|
|
| səhifə | 14/53 | | tarix | 31.12.2018 | | ölçüsü | 4,04 Mb. | | #88583 |
|
CE2.4 Adaptive Loop Filters
Test#
|
Description
|
Document#
|
AhG13
|
Coding gain of ALF in BMS
|
JVET-K0013
|
2.4.1.1.a
|
Subsampled sum-modified-Laplacian with 4×4 level classification
|
JVET-K0164
|
2.4.1.1.b
|
Subsampled sum-modified-Laplacian with 2×2 level classification
|
JVET-K0164
|
2.4.1.2
|
Adaptive Loop Filter Simplification
|
JVET-K0327
|
2.4.1.3
|
ALF with Multiplication Replaced by Bit-Shifting
|
JVET-K0215
|
2.4.1.4.a
|
luma 7×7, classifier 2×2, chroma 5×5
|
JVET-K0371
|
2.4.1.4.b
|
luma 7×7, classifier 2×2, chroma 7×7
|
JVET-K0371
|
2.4.1.4.c
|
luma 7×7, classifier 4×4, chroma 5×5
|
JVET-K0371
|
2.4.1.4.d
|
luma 7×7, classifier 4×4, chroma 7×7
|
JVET-K0371
|
2.4.1.4.e
|
luma 9×9, classifier 2×2, chroma 5×5
|
JVET-K0371
|
2.4.1.4.f
|
luma 9×9, classifier 2×2, chroma 7×7
|
JVET-K0371
|
2.4.1.4.g
|
luma 9×9, classifier 4×4, chroma 5×5
|
JVET-K0371
|
2.4.1.4.h
|
luma 9×9, classifier 4×4, chroma 7×7
|
JVET-K0371
|
2.4.1.4.i
|
deblocking TC offset setting (-6 for AI, and TC offset -2 for RA and LB)
|
JVET-K0371
|
2.4.2.1.a
|
Multiple-feature based adaptive loop filter (MCALF)
with 4 classifiers (whole proposal)
|
JVET-K0285
|
2.4.2.1.b
|
MCALF with two classifiers  (GALF classification) and 
(sample-based classification, like BO)
|
JVET-K0285
|
2.4.2.1.c
|
MCALF with two classifiers and 
(ranking with sample-based classification, 3×3 pattern)
(counting number of samples that are larger or smaller)
|
JVET-K0285
|
2.4.2.1.d
|
MCALF with two classifiers and 
(ranking with local variation-based, cross-sign pattern)
(checking difference in neighbourhood)
|
JVET-K0285
|
2.4.2.2.a
|
CTB-based ALF with slice filter sets, 3 classification methods (whole proposal)
( IntensitySA (like BO), SimilaritySA(5x5 diamond),
and GeometricBA (GALF classification))
|
JVET-K0235
|
2.4.2.2.b
|
CTB-based ALF with slice filter sets, only IntensitySA and GeometricBA
|
JVET-K0235
|
2.4.2.2.c
|
CTB-based ALF with slice filter sets, only GeometricBA
|
JVET-K0235
|
2.4.2.2.d
|
CTB-based ALF with slice filter sets, only IntensitySA
|
JVET-K0235
|
2.4.2.2.e
|
CTB-based ALF with slice filter sets, single filter is selected at CTB
(including merge with neighbour signalling)
|
JVET-K0235
|
2.4.2.3
|
Unified Adaptive Loop Filter for Luma and Chroma
|
JVET-K0132
|
2.4.2.4
|
Modified ALF classification: horizontal and vertical gradients
are calculated using Sobel filter
|
JVET-K0151
|
Properties of Methods:
Test#
|
Block classification granularity
|
|
# Samples for 1-D Laplacian value calculations
(worst case)
|
Line buffer size
|
Filter supports
|
FF
|
Filter
|
Max. Num. of filters per slice to be stored
|
Classification method selection
|
Filter decision require whole slice data?
|
BMS
|
2×2 (L), N/A (C)
|
|
# samples in the whole slice (M)
|
8 (L), 4 (C)
|
9×9/7×7/5×5 diamond (L);
5×5 diamond (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.1.a
|
4×4 (L), N/A (C)
|
|
M/2
(one dimension
subsample by 2)
|
8 (L), 4 (C)
|
Same as BMS
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.1.b
|
2×2 (L), N/A (C)
|
|
M/2
(one dimension
subsample by 2)
|
8 (L), 4 (C)
|
Same as BMS
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.2
|
2×2 (L), N/A (C)
|
|
M/4 for highest temporal layer; and M for others
(two dimensions subsampled by 2)
|
8 (L), 4 (C)
|
Same as BMS
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.3
|
2×2 (L), N/A (C)
|
|
M
|
8 (L), 4 (C)
|
Same as BMS
|
N
|
× and <<
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.4.a
|
2×2 (L), N/A (C)
|
|
M
|
6(L), 4(C)
|
7×7 (L), 5×5 (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.4.b
|
2×2 (L), N/A (C)
|
|
M
|
6(L), 6(C)
|
7×7 (L), 7×7 (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.4.c
|
4×4 (L), N/A (C)
|
|
M
|
6(L), 4(C)
|
7×7 (L), 5×5 (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.4.d
|
4×4 (L), N/A (C)
|
|
M
|
6(L), 6(C)
|
7×7 (L), 7×7 (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.4.e
|
2×2 (L), N/A (C)
|
|
M
|
8(L), 4(C)
|
9×9 (L), 5×5 (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.4.f
|
2×2 (L), N/A (C)
|
|
M
|
8(L), 6(C)
|
9×9 (L), 7×7 (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.4.g
|
4×4 (L), N/A (C)
|
|
M
|
8(L), 4(C)
|
9×9 (L), 5×5 (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.4.h
|
4×4 (L), N/A (C)
|
|
M
|
8(L), 6(C)
|
9×9 (L), 7×7 (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.1.4.i
|
|
Same as in BMS
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.2.1
|
2×2, 1×1 (L),
N/A (C)
|
|
M
|
8 (L), 4 (C)
|
Same as BMS
|
Y
|
×
|
16 (L)
1 (C)
|
Slice-level
|
Y
|
2.4.2.2.a
|
2×2, 1×1 (L),
N/A (C)
|
|
M
|
8(L), 8(C)
|
9×9 Cross +3×3Square with half or symmetry
|
N
|
×
|
16 (L)
2 (C)
|
Slice-level
|
Y
|
2.4.2.2.b
|
2×2, 1×1 (L),
N/A (C)
|
|
M
|
8(L), 8(C)
|
9×9 Cross +3×3Square with half or symmetry
|
N
|
×
|
16 (L)
2 (C)
|
Slice-level
|
Y
|
2.4.2.2.c
|
2×2 (L)
N/A (C)
|
|
M
|
8(L), 8(C)
|
9×9 Cross +3×3Square with half or symmetry
|
N
|
×
|
16 (L)
2 (C)
|
Slice-level
|
Y
|
2.4.2.2.d
|
1×1 (L),
N/A (C)
|
|
M
|
8(L), 8(C)
|
9×9 Cross +3×3Square with half or symmetry
|
N
|
×
|
16 (L)
2(C)
|
Slice-level
|
Y
|
2.4.2.2.e
|
N/A (L),
N/A (C)
|
|
0
|
8(L), 8(C)
|
9×9 Cross +3×3Square with half or symmetry
|
N
|
×
|
16L + 2C + (# of CTUs in one CTU row) * 3 (L)
2 (C)
|
N/A
|
Y
|
2.4.2.3
|
2×2 (L),
1×1 (C, inherited from luma)
|
|
M
|
8 (L), 4 (C)
|
9×9/7×7/5×5 diamond (L);
5×5 diamond (C)
|
Y
|
×
|
25 (L)
1 (C)
|
N/A
|
Y
|
2.4.2.4
|
|
Same as BMS (except gradient calculation is replaced by Sobel filter)
|
Results vs. VTM:
|
AI
|
RA
|
LDB
|
Test#
|
Y
|
U
|
V
|
EncT
|
DecT
|
Y
|
U
|
V
|
EncT
|
DecT
|
Y
|
U
|
V
|
EncT
|
DecT
|
AhG13
|
-3.30%
|
-3.55%
|
-4.06%
|
101%
|
155%
|
-5.34%
|
-2.21%
|
-1.66%
|
108%
|
184%
|
-
|
-
|
-
|
-
|
-
|
2.4.1.1.a
|
-3.11%
|
-3.55%
|
-4.06%
|
110%
|
205%
|
-5.18%
|
-2.26%
|
-1.65%
|
113%
|
268%
|
-4.41%
|
-1.80%
|
-1.85%
|
119%
|
216%
|
2.4.1.1.b
|
-3.19%
|
-3.55%
|
-4.06%
|
111%
|
209%
|
-5.25%
|
-2.19%
|
-1.64%
|
114%
|
273%
|
-4.45%
|
-1.81%
|
-1.94%
|
120%
|
219%
|
2.4.1.2
|
-3.30%
|
-3.55%
|
-4.06%
|
102%
|
170%
|
-5.26%
|
-2.23%
|
-1.67%
|
109%
|
200%
|
-4.55%
|
-1.80%
|
-1.95%
|
114%
|
170%
|
2.4.1.3
|
-3.04%
|
-3.44%
|
-3.95%
|
102%
|
149%
|
-5.15%
|
-2.19%
|
-1.56%
|
107%
|
176%
|
-4.52%
|
-1.79%
|
-2.16%
|
112%
|
154%
|
2.4.1.4.a*
|
-3.10%
|
-3.56%
|
-4.07%
|
101%
|
124%
|
-5.10%
|
-2.29%
|
-1.69%
|
105%
|
135%
|
-4.32%
|
-1.79%
|
-1.77%
|
106%
|
119%
|
2.4.1.4.b*
|
-3.09%
|
-4.20%
|
-4.73%
|
101%
|
123%
|
-5.11%
|
-3.16%
|
-2.66%
|
104%
|
134%
|
-4.31%
|
-2.63%
|
-2.72%
|
107%
|
122%
|
2.4.1.4.c*
|
-2.99%
|
-3.56%
|
-4.07%
|
101%
|
114%
|
-4.99%
|
-2.24%
|
-1.66%
|
104%
|
122%
|
-4.19%
|
-1.64%
|
-1.85%
|
107%
|
114%
|
2.4.1.4.d*
|
-2.98%
|
-4.20%
|
-4.73%
|
102%
|
114%
|
-4.98%
|
-3.14%
|
-2.66%
|
104%
|
122%
|
-4.23%
|
-2.44%
|
-2.57%
|
107%
|
116%
|
2.4.1.4.e*
|
-3.28%
|
-3.54%
|
-4.05%
|
103%
|
130%
|
-5.33%
|
-2.21%
|
-1.63%
|
108%
|
143%
|
-4.56%
|
-1.81%
|
-2.01%
|
110%
|
125%
|
2.4.1.4.f*
|
-3.27%
|
-4.19%
|
-4.71%
|
102%
|
129%
|
-5.35%
|
-3.11%
|
-2.57%
|
107%
|
142%
|
-4.63%
|
-2.61%
|
-2.88%
|
111%
|
128%
|
2.4.1.4.g*
|
-3.17%
|
-3.54%
|
-4.05%
|
102%
|
119%
|
-5.21%
|
-2.20%
|
-1.66%
|
107%
|
129%
|
-4.46%
|
-1.83%
|
-2.01%
|
110%
|
117%
|
2.4.1.4.h*
|
-3.16%
|
-4.19%
|
-4.71%
|
103%
|
121%
|
-5.23%
|
-3.15%
|
-2.63%
|
108%
|
129%
|
-4.49%
|
-2.59%
|
-2.72%
|
111%
|
116%
|
2.4.1.4.i
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
2.4.2.1.a
|
-3.32%
|
-3.55%
|
-4.06%
|
103%
|
162%
|
-5.61%
|
-2.22%
|
-1.66%
|
111%
|
197%
|
-5.19%
|
-1.74%
|
-2.04%
|
117%
|
167%
|
2.4.2.1.b
|
-3.31%
|
-3.55%
|
-4.06%
|
102%
|
165%
|
-5.53%
|
-2.21%
|
-1.67%
|
109%
|
200%
|
-4.97%
|
-1.96%
|
-2.13%
|
115%
|
167%
|
2.4.2.1.c
|
-3.31%
|
-3.55%
|
-4.06%
|
102%
|
164%
|
-5.60%
|
-2.21%
|
-1.64%
|
109%
|
197%
|
-5.14%
|
-1.99%
|
-1.94%
|
113%
|
165%
|
2.4.2.1.d
|
-3.30%
|
-3.55%
|
-4.06%
|
102%
|
166%
|
-5.53%
|
-2.24%
|
-1.66%
|
110%
|
201%
|
-4.90%
|
-1.67%
|
-2.04%
|
115%
|
169%
|
2.4.2.2.a
|
-2.70%
|
-4.94%
|
-5.69%
|
102%
|
133%
|
-4.85%
|
-5.25%
|
-5.08%
|
101%
|
145%
|
-4.60%
|
-7.42%
|
-7.19%
|
102%
|
137%
|
2.4.2.2.b
|
-2.69%
|
-4.94%
|
-5.69%
|
101%
|
132%
|
-4.82%
|
-5.24%
|
-5.09%
|
101%
|
144%
|
-4.52%
|
-7.51%
|
-6.97%
|
101%
|
135%
|
2.4.2.2.c
|
-2.66%
|
-4.94%
|
-5.69%
|
101%
|
134%
|
-4.66%
|
-5.26%
|
-5.10%
|
100%
|
148%
|
-4.19%
|
-7.48%
|
-6.97%
|
101%
|
136%
|
2.4.2.2.d
|
-2.07%
|
-4.96%
|
-5.71%
|
101%
|
117%
|
-3.72%
|
-5.12%
|
-4.95%
|
100%
|
123%
|
-3.38%
|
-7.30%
|
-6.96%
|
101%
|
118%
|
2.4.2.2.e
|
-1.89%
|
-4.97%
|
-5.71%
|
101%
|
114%
|
-3.41%
|
-5.16%
|
-4.96%
|
100%
|
118%
|
-2.94%
|
-7.33%
|
-6.95%
|
101%
|
113%
|
2.4.2.3
|
-3.24%
|
-4.79%
|
-5.47%
|
105%
|
144%
|
-5.31%
|
-6.69%
|
-5.38%
|
111%
|
173%
|
-4.62%
|
-8.68%
|
-7.52%
|
116%
|
154%
|
2.4.2.4**
|
-3.25%
|
-3.51%
|
-4.02%
|
109%
|
268%
|
-5.34%
|
-2.21%
|
-1.63%
|
110%**
|
370%
|
-4.68%
|
-1.55%
|
-2.04%
|
124%
|
275%
|
Dostları ilə paylaş: |
|
|
