Project development, status, and guidance

Project development, status, and guidance

1. Communication by parent bodies

15.0.1.1.1.1.1.1.16m30385 KNB comment on 3D-AVC depth tools performance [Kyeongok Kang on behalf of the KNB]

1. 3D video application areas

(MPEG Requirements contribution; was presented for information in JCT-3V.)

This contribution encourages the MPEG community to consider the use case that is introduced in this document. This contribution is proposing the MPEG-3DV group to move to a next phase that seeks standard for compression of contents of super/ultra-multiview 3D displays.

MPEG community is currently standardizing a new codec for autostereoscopic displays, where the test input signal is 2view+2depth or 3view+3depth. In this scenario, at most about 30 views can reportedly be generated and tested at the display side. Recently, super/ultra-multiview 3D displays are emerging and reportedly widely, anticipated as the next generation of auto-stereoscopic display. Since these super/ultra multiview 3D displays require a large number of multi-view images in real-time, the contributor asserted that a new coding standard is essential to realize their services and products in the market.

15.0.1.1.1.1.1.1.18JCT3V-E0258 Proposal on a New Activity for the Third Phase of FTV [Masayuki Tanimoto, Takanori Senoh, Marek Domański, Anthony Vetro, Marius Preda] [late]

(MPEG Requirements contribution; was presented for information in JCT-3V.)

The contibutors assert that the standardization of MVC and 3DV has been conducted successfully, and that MVC was the first phase of FTV (Free-viewpoint Television) and 3DV is the second phase of FTV. Based on the recent development of 3D technology, this document proposes a new 3D standardization activity targeting "super multi-view" and free navigation applications for the third phase of FTV.

15.0.1.1.1.1.1.1.19JCT3V-E0232 Multi-layer MVD data coding for DIBR of complex 3D scenes [D. Rusanovskyy (LGE)] [late]

This document consider coding of multilayer MVD data for purposes of enabling DIBR of complex 3D video scenes with transparent and semi-transparent objects through utilization of multi-layer MVD plus Alpha scene representation (MVD+A).

It is proposed to extend the 3DV test set and CTC to include 3D scene with transparent or semi-transparent objects with (MVD+A) data representation.

It is also proposed to establish AhG to study DIBR performance and coding approaches for such type of data.

It is also proposed to enable in 3DV standards support for coding of multi-layer MVD+A data represented in YUV color pipeline (depth and alpha information combined into a single YUV container).

Not intended to use alpha map for better compression.

Such an approach would be applicable using auxiliary pictures (as currently defined in the RExt extension). However, the contribution states that it would be inefficient to use “as is”.due to the necessity of using multiple decoders.

More evidence required how solutions on this could be optimized with existing auxiliary picture concepts or extensions thereof; no test material available currently.

Establish new AHG (D. Rusanovskyy, S. Yea, S. Shimizu) to study “extensions of DIBR concepts using data beyond depth”.

2. Performance evaluation

JCT-3V has not decided whether the Stereo Main profile of MV-HEVC includes the support of AVC-coded base view. To provide input for the decision-making, this contribution presents results on 2-view coding with AVC base view and MV-HEVC non-base view compared against MVC. In the presented results, the Bjontegaard delta bitrate was -18.5% on average.

This would require normative specification. Potentially such a hybrid combination of two desoders (AVC and HEVC) will be specified in the context of SHVC, and it might then simple to also use it for multi-view (would require a new profile).

From the discussion, it is noted that in the current (HEVC only) stereo profile definition the avc_base_flag which exists should be disabled.

3. Asymmetric coding modes

In this document, an enhanced evaluation of impact of coded depth resolution on performance of the current AVC-based 3D video coding technology implemented in 3D-ATM is presented.

Just like in case of previous works presented in JCT3V-D0080, the evaluation has been done with respect to Common Test Conditions in HP and EHP profiles. Two depth configurations has been tested - full depth resolution and half depth resolution. In this work, additionally, comparison has been done for operation point with maximized codec performance for each case. Maximization of coding performance was done by selection of optimal quantization parameter for both texture and depth views. Also, optimized quantization parameters curves were provided.

The results show that coding with half resolution of depth (currently used in CTC) provides better results in comparison to coding with full resolution of depth..

Gain in CTC is around 1.5%, however the new settings are trained within the set of test sequences (but not individually for the single sequences)

It is reported that the gap between full and half res is around 12% in HP and 18% in EHP under normal CTC; this somehow contradicts the number reported under E0245.

The benefit in CTC is not dramatic, and some concern is expressed that this may be over-trained to the set of test sequences

For full-resolution depth maps, there seems to be a systematic tendency that the QP of depth maps should be increased (i.e. lower quality). Further investigate in the context of the quality investigation AHG.

29. 
    
    Directory: sc29 -> open -> 29view
    29view -> International organisation for standardisation organisation internationale de normalisation
    29view -> International organisation for standardisation organisation internationale de normalisation
    
    
    
    Download 8.47 Mb.
    
    Share with your friends: