International organisation for standardisation organisation internationale de normalisation

5.1.93D-AVC-CE9: Texture-based prediction for depth coding

m23683	Jaejoon Lee	3D-AVC-CE9 summary report on texture based prediction for depth coding
m23663	Jin Young Lee, HoCheon Wey, Jaejoon Lee, Du Sik Park	3D-AVC-CE9 results on Samsung’s depth-block skip method using a temporal correlation of texture images
m23841	J. Jung, E. Mora	Cross check of 3D-AVC-CE9 results of Samsung

Summary report: There was one proposal from Samsung to evaluate texture-based prediction for depth coding. The technique is based on temporal correlation of texture images, whereby depth blocks are skipped adaptively.

The performance was compared with inside view motion vector prediction (IVMP). The proposed method reportedly achieves 1.03% and 0.74% gain for the decoded texture views and 0.79% and 0.46% for the synthesized views for 2 view and 3 view cases on average when IVMP is turned off. The gains of the proposed method are much smaller relative to the case that when IVMP is turned on.

The proponents recommended adopting this proposal into the 3D-ATM.

The proposal requires temporal analysis of the difference of two frames to infer whether a block should be skipped or not. No skip flag is transmitted as part of the bitstream.

There were concerns that the complexity is rather high relative to the gains. Also, there seemed to be an impact on error resilience. It was also remarked that there is no RD guarantee that the coding efficiency performance would improve. The group was not aware of any tools in single-view decoding that require a block-level coding mode decision to be based on data or information from other access units.

A cross-check was conducted by Orange (as reported in m23841).

Conclusions:

• 
  Discontinue CE
  
• 
  As a general design principle, parsing should not be dependent on signal processing applied to decoded pictures.