International organisation for standardisation organisation internationale de normalisation

5.1.23D-AVC-CE2: Depth-based prediction

m23787

Olgierd Stankiewicz

3D-AVC-CE2 summary report: Depth-based prediction

Summary report: There was insufficient time to complete the work among the proponents. The proponents recommended continuing the CE period until the next meeting.

Conclusion:

• 
  Continue CE.
  
• 
  Coordinator: Olgierd (Poznan).
  

5.1.33D-AVC-CE3: Depth representation and coding

m23785	Olgierd Stankiewicz	3D-AVC-CE3 summary report: Depth Representation & Coding
m23659	Byung Tae Oh, Jaejoon Lee, Du Sik Park	3D-AVC-CE3 results on Samsung adaptive depth quantization
m23670	HoCheon Wey, Jaejoon Lee, Du Sik Park	Cross check of 3D-AVC-CE3 Poznan Univ's depth representation by Samsung
m23788	Olgierd Stankiewicz, Krzysztof Wegner, Maciej Kurc	3D-AVC-CE3 results on Nonlinear Depth Representation & Coding
m23790	Olgierd Stankiewicz, Krzysztof Wegner, Maciej Kurc	3D-AVC-CE3 cross-check of adaptive depth quantization
m23791	Olgierd Stankiewicz, Krzysztof Wegner, Marek Domanski	Nonlinear depth representation – extended results
m23904	Fang-Chu Chen, Ching-Chieh Lin, Jih-Sheng Tu	Cross check of 3D-AVC-CE3 results on Poznan's proposal by ITRI

Summary report (3 contributions):

• 
  Poznan considered a unified depth representation that was very similar to the joint view filter which is already supported in software; as a result, consideration of this was abandoned.
  
• 
  Poznan considered a non-linear depth representation with an aim to improve subjective quality by changing the allocation of bits between foreground and background in the depth representation. Some mismatches were reported in the cross-checks and were likely due to view synthesis operations. Subjective evaluation was recommended. Extended results were reported in m23791.
  
• 
  Samsung proposed an adaptive depth quantization based on corresponding texture. This would assign more bits for complex textures and fewer bits for non-complex textures. There reportedly 1.48% and 1.15% gain in bit rate for 2-view and 3-view cases. The complexity increase was relatively low – not more than 10%.
  
• 
  The CE report recommended viewing the results of non-linear depth representation. It may be interesting to study the combination of both tools as part of a future CE.
  

m23791 (Poznan): this document was provided for information. It provides an extended subjective evaluation of HEVC related results of Poznan response to CfP and non-linear depth representation. The findings reportedly show that 10-40% of the gains shown by the original proposal come from non-linear depth representation. The tool provides approximately 12% bit rate reduction of depth with no computational cost.

Subjective viewing of non-linear depth representation results for AVC was conducted, and it was found that the subjective quality was approximately indistinguishable, while the bit rate of proposed tool was lower (up to 20% lower bit rate for depth, 6% on average for depth) – the result seemed inconclusive for now.

Conclusions:

• 
  Continue CE.
  
• 
  Evaluate adaptive depth quantization.
  
• 
  Evaluate non-linear depth representation.
  
• 
  Evaluate the combination of adaptive depth quantization and non-linear depth.
  
• 
  Consider the potential to use non-linear depth representation as pre/post-processing.
  
• 
  Clarify the metric to be used for evaluation: subjective and/or BD rate.
  
• 
  Coordinator: Olgierd (Poznan)