Of itu-t sg16 wp3 and iso/iec jtc1/SC29/WG11


Tool experiments (in SHVC) 5.1TE1: Upsampling filter



Yüklə 2,33 Mb.
səhifə9/37
tarix17.01.2019
ölçüsü2,33 Mb.
#99028
1   ...   5   6   7   8   9   10   11   12   ...   37

5Tool experiments (in SHVC)

5.1TE1: Upsampling filter

5.1.1TE1 summary and general discussion


JCTVC-L0021 TEA1: Summary Report Upsampling Filter [A. Segall, J. Chen, J. Dong, E. Alshina]



Test

Proponent

Avg.

BD-rate

x2

Avg.

BD-rate

x1.5

Enc.Time

Dec.Time

#

 

Y

U

V

Y

U

V

 

 

test 4.1 (L0083)

Samsung

−17.1

−12.4

−11.9

−26.9

−23.4

−22.3

111

126

test 4.2_(L0054) 6bit luma + K0378 chroma

InterDigital

−17.1

−12.4

−11.9

−26.9

−23.5

−22.4

107

125





























































test 4.3 (L0335)

Qualcomm

−17.1

−12.4

−11.9

−26.9

−23.4

−22.3

111

128

test 4.4 (L0083)

Samsung

−17.1

−12.4

−11.9

−26.9

−23.4

−22.3

111

124

Performance-wise, there was basically no difference.

Current “default” (4.1) as used in TE2+ was a design with 6 bit precision for luma and chroma (which is identical with the MC interpolation filter of HEVC in case of 2X, where chroma is using 5 bits)

For 2X scalability luma, there was no difference between the proposals. For chroma, 4.1-4.3 are identical, and 4.4 uses a different phase calculation for vertical upsampling (but still using one of the MC chroma interpolation filters)

For 1.5X scalability (which requires 1/3 phase position for luma and some more for vertical chroma which are not existing as MC interpolation filters), 4.1, 4.3 and 4.4 use the same luma filter, 4.2 uses a different one.

From the frequency response presented in the cross-check document JCTVC-L0088, the design of 4.1/4.3/4.4 seems to be closer to the MC interpolation filters (avoiding the “bumps” in the passband).

Decision: Adopt solution from 4.1/4.3/4.4 as default luma filter.

For 1.5X chroma where 1/3, 7/12 and 11/12 phase positions are additionally required, 4.3 uses 5 bit rounding, whereas 4.1/4.2/4.4 use 6 bit (and 4.4 uses different phase calc. with less accuracy).

5 bit rounding is more consistent with the MC interpolation filter design.

Decision: Adopt solution from 4.3 as default chroma upsampling filter.

Further study (AHG12) was planned on accuracy of chroma phase shift for possible simplification.

5.1.2TE1 primary contributions


JCTVC-L0054 TE1: Results of test 4.2 on fixed upsampling filter [J. Dong, Y. Ye (InterDigital)]
JCTVC-L0083 TEA1: Fixed up-sampling filter tests by Samsung (tests 4.1 and 4.4) [E.Alshina, A.Alshin, J.H.Park (Samsung)]
JCTVC-L0335 TE-A1: Results of test 4.3 on fixed up-sampling filter [J. Chen, R. Krishna, M. Karczewicz (Qualcomm)]

5.1.3TE1 cross checks


JCTVC-L0088 TEA1: Cross-check of fixed 6 and 7 bit-depth up-sampling filters performance (test 4.2) [E.Alshina, A.Alshin, J.H.Park (Samsung)]
JCTVC-L0089 TEA1: Cross-check of low bit-depth Chroma up-sampling filter (test 4.3) [E.Alshina, A.Alshin, J.H.Park (Samsung)]
JCTVC-L0396 TEA1: Cross check for Test 4.4 on chroma up-sampling filter (JCTVC-L0083) [M. Budagavi (TI)] [late]


Yüklə 2,33 Mb.

Dostları ilə paylaş:
1   ...   5   6   7   8   9   10   11   12   ...   37




Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©muhaz.org 2024
rəhbərliyinə müraciət

gir | qeydiyyatdan keç
    Ana səhifə


yükləyin