International organisation for standardisation organisation internationale de normalisation

NB positions

The following NB inputs were reviewed, and dispositions were prepared in coordination with JCT-VC, JCT-3V and other subgroups, as applicable.

1.1.1.1.1.14m32712 Extensions of HEVC [A. G. Tescher for USNB]

(Responded in N14433.)

1.1.1.1.1.15m33117 JPNB Comments on Request for Subdivision: ISO/IEC 14496-10:2012/Amd6: Texture and Depth View Packing SEI Message in 3D-AVC [Takanori Senoh on behalf of JPNB]

(Responded in N14417.)

1.1.1.1.1.16m33346 JPNB comment on resolution 11.2.2 [Japan National Body]

(Responded in N14433.)

1.1.1.1.1.17m33347 JPNB comment on resolution 11.2.7 [Japan National Body]

(Responded in N14433.)

1.1.1.1.1.18m33399 JPNB comments on N14256 (Request for Subdivision: ISO/IEC 14496-31) [Japan National Body]

(Responded in N14432.)

6. Joint Meetings

The following joint meetings were held (see notes, as applicable, in the corresponding sections)

• 
  Requirements, JCT-VC Mon 16-18 SCC, RExt (Requirements Room)
  
• 
  Requirements, JCT-VC, JCT-3V Tue 16-18 SHVC, MV-HEVC etc. (Requirements Room)
  
• 
  3DG, JCT-3V Thu 10-11 AR / 3D Video (3DG Room)
  
• 
  3DG, Video Thu 11-12 AR / CDVS (3DG Room)
  
• 
  Requirements, JCT-VC, Thu 15-17 continuation (Requirements Room)
  
• 
  11:30-12, Wed assets: As a follow-up action, an AHG was installed to collect information about test sequence assets and transfer them to the MPEG SVN.
  
• 
  17-18, Wed editing advice
  

7. Output Documents planned

A provisional list of output documents was set up in the opening plenary and further updated during the Wednesday plenary. See final list of output documents under sec. 11.1.

2. MPEG-7 Visual XM software part 6

1.1.1.1.1.19m33434 MPEG-7 eXperimentation Model Update and Testing [Karol Wnukowicz, Stavros Paschalakis]

Considerable progress had been made in the context of the AHG work on updating the Visual XM software. A WD for a new edition was issued as N14390.

3. CDVS

The BoG on CDVS was chaired by Miroslaw Bober. The notes in this section were primarily recordings from the BoG, which met Monday-Thursday, usually for the whole day (except video plenary times).

The DIS was issued as N14392 with DoC N14391.

1. CE and other technical inputs

   1. CE1: Key point extraction

1.1.1.1.1.20m33159 CDVS CE1: A low complexity detector ALP_BFLoG [Jie Chen, Ling-Yu Duan, Tiejun Huang, Wen Gao, Alex C. Kot, Massimo Balestri, Gianlucca Francini, Skjalg Lepsoy]

Key objective: to reduce memory and achieve speedup of the ALP detector and descriptor extraction module. Memory reduced to 1.29MB and 14% faster then TM9 @ 1.29MB.

Some filters were moved to the frequency domain. Maxima detection was performed on sets of 3 lines: 312kb-> 9kB.

The added benefit is that block-based approach can be executed in parallel on multiple processors.

Algorithmically equivalent to the default TM9 ALP with filtering done in the frequency domain.

Crosschecks successful: M33215, M33099.

Recommendation: Include as fast option in the TM software. No changes to the text of the CVDS standard required.

1.1.1.1.1.21m33099 CDVS: Cross-Check of Peking University’s Proposal m33159 [Zheng Liu, Giovanni Cordara, Bin Chen]

Crosscheck successful.

1.1.1.1.1.22m33215 CDVS: Crosscheck of m33159 [Alessandra Mosca]

Crosscheck successful.

2. CE2: Global descriptor

1.1.1.1.1.23m33189 CDVS CE2: Improved SCFV [Zhe Wang, Ling-Yu Duan, Jie Lin, Tiejun Huang, Wen Gao, Alex C. Kot, Miroslaw Bober]

The document reported improvements to SCVF. Two changes were proposed: (1) variance within each GMM is used to select clusters improving performance. The number of Gaussian functions was increased to 512.

Number of local vectors used for aggregation reduced to 250 (fixed) - reducing aggregation complexity.

Some increase in memory: 21k->41k. Improvements to mAP +0.87%, TOP +0.55%. No change to TPR and localisation. Crosscheck successful: M33098.

The group needs to understand & balance complexity increase with performance increase. Following AHG the proponents started simulations to provide profiling info (TM8, TM9 and proposal), including performance for 256 cluster centers with the new selection method and 300 local vectors.

Some initial profiling results were provided, indicating that Global descriptor from TM9 is about 3x slower compared to TM8.

Further results were presented showing that GD 512 GMM functions offers +0.87% mAP over the 256

	mAP	TOP	TPR	Loc
TM9	83.72%	90.16%	93.22	0.8225
M33189 (512)	84.59%	90.71%	93.18	0.8185
256 with new selection and 300 vectors	83.56%		93.57

The group decided to adopt m33189 into the DIS.

1.1.1.1.1.24m33098 CDVS: Cross-Check of Peking University’s Proposal m33189 [Zheng Liu, Giovanni Cordara, Bin Chen]

Crosscheck successful.

3. Other

1.1.1.1.1.25m33160 CDVS: Evaluate the performance of MBIT algorithm in TM9.0 [Zhe Wang, Ling-Yu Duan, Jie Lin, Tiejun Huang, Wen Gao, Alex C. Kot]

Speedup confirmed, up to 15 times with 1% loss of mAP performance.

1.1.1.1.1.26m33342 CDVS: Cross-check of m33160 [Keundong Lee, Seungjae Lee, Sang-il Na]

Crosscheck successful.

1.1.1.1.1.27m33447 CDVS: Cross-Check of Peking University’s Proposal m33160 [Zheng Liu, Giovanni Cordara, Bin Chen]

Crosscheck successful.

1.1.1.1.1.28m33489 CDVS: Editor’s proposed improvements on Study text of ISO/IEC CD15938-13 Compact Descriptors for Visual Search.

The group reviewed the recommendation from the editors and accepted all except for the introduction of the ConvertedImageXResolution and ConvertedImageYResolution descriptor components.

2.