International organisation for standardisation organisation internationale de normalisation

ITM6 results viewing

(Viewing was conducted by Sang-hyo Park based on advice by Vittorio Baroncini (for viewing test environment) and Pierrick Philippe (for cross-checking of the results).

1. 
   Participants
   

1. 
   7 participants from 5 different organizations
   

	Name	Organization
Viewer1	Kiho Choi	Hanyang University
Viewer2	Hyungyu Kim	Hanyang University
Viewer3	Pierrick Philippe	Orange
Viewer4	Yun He	Qinghua University
Viewer5	Ronggang Wang	Peking University
Viewer6	Dong-Hyun Kim	Korea Aerospace University
Viewer7	Jae-Gon Kim	Korea Aerospace University

2. 
   Bitrate of sequences
   

Sequences and bit rate cases were selected where IVC coding came close to the rates of the AVC (HP CS1 and CBP CS2) anchors that had been produced for the recent CfP.

	Target rate (kbit/s)	Anchor (CBP CS2)	ITM 6.0 CS1	Diff	ITM 6.0 CS2	Diff	Anchor (HP CS1)	Diff
Kimono	2400	2462.62	2492.34	1.2%	2462.62	0.0%	2493.93	1.3%
	1000	1011.05	1046.66	3.5%	1023.28	1.2%	997.34	-1.4%
ParkScene	4000	4058.98	4116.4	1.4%	4032.25	-0.7%	3987.78	-1.8%
	1600	1637.77	1597.16	-2.5%	1606.30	-1.9%	1589.84	-2.9%
BasketballDrill	2600	2665.67	2693.24	1.0%	2672.41	0.3%
	760	759.50	766.7	0.9%	766.05	0.9%	766.84	1.0%
RaceHorses	1800	1874.36	1815.13	-3.2%	1800.56	-3.9%	1996.27	6.5%
	765	764.23	744.01	-2.6%	734.25	-3.9%	765.84	0.2%

3. 
   Coding conditions
   

AVC BP HR	AVC Baseline profile with high rate (in N12204)
AVC BP LR	AVC Baseline profile with low rate (in N12204)
ITM CS1 HR	ITM 6.0 Constrained set 1 with high rate (in N12204)
ITM CS1 LR	ITM 6.0 Constrained set 1 with low rate (in N12204)
ITM CS2 HR	ITM 6.0 Constrained set 2 with high rate (in N12204)
ITM CS2 LR	ITM 6.0 Constrained set 2 with low rate (in N12204)
AVC HP HR	AVC High profile with high rate (in N13546)
AVC HP LR	AVC High profile with low rate (in N13546)
Original YUV

4. 
   Test results
   

The following results give an informal indication about the performance of the codecs. Neither can the quantitative 1-5 axis be interpreted as MOS values, nor can the bars be interpreted as confidence intervals (they exist, because the software used to produce the viewgraphs automatically plots them, but the number of participants is too low to actually compute confidence values).

1. 
   HD 1080p case (class A)
   

2. 
   WVGA case (class B)
   

As a general conclusion, several experts expressed an opinion that for the cases viewed the quality of IVC has significantly increased during the last year. Whereas the viewing of ITM3 that had been performed during the January 2013 meeting unveiled clearly worse performance than the AVC CBP anchors used by that time, the quality of ITM6 is now roughly similar in many of the cases.

5. 
   Configuration
   
   1. 
      Anchor (AVC BP CS2)
      

Conformance with Constrained Baseline Profile IpPpP coding structure (num_reorder_frames=0, P denotes reference frames, p non-reference) hierarchical 4-frame GOP QP assignment: QP for I, QP+1 for layer 0 P, QP+4 for layer 1 P, and QP+5 for layer 2 p) 2 reference pictures (max_ref_frames = 2) RD Optimization enabled Fast motion estimation (range 64x64) RDOQ disabled Adaptive rounding enabled Frame-level multipass optimizations disabled

2. 
   ITM CS1
   

Constraint set 1: structural delay of processing units not larger than 8-picture "group of pictures (GOPs)" and random access intervals of 1.1 seconds or less. NumberReferenceFrames = 1 # Number of previous frames used for inter motion search NumberBFrames = 2 or 3 # Number of B frames inserted FrameSkip = 2 or 3 # Number of frames to be skipped PSubType = 0 # using non-reference P frame coding

3. 
   ITM CS2
   

Constraint set 2: no picture reordering between decoder processing and output, with bit rate fluctuation characteristics and no multi-pass encoding. A metric will be applied to measure bit rate fluctuation characteristics. NumberReferenceFrames = 5 # Number of previous frames used for inter motion search NumberBFrames = 0 # Number of B frames inserted FrameSkip = 0 # Number of frames to be skipped

4. 
   AVC HP
   

Conformance with High Profile Hierarchical B pictures IbBbBbBbP (8) coding structure – each picture uses at most 4 reference pictures in each list for inter prediction Open GOP structuring with an Intra picture every 24, 32, 48 and 64 pictures for 24 fps, 30 fps, 50 and 60 fps sequences, respectively num_reorder_frames = 3 ("GOP length 8") max_ref_frames = 4 QP scaling: QP (I picture), QP+1 (P picture), QP+2 (first B layer), QP+3 (second B layer), QP+4 (third B layer) CABAC, 8x8 transforms Flat quantization weighting matrices RD Optimization enabled RDOQ enabled (fast mode, NUM=1) Adaptive rounding disabled Weighted prediction enabled Fast motion estimation (range 128x128)

7.