14.1.1.1.1.1.1.1.52m31396 CDVS: extended dataset for feature selection training [Gianluca Francini, Massimo Balestri, Skjalg Lepsoy]
Extended dataset for training the feature selection, in order to avoid the problems with copyright protected data we had so far. The new training dataset is be made available to the group, with crosschecked results, and already used for the selection of ALP keypoints.
14.1.1.1.1.1.1.1.53m31652 Verification test related to m31396 [Alessandra Mosca, Massimo Mattelliano]
Croscheck noted.
14.1.1.1.1.1.1.1.54m31420 CDVS: enhancements to m31369 [Massimo Balestri, Gianluca Francini, Skjalg Lepsoy]
Further improvements of ALP keypoint detector going beyond the Core Experiments. In particular, multithreading extraction, fast scalar quantization, improvements in the feature extraction part, reduced block width and bitrate control mechanism. Slight improve in localization and software optimization. Only software improvements, will be integrated into the new TM, that will be released one week after the meeting.
14.1.1.1.1.1.1.1.55m31488 CDVS: Retrained Feature Selector using updated dataset [Keundong Lee, Sang-il Na, Seungjae Lee, Weon-Geun Oh]
Confirmation of results for new training dataset for feature selection.
14.1.1.1.1.1.1.1.56m31489 CDVS: block-wise approach of SCLoG [Sang-il Na, Seungjae Lee, Keundong Lee, Weon-Geun Oh, Insu Won, Dong-Seok Jeong]
Block based implementation – descriptor part is M31486. Simple to understand. Easy to implement.
Comparable performance to BFLoG. Interoperable with M31486.
No cross-check.
WD and TM Development
14.1.1.1.1.1.1.1.57m31482 Evaluation of the MPEG CDVS TM7.0 performance vs state of the art [Syed Husain, Miroslaw Bober, Stavros Paschalakis, Karol Wnukowicz]
Document withdrawn.
14.1.1.1.1.1.1.1.58m31615 CDVS: interest point detection, a concise text to help further editing [Gianluca Francini, Massimo Balestri, Skjalg Lepsoy]
The document text used as a starting point for CD drafting.
Implementation study
14.1.1.1.1.1.1.1.59m31313 CDVS: STM Detector HW accelerator feasibility study and demonstrator [Thomas Boesch, Giuseppe Desoli, Gian Domenico Licciardo, Mario Vigliar, Claudio Parrella, Danilo Pau, Arcangelo Bruna]
FPGA draft accelerator. Complexity analysis: detector takes approx. 60% of the time. GOPS/mm2/W two orders of magnitude compared to embedded CPU at 32nm node. One pipeline, 10fps.
CE definition
Continuing CEs in a collaborative approach to further stabilize/harmonize the entire chain (see N13927)
-
CE1 investigating simpler and more precise specification of filters and other parts of the keypoint extraction process, also investigate possibility of interoperability of different filters, which eventually would not make it necessary to specify them precisely.
-
CE2 (global descriptor) – further assess benefits of the recently proposed simplifications
-
CE3 (new) looks into possibility of not specifying the keypoint detection in a normative way
Internet Video Coding Exploration
The BoG on IVC was chaired by Kiho Choi. The notes in this section were primarily recordings from the BoG, which met Monday-Thursday, usually whole day (except during plenary times and during times where subjective viewing was performed).
The following issues were initially identified to be resolved during the meeting:
-
Visual testing: Informal expert viewing for assessing the status of IVC in terms of visual quality.
-
Compare IVC LD, IVC RA, AVC BP LD (old anchors), AVC HP RA (anchors from recent CFP). Purpose of this is not reaching the HP quality by IVC RA, but rather having a second comparison point to know what the gap compared to non-type-1 technology is.
-
Find out whether it is possible to make a BD rate comparison between the sequences of ITM4 of January (when the last informal expert viewing was made and the current status)
-
Assessment of prior-art status of new proposals still to be made (which should include the analysis that the reported prior art implements the method identically).
The following input documents were reviewed and recordings of discussions were made in the BoG.
General
14.1.1.1.1.1.1.1.60m31556 Technology-IPR mapping management for Type-1 standard [Kiho Choi, Euee S. Jang]
-
Summary
-
This contribution proposes technology-IPR mapping management for Type-1 coding standards.
-
To manage the adopted tools and the corresponding IPR information, this contribution suggests that a document containing the IPR information of adopted technologies should be maintained.
-
Comments
-
This should be proposed to the video plenary.
-
As a subsequent action, the output document “Collection of information related to adopted IVC technologies” (N13946) was approved.
14.1.1.1.1.1.1.1.61m31626 Information of IVC anchors for subjective assessment test [Sang-hyo Park, Kiho Choi, Minsoo Park, Euee S. Jang]
-
Summary
-
This contribution provides the bitstream of the anchor of IVC for subjective assessment test.
-
As a subsequent action, subjective viewing was performed.
EE
14.1.1.1.1.1.1.1.62m31557 IVC-EE: Reference frame improvement [Minsoo Park, Kiho Choi, Jin Yeon Choi, Sang-hyo Park, Euee S. Jang]
-
Summary
-
This contribution presents a method that uses more than one frame as the reference frame in the random access case.
-
This contribution extends the use of MRF of P-frame to the random access case.
-
The experimental result shows that the proposed method achieves 4.6% in coding gain on average compared to that of ITM 6.0. Comments
-
Due diligence has been reported by providing the related IPR information.
-
Comments
-
How to resolve “Random Access” problems involved by MRF?
-
Conclusion
-
Adopt it in the condition to refine an I-frame as closed key frame.
-
Adopt bug-fix one.
14.1.1.1.1.1.1.1.63m31597 IVC-EE: Cross-check of Reference frame improvement (m31557) [Dong-Hyun Kim, Jae-Gon Kim]
-
Summary
-
This contribution is cross verification for “IVC-EE: Reference frame improvement” of M31557.
-
The results of cross verification are exactly matched with the provided results by the proponents.
14.1.1.1.1.1.1.1.64m31515 Smoothed Planar Intra prediction for Internet Video Coding [Dong-Hyun Kim, Jae-Gon Kim]
-
Summary
-
This document proposes a smoothed planar intra prediction to improve the coding efficiency of intra coding in ITM 6.0.
-
Experimental results show that the proposed scheme gives an average of 1.9% and 0.7% bit saving in CS1 and CS2, respectively. It is recommended to adopt this proposal to ITM 7.0.
-
Due diligence has been reported by providing the related IPR information.
-
Comments
-
Q: Is there any IPR investigation?
-
A: The information is provided in the contribution.
-
Q: Is the method similar with that in HEVC? What is a difference compared to planar mode in HEVC?
-
A: The proposed method is similar with that in HEVC. The proposed method is based on the basic concept of planar prediction with different prediction algorithms.
-
Q: Do you have syntax changes?
-
A: There are no syntax changes but there are changes in decoding process.
-
Conclusion
-
Further investigation on prior-arts is necessary on this.
14.1.1.1.1.1.1.1.65m31558 IVC-EE: Cross check of m31515 (Smoothed Planar Intra prediction for Internet Video Coding) [Kiho Choi, Minsoo Park, Sang-hyo Park, Euee S. Jang]
-
Summary
-
This contribution is cross verification for “Smoothed Planar Intra prediction for Internet Video Coding” of M31515.
-
The results of cross verification are exactly matched with the provided results by the proponents.
-
Summary
-
This contribution is cross verification for “Smoothed Planar Intra prediction for Internet Video Coding” of M31515.
-
The results of cross verification are exactly matched with the provided results by the proponents.
14.1.1.1.1.1.1.1.66m31516 Improvement of intra prediction with multiple modes for Internet Video Coding [Dong-Hyun Kim, Jae-Gon Kim]
-
Summary
-
This document proposes a multiple-mode intra prediction to improve the coding efficiency of intra coding in ITM 6.0.
-
The proposed scheme uses four prediction modes for 8x8 block size: DC, planar, horizontal, and vertical.
-
Experimental results show that the proposed scheme gives an average of 4.4% and 0.2% bit saving in CS1 and CS2, respectively. It is recommended to adopt this proposal to ITM 7.0.
-
Due diligence has been reported by providing the related IPR information.
-
Comments
-
Q: The test result seems to be strange because of the result in Class 0, A, and B.
-
A: The investigation of the reason would be needed.
-
Q: How much the gain from DC mode changes?
-
A: 0.2 % gain in all intra case.
-
Q: Is this meaning 0.2 % gain while increasing computational complexity?
-
Conclusion
-
The modified DC mode and is adopted on the next version of ITM with the intra prediction of the m31256.
-
Further study necessary on IPR status of smooting filtering in intra prediction
14.1.1.1.1.1.1.1.67m31559 IVC-EE: Cross check of m31516 (Improvement of intra prediction with multiple modes for Internet Video Coding) [Minsoo Park, Kiho Choi, Jin Yeon Choi, Sang-hyo Park, Euee S. Jang]
-
Summary
-
This contribution is cross verification for “Improvement of intra prediction with multiple modes for Internet Video Coding” of M31516.
-
The results of cross verification are exactly matched with the provided results by the proponents.
Dostları ilə paylaş: |