Back to Search
Document details
Non-EE2: CCCM using multiple downsampling filters
Abstract
In this contribution, it is proposed to apply multiple downsampling filters to reconstructed luma samples corresponding to the prediction shape of a convolutional cross-component model for chroma intra prediction. The linear combination of these variously downsampled samples separately multiplied by derived coefficients composes a chroma predictor. The experimental results over ECM-7.0 are summarized as follows:
JVET-AC0148 Non-EE2: CCCM using multiple downsampling filters [Y.-J. Chang, V. Seregin, M. Karczewicz (Qualcomm)]
In this contribution, it is proposed to apply multiple downsampling filters to reconstructed luma samples corresponding to the prediction shape of a convolutional cross-component model for chroma intra prediction. The linear combination of these variously downsampled samples separately multiplied by derived coefficients composes a chroma predictor. The experimental results over ECM-7.0 are summarized as follows:
- AI: -0.02 % (Y), -1.05 % (U), -0.89 % (V), 104 % (EncT), 100 % (DecT)
- AI-F: -0.19 % (Y), -1.07 % (U), -0.98 % (V), 103 % (EncT), 101 % (DecT)
- AI-TGM: -0.75 % (Y), -1.14 % (U), -1.37 % (V), 101 % (EncT), 100 % (DecT)
- RA: 0.01 % (Y), -0.69 % (U), -0.57 % (V)
- RA-F: -0.28 % (Y), -1.15 % (U), -0.95 % (V)
- RA-TGM: -0.2 % (Y), -0.44 % (U), -0.39 % (V)
It was expected (also by the cross-checker) that the gains would still be retained in context with new adoptions to ECM. It was suggested to investigate this in an EE. Reduction of encoder run time was requested to improve the tradeoff with the compression gain.
Cited:
Decisions
adopted
It was expected (also by the cross-checker) that the gains would still be retained in context with new adoptions to ECM. It was suggested to investigate this in an EE. Reduction of encoder run time was requested to improve the tradeoff with the compression gain.
Citation