Witness the development of H.265

Next Step of JCT-VC: Scalable Extensions of HEVC

2012-12-24 SHVC View Comments Views(9,887)

As the first phase of HEVC standard (single layer video coding standard) comes close to finalization, scalable extensions of HEVC have been under discussion since the 7th JCT-VC meeting. The final call for proposal (CfP) [1] was issued jointly by ITU-T/VCEG and ISO/IEC/MPEG in July 2012 after the 10th JCT-VC meeting, and companies and organizations are invited to submit proposals in response to this CfP.

The CfP includes two categories using HEVC and H.264/AVC as the base layer formats, respectively. In each category, the coding parameters and conditions are specified differently, such as scalability type (spatial scalability and SNR scalability), base/enhancement layer spatial resolution ratio (1.5x and 2x), and coding configuration (all intra (AI) and random access (RA)). In combination of differet coding conditions, there are in total seven sets of performance data: five for Category 1, using HEVC base layer, (AI 2x, AI 1.5x, RA 2x, RA 1.5x, and RA SNR) and two for Category 2, using H.264/AVC base layer, (RA 2x and RA 1.5x). According to the CfP, a response may provide as few as only one set of performance data and as many as all seven sets. The test sequences are HD and ultra-HD videos, of which the resolutions are shown as below. They can be downloaded from the ftp site ftp.tnt.uni-hannover.de .

 

 

Test Sequence

Original

2x base layer

1.5x base layer

Class A

PeopleOnStreet

3840×2160

1920×1080

N/A

Traffic

3840×2160

1920×1080

N/A

Class B

BasketballDrive

1920×1080

960×540

1280×720

BQTerrace

1920×1080

960×540

1280×720

Catus

1920×1080

960×540

1280×720

Kimono

1920×1080

960×540

1280×720

ParkScene

1920×1080

960×540

1280×720

 

Responding the CfP, 18 organizations (Canon, ETRI, Ghent Univ., HHI, Huawei, Intel, InterDigital, KDDI, Kwangwoon Univ., LG, MediaTek, Nokia, Qualcomm, Samsung, Sharp America, Sharp Japan, TI, and Vidyo) submitted 20 proposals by the deadline of 10/1/2012; all were reviewed in the 11th JCT-VC meeting, held in Shanghai CN from Oct. 9 to 19 2012. The following figure gives a general idea of how each proposal performs under the random access configuration with HEVC base layer, where the x-axis shows the proposal numbers and the y-axis shows the bit-rate saving. Interested readers are referred to [2] for more detailed performance comparison and the summary of the coding tools used in each proposal. 

 

untitled

 

To pursue good performance numbers, many proposals, especially the ones with “leading” performance numbers, applied single layer coding tools, such as adaptive loop filter (ALF), internal bit depth increase (IBDI), multi-hypothesis prediction, and optical flow, and encoder-side optimization, such as adaptive GoP sizes and lambda values, to improve the enhancement layer coding efficiency, and such coding tools are actually out of the scope of SHVC standardization. The SHVC development plan will only consider core scalable coding technologies. The performance difference among the proposals is not clear, if the effect of the single layer coding tools and encoder-side optimization is taken out and only core scalable coding technologies are considered.

Finally, the committee could not pick out the winning proposal, of which its software is supposed to serve as the starting point for future SHVC development, at the meeting without proper testing and rigorous evaluation. Instead, the committee agreed to start from the minimalist approach including only one coding tool. That tool is essential to inter-layer prediction, which uses a block level flag to indicate whether the current enhancement layer block is predicted using the texture data of the co-located base layer block. The software starting point, realizing this simple and clean approach, is named SMuC for “SHVC test Model under Consideration” and was released one month after the meeting [Check out SMuC].

The individual coding tools used in the 20 proposals were roughly classified into six categories, and planned to be further investigated in six corresponding tool experiments (TE), with SMuC serving as the code base. The experiments are named as TE instead of CE (core experiments), because the test model for SHVC, which is the SHVC reference software, has not been formed yet. According to the published timeline, the SHVC test model will be formed at the 12th meeting by examining the coding tools in the TEs and adopting a subset of them to be integrated.

  • TE1: Upsampling filter for SHVC
  • TE2: Inter-layer texture prediction signaling in SHVC
  • TE3: Combined prediction in SHVC
  • TE4: Inter-layer filtering in SHVC
  • TE5: Inter-layer syntax prediction using HEVC base layer
  • TE6: Inter-layer syntax prediction using AVC base layer

The scalable extensions of HEVC has been named SHVC for “Scalable High-efficiency Video Coding”, and its standardization will continue under JCT-VC. The tentative timeline of SHVC is shown as below. 

  • January 2013, Test Model
  • April 2013, first Working Draft
  • June 2014, final standard draft

[1] ISO/IEC JTC-1/SC29/WG11 w12957, Joint Call for Proposals on Scalable Video Coding Extensions of High Efficiency Video Coding (HEVC), July 2012.
[2] A. Segall, “BoG report on HEVC scalable extensions”, JCT-VC document, JCTVC-K0354, Oct. 2012, Shanghai, China.

Permanent Link: Next Step of JCT-VC: Scalable Extensions of HEVC

Post Comment

*
To prove you're a person (not a spam script), type the security word shown in the picture.
Anti-spam image