Abstract not available in document
JVET-D0052 AHG5: Experiment on JEM default setting [K. Choi, E. Alshina, C. Kim (Samsung)]
In this contribution, test results on various configuration settings are provided. The Test6 setting (i.e., CTU=256, MaxBTDepth=2) in this contribution showed the fastest encoding time that reduced encoding time about 37% with 1.2% luma coding loss. In the Test4 setting (i.e., CTU=256), coding improvement was achieved about 0.4% in RA without increase of encoding time. Considering the trade-off between encoding time and coding loss, the contributor suggested to use Test4 or Test6 setting for the next default setting of JEM software.
(Discussion was chaired by Yu-Wen Huang)
Param. | Anchor | Test1 | Test2 | Test3 | Test4 | Test5 | Test6 |
CTUSize | 128 | 128 | 128 | 256 | 256 | 128 | 256 |
MinQTLumaISlice | 8 | 16 | 8 | 8 | 8 | 8 | 8 |
MinQTChromaISlice | 4 | 4 | 4 | 4 | 4 | 4 | 4 |
MinQTNonISlice | 8 | 16 | 8 | 8 | 8 | 8 | 8 |
MaxBTDepthISliceL | 3 | 4 | 2 | 2 | 3 | 2 | 2 |
MaxBTDepthISliceC | 3 | 0 | 2 | 2 | 3 | 0 | 0 |
MaxBTDepth | 3 | 4 | 2 | 2 | 3 | 2 | 2 |
MaxBTSizeISliceL | 32 | 32 | 32 | 32 | 32 | 32 | 32 |
MaxBTSizeISliceC | 16 | 16 | 16 | 16 | 16 | 16 | 16 |
MaxBTSizeBPSlice | 128 | 128 | 128 | 128 | 128 | 128 | 128 |
RA, Y / U/ V BD-rate
Detail | Coding performance | Encoding Time | Decoding Time |
Test1 | −0.77% / 3.04% / 3.27% | 133% | 100% |
Test2 | 1.46% / 1.93% / 1.97% | 66% | 107% |
Test3 | 1.28% / 0.89% / 1.13% | 66% | 106% |
Test4 | −0.43% / −1.07% / −0.95% | 105% | 108% |
Test5 | 1.37% / 5.90% / 6.08% | 63% | 100% |
Test6 | 1.23% / 4.82% / 5.21% | 63% | 100% |
The proponents preferred Test 4 or Test 6.
It was asked why Test 4 did not change Class D results. This is because enlarging CTU to 256 has no effect for Class D in JEM-3.0.
It was asked why Test 4 Class D has higher decoding time. The proponent was not sure about the root cause.
One expert commented Test 1 and Test 4 are out of the BoG scope. One expert commented that Test 4 shows a very good tradeoff.