Pier
14th October 2013, 03:13
Hi, I was confused about some points of the H264 coding and I was wondering if someone could clarify them for me. I may also think that something is clear to me, while it is not, as a result there is the risk that i will make mindless questions, if I do, please point it out.
-A 16x16 macroblock gives the best compression (entropy wise) and the worst spatial/temporal prediction, while a 4x4 sub-block would give the best spatial/temporal prediction and the worst compression (entropy wise).
--Here the point: How is dicided if a macroblock must be divided into sub-blocks?
Here what i could understand: A s/t prediction is made on a macroblock, after that another s/t prediction is made on the 16x8,8x16,8x8 sub-blocks; if the extra precision from the sub-prediction is not much the macroblock is kept, otherwise the block is divided into 2 8x16/16x8 sub-blocks, or 4 8x8 sub-blocks and in this case, for each of the 8x8 blocks the above check is repeted with the 8x4,4x8,4x4 sub-sub-blocks, keeping the best compromise between compression and prediction. Am i correct? Also, an inter macroblock/sub-block can make both spatial and temporal prediction, right?
-Into an H264 encoder (x264 as example) if i use --ref 5 it would mean that i'm using max 5 reference frames for a P-Slice, and 5 for each dimension into a B-Slice, right?
-When a block is compensated using the s/t prediction the DC value is calculated on the original block or on the remaining after the prediction?
-I never seen a 4x8,8x4 QM, are they made from the 8x8 QM?
-A 16x16 QM dosn't exist becouse the HCT and the WHT is made using the 4x4 QM, right? (This one is really confusing me :p)
-If the deblocking filter is used, the s/t predition is made on the filtered blocks right?
-If i'm a person that does his encodes using negative values for the deblocking filter or even without using it at all, would be a good idea to use an high value like 1:1 for a better prediction while disabling it during playback, so that i can keep my details/original quality with a less filesize due the better prediction?
-Into an H264 encoder, the qpstep means:
1-How much the average QP between the current and the next frame can change.
2-How much the QP of the current block and the next block in the same x,y position can change.
3-How much the min/max QP of the frame can change.
I'm stupid*
-A 16x16 macroblock gives the best compression (entropy wise) and the worst spatial/temporal prediction, while a 4x4 sub-block would give the best spatial/temporal prediction and the worst compression (entropy wise).
--Here the point: How is dicided if a macroblock must be divided into sub-blocks?
Here what i could understand: A s/t prediction is made on a macroblock, after that another s/t prediction is made on the 16x8,8x16,8x8 sub-blocks; if the extra precision from the sub-prediction is not much the macroblock is kept, otherwise the block is divided into 2 8x16/16x8 sub-blocks, or 4 8x8 sub-blocks and in this case, for each of the 8x8 blocks the above check is repeted with the 8x4,4x8,4x4 sub-sub-blocks, keeping the best compromise between compression and prediction. Am i correct? Also, an inter macroblock/sub-block can make both spatial and temporal prediction, right?
-Into an H264 encoder (x264 as example) if i use --ref 5 it would mean that i'm using max 5 reference frames for a P-Slice, and 5 for each dimension into a B-Slice, right?
-When a block is compensated using the s/t prediction the DC value is calculated on the original block or on the remaining after the prediction?
-I never seen a 4x8,8x4 QM, are they made from the 8x8 QM?
-A 16x16 QM dosn't exist becouse the HCT and the WHT is made using the 4x4 QM, right? (This one is really confusing me :p)
-If the deblocking filter is used, the s/t predition is made on the filtered blocks right?
-If i'm a person that does his encodes using negative values for the deblocking filter or even without using it at all, would be a good idea to use an high value like 1:1 for a better prediction while disabling it during playback, so that i can keep my details/original quality with a less filesize due the better prediction?
-Into an H264 encoder, the qpstep means:
1-How much the average QP between the current and the next frame can change.
2-How much the QP of the current block and the next block in the same x,y position can change.
3-How much the min/max QP of the frame can change.
I'm stupid*