rudyb
22nd August 2014, 09:04
Hi everyone,
I would really appreciate if someone verifies my understanding of PU partitioning.
I believe this is also explained in section 8.4.1 (sub-section 2) of the following excellent source:
http://f265.org/f265/static/txt/h265_companion.html
We know that for Intra-Prediction, we always have partition size of Part_2Nx2N (well except smallest CU size). Even though this implies that our PU size will be the same as CU size, but all this REALLY implies is that we use this PU size to determine the correct mode (angle); but in reality this PU size will possibly split (depending on the TU size), although it will use the same mode that was determined at Part_2Nx2N level for all the sub-PUs.
For example, if we have a CU size of 32x32, then we will use a PU size of 32x32 to find the best angle (mode). Let's say the best mode turns out to be angle 26. And let's say that we want to use TU size of 16x16. Then this means that, in reality, we use four PU size of 16x16, but with the best mode that was determined at a PU size of 32x32 (e.g. angle 26).
So, we will use a PU size of 16x16 for each TU size of 16x16, but we don't compute the best mode again, and will use the best mode (angle 26) that was already determined at PU size of 32x32, for each one of 16x16 sub-PUs.
Is this correct?
Many thanks in advance.
I would really appreciate if someone verifies my understanding of PU partitioning.
I believe this is also explained in section 8.4.1 (sub-section 2) of the following excellent source:
http://f265.org/f265/static/txt/h265_companion.html
We know that for Intra-Prediction, we always have partition size of Part_2Nx2N (well except smallest CU size). Even though this implies that our PU size will be the same as CU size, but all this REALLY implies is that we use this PU size to determine the correct mode (angle); but in reality this PU size will possibly split (depending on the TU size), although it will use the same mode that was determined at Part_2Nx2N level for all the sub-PUs.
For example, if we have a CU size of 32x32, then we will use a PU size of 32x32 to find the best angle (mode). Let's say the best mode turns out to be angle 26. And let's say that we want to use TU size of 16x16. Then this means that, in reality, we use four PU size of 16x16, but with the best mode that was determined at a PU size of 32x32 (e.g. angle 26).
So, we will use a PU size of 16x16 for each TU size of 16x16, but we don't compute the best mode again, and will use the best mode (angle 26) that was already determined at PU size of 32x32, for each one of 16x16 sub-PUs.
Is this correct?
Many thanks in advance.