.

 

Furthermore, authors propose a method to prune decisions for reference list when possible. Reviewer believes that this subsection

needs to be seen more carefully. Main problems founded are the following:

1. More analysis about how and why Eq. 19 and 20 measure content similarity. It is necessary to know the intuition that led to these equations.
2. Regarding PCS, is this a vector value? Do you compare each value (horizontal and vertical) with threshold, or there is a way to have a unique value?
3. Define threshold m better.
4. Figure 8 in the manuscript is corrupted. Please change it.

 

Regarding Introduction, it should be noted that this section usually ends with a short description of how the rest of the paper is organized. It highly recommended authors follow this good practice.

Regarding Experimental results, is Ts presented in Tables the average for all QPs ? If yes, it should be mentioned in the paper. If not please specify.

Regarding Conclusions authors mention : “The algorithm uses the characteristics of VVC partition mode decision to construct the MPPPL”. Reviewer believes that this statement is rather short and not precise to describe the work presented in the manuscript. Perhaps, authors should give more emphasis on the use of texture CU info, after a statistical analysis of its impact on CU partitioning.

Finally, as a general comment, there are many syntax errors in the manuscript. As the errors are quite a lot the reviewer cannot list them all. It is highly recommended to the authors to pay more attention the use of English language.

  

There are many syntax errors in the manuscript. As the errors are quite a lot the reviewer cannot list them all. It is highly recommended to the authors to pay more attention the use of English language.

 

Author Response

1. In Eq. 1 and 2 letter p is used both for denoting pixel and coordinate. 

Thank you for your correction. We have made modifications to this

12. The dominator is the same to all eq. 8 -12. It would be better to use an Equation to express the dominator and use the equivalent instead.

Thank you for your correction. We have made modifications to this

 

1. In Eq. 262 and 266 min is not properly typed.

Thank you for your correction. We have made modifications to this

The main problem with the proposed algorithm in the manuscript is flow chart in Fig.6

1. In many cases after Mode-a (or b - e)  a symbol with C is present. Please remove it.
2. The last part of the chart would be better to change. It is practical a for loop with early breaking condition. You should not unfold the loop.

Thank you for your correction. We have redrawn Figure 6 to make it look more reasonable

More analysis about how and why Eq. 19 and 20 measure content similarity. It is necessary to know the intuition that led to these equations.

Thank you for your question. We have further explained the meanings of these two in the text

 Regarding PCS, is this a vector value? Do you compare each value (horizontal and vertical) with threshold, or there is a way to have a unique value?

Thank you for your question. PCS is not a vector, it is a ratio. We will compare it with the threshold m.

Define threshold m better.

Thank you for your error. We have redefined Re for this

Figure 8 in the manuscript is corrupted. Please change it.

Thank you for your error. We have made modifications to this

Round 2

Reviewer 2 Report

The authors have made some revisions. However, there are still some remaining issues:

 

1) It is unclear how the comments #2 was addressed, "VVC has some exceptions (as presented in Section 3.2.4 and 3.2.5 in [1]). Please present how the proposed algorithm will handle these exceptions?", please make clear in both paper and the responses to reviewers;

 

2) In comments #5, "his paper proposes two algorithms: fast partition mode decision and skip reference line 3 decision. Please indicate the contribution of each algorithm to the overall result", it is suggested to include the results for 20 seq in the test set;

 

- "Please provide some figures to compare the quality of the reconstructed frame", here the key point is the quality of the reconstructed frames, not the relevant bitrate.

No comments

Author Response

1) It is unclear how the comments #2 was addressed, "VVC has some exceptions (as presented in Section 3.2.4 and 3.2.5 in [1]). Please present how the proposed algorithm will handle these exceptions?", please make clear in both paper and the responses to reviewers;

Response: Thank you for your question. After the MPPPL construction is completed, we will delete the partitioning modes that are not allowed in MPPPL to avoid conflicts with the partitioning restrictions of VVC according to the partitioning rules. The explanation are added in line226-229

 

 

2) In comments #5, "his paper proposes two algorithms: fast partition mode decision and skip reference line 3 decision. Please indicate the contribution of each algorithm to the overall result", it is suggested to include the results for 20 seq in the test set;

Response:: Thanks for your suggestion. We have given the experimental results of the two methods in Table 3. The descriptions are added in line368-371.

 

 

3)"Please provide some figures to compare the quality of the reconstructed frame", here the key point is the quality of the reconstructed frames, not the relevant bitrate.

Response: Thank you for your suggestion. We have added the quality comparison of the reconstructed frames in figure11. The descriptions are added in line413-416