6Explorations
After the Draft Call for Proposals on multiview video coding, as issued by the Hongkong meeting (N6910), a decision was to be made to proceed with the Call according to the status of availability of suitable test sequences (including camera parameters). Before and at the Busan meeting, new test sequences were provided by 6 companies. Therefore, it was decided to proceed and issue a Preliminary CfP (N7094). The available material was extensively reviewed and it was decided to define four different scenario classes, including rectified and non-rectified sets, linear and half-circle camera arrangements:
-
linear setting, calibrated and registered (rectified) for 3D display
-
linear 2D setting, calibrated and registered (rectified) for 3D display
-
half circle setting, calibrated
-
linear convergent setting, (no preprocessing).
The rectified sets are useful for applications supporting view adaptation on stereoscopic 3D displays (t.b.d. whether stereo displays will be used during the tests), while other scenarios would rather support look-around views. As the new test sequences were only available shortly before the Busan meeting, it was not yet possible to define the actual rates at which the CfP results will have to be submitted. This will be tested within the AHG period until the Poznan meeting, taking into account the results gained by AVC anchors with appropriate settings (preliminary definition of settings are given in the Pre-Call, with indication that they will be refined). A detailed description of the evaluation procedure, material to be sent by proponents etc. is also provided.
Two technical input documents reported improvements of multiview video coding technology, as variants of methods that had been presented by previous meetings.
Documents reviewed:
a) Related to the CfP and test sequences
11829
|
Fons Bruls
|
Testing & evaluation of a possible 3DAV stereoscopic extension
Proposal to evaluate N-views by randomly selecting a stereoscopic pair – claim importance of stereoscopic setup for the application. May be useful to perform visual comparison with shutter glass system – to be clarified by Tobias whether this can be implemented without any problems.
|
11894
|
Janine Daase
Ullrich Goelz
Peter Kauff
Karsten Mueller
Oliver Schreer
Aljoscha Smolic
Ralf Tanger
Thomas Wiegand
|
Fraunhofer HHI test data sets for MVC
|
11981
|
Daehee Kim
Kugjin Yun
Sukhee Cho
Namho Hur
Soo In Lee
|
Remarks about Implementing Multi-view Video Codec
Problem of rectification: holes at the edges, distortions. Rectification better to be used as tool for rendering or post processing, not proper as input for video coding. Definition of spatial prediction structures with up to two temporal and one spatial vector, provides structured rule for GOP and picture ordering. Requirements for camera parameters and camera geometric relation information (to decide on a high level which camera views to select).
|
12003
|
Ted Square
Tsuhan Chen
|
CMU Ultimate Eye Multi-Camera Data
|
12022
|
Masayuki Tanimoto
Toshiaki Fujii
Terumasa Aoki
|
Test Sequences for 3DAV Call for Proposal on Multiview Video Coding
|
12030
|
Toshiaki Fujii
Masayuki Tanimoto
|
Comments on Input and Output Format of MVC
Different camera and display methods are explained.
|
12077
|
Anthony Vetro
Morgan McGuire
Wojciech Matusik
Alexander Behrens
Jinho Lee
Hanspeter Pfister
|
Multiview Video Test Sequences from MERL
|
b) other technical input
11916
|
Yo-Sung Ho
Seung-Uk Yoon
Sung-Yeol Kim
|
Preliminary Results for Multi-view Video Coding using Layered Depth Image
LDI usually constructed from CG, but here it is applied to natural video. Key issue is quality of the depth maps (MS test set provides high-quality depth maps and camera parameters). These were estimated automatically. Raw amount of data is significantly lower than for simulcast or separate representation of texture and depth, but conventional video coding can be applied.
|
11946
|
Kwanghoon Sohn
Yontae Kim
Hyungap Seo
Jungdong Seo
Changseob Park
Jaeho Lee
Shinil Chung
Junyong Lee
Seungjin Nam
|
Multi-view video coding using inter-view balanced disparity estimation
GGOP contains GOPs for all different views. Only base GOP (for one view) includes I frame. Scalable in terms of number of views. IBDE: Balancing parameters are estimated to compensate illumination variations across views. This is also used in disparity estimation; improves the quality of estimated disparity and of interpolated views.
|
Output documents:
No.
|
Title
|
TBP
|
Available
|
|
3D AV
|
|
|
7094
|
Preliminary Call for Proposals on Multi-View Video Coding
|
Yes
|
05/04/22
|
Share with your friends: |