International organisation for standardisation organisation internationale de normalisation

5.2.33D-HEVC-CE3: Motion Parameter Prediction and Coding (Independent of Depth)

m23605	Teruhiko Suzuki	3D-HEVC-CE3 summary report
m23620	Junghak Nam, Hyomin Choi, Sunmi Yoo, Woong Lim, Donggyu Sim	3D-HEVC-CE3 result on KWU’s advanced motion and disparity prediction method based on global disparity
m23625	Junghak Nam, Hyomin Choi, Sunmi Yoo, Woong Lim, Donggyu Sim	Cross check of 3D-HEVC-CE3 results of HHI from KWU
m23639	Yoshitomo Takahashi, Teruhiko Suzuki	3D-HEVC-CE3 results on motion parameter prediction by Sony
m23640	Yoshitomo Takahashi, Teruhiko Suzuki	Cross check of 3D-HEVC-CE3 results of 3D-HTM
m23641	Yoshitomo Takahashi, Teruhiko Suzuki	Cross check of 3D-HEVC-CE3 results of KWU
m23652	Byeongdoo Choi, Chanyul Kim, Jeonghoon Park	Cross check of 3D-HEVC-CE3 results on joint proposal of HHI and Sony
m23698	Heiko Schwarz, Heribert Brust, Thomas Wiegand	3D-HEVC-CE3 results on inter-view motion parameter prediction by HHI
m23699	Heiko Schwarz	Cross check of 3D-HEVC-CE3 results of LG
m23700	Jaewon Sung, Sehoon Yea	Cross check of 3D-HEVC-CE3 Sony’s vector prediction method by LG
m23701	Jaewon Sung, Sehoon Yea	3D-HEVC-CE3 Results on Multi-view Motion Information Reuse Method
m23817	Yoshitomo Takahashi, Heiko Schwarz	3D-HEVC-CE3 results on joint proposal of HHI and Sony

Summary report:

• 
  Regarding common conditions (N12352), all other tools are turned off from the 3DV-HTM. Each tool for inter-prediction for texture was tested and compared against D-MVP in the 3DV-HTM. Note that the depth components will be always coded at full resolution.
  
• 
  The reference was the 3D-HTM without inter-view motion prediction.
  

KWU proposed transmission of a global disparity vector for each frame, and usage of depth data at the encoder, with gains of 3.5% and 2.5% for the 3-view and 2-view cases.

LG proposed multi-view motion information reuse, with one global disparity per GOP, and had gains of 1.9% and 1.4% for 3-view and 2-view cases.

Sony proposed an extended collocated vector decision, with scaling of the inter-view vector based on ViewOrderIdx, and had gains of 1.5% and 1.0% for 3-view and 2-view cases.

HHI proposed derivation of the disparity map estimate during encoding, downscaling of the depth map estimate, and reported complexity reduction with gains of 4.8% and 3.2% for 3-view and 2-view cases.

A combined Sony+HHI scheme involving a combination of proposals had gains of 5.0% and 3.4% for 3-view and 2-view cases.

Conclusions:

• 
  Include combined proposal in HTM (m23817).
  
• 
  Continue CE.
  
• 
  Focus on benefits of global disparity vector (LG, KWU).