Teegedeck
7th March 2002, 01:21
Hi,
I got a daft idea for making curve-compression possibly a bit more intuitive. Just say 'daft' if you think it is lame or won't work. :)
Instead of having to enter curve-compression-values, the values could be calculated based upon a user-defined value
'max. big frame quant=max. small frame quant+(user-defined value)'
IF this is mathematically possible! You would have to decide what are considered smallest (average size of the smallest 10%?) or largest frames. If it worked, we could drop CC-percentages and maybe any absolute quant-limitations.
The idea stems from several thoughts. If experience from Nandub counts anything, then we aren't likely to find ideal values for CC EVER. At the moment I find myself searching for ideal values for every individual movie, just like in the dreaded old days. If we consequently aren't gonna hardcode the values any time soon, people are likely to screw up the settings they don't understand and blame the codec. One way to find pseudo-universally good CC-values was using a crosspoint in Nandub, introduced by R.H. But the merits of this value were questionable, too, they were rooted in theory, not very close to the actual problem. So me thinks, if the whole problem - and the thing I'm trying to sort out in every encoding - is to find a CC that triggers quants that are as low as possible on big frames while still keeping the quants on small frames low enough to make them look decent (I remembered that setting CC to '0' like I did in Nandub only worked because I could specify a minimum bitrate), the one thing that doesn't rely on the bitrate here, is my choice on how much these quants are allowed to differ. If I'd choose 'max. big frame quant=max. small frame quant+(0)', I'd have perfect VBR, all other positive values would allow me to treat small frames more benignly in a way that I would still call intuitive. And it would be easier to find values good enough for a possible default. And maybe all this is nonsense and wouldn't work at all. ;) Just let me know what you think.
Please forgive me if the post is a bit incoherent; it's late.
I got a daft idea for making curve-compression possibly a bit more intuitive. Just say 'daft' if you think it is lame or won't work. :)
Instead of having to enter curve-compression-values, the values could be calculated based upon a user-defined value
'max. big frame quant=max. small frame quant+(user-defined value)'
IF this is mathematically possible! You would have to decide what are considered smallest (average size of the smallest 10%?) or largest frames. If it worked, we could drop CC-percentages and maybe any absolute quant-limitations.
The idea stems from several thoughts. If experience from Nandub counts anything, then we aren't likely to find ideal values for CC EVER. At the moment I find myself searching for ideal values for every individual movie, just like in the dreaded old days. If we consequently aren't gonna hardcode the values any time soon, people are likely to screw up the settings they don't understand and blame the codec. One way to find pseudo-universally good CC-values was using a crosspoint in Nandub, introduced by R.H. But the merits of this value were questionable, too, they were rooted in theory, not very close to the actual problem. So me thinks, if the whole problem - and the thing I'm trying to sort out in every encoding - is to find a CC that triggers quants that are as low as possible on big frames while still keeping the quants on small frames low enough to make them look decent (I remembered that setting CC to '0' like I did in Nandub only worked because I could specify a minimum bitrate), the one thing that doesn't rely on the bitrate here, is my choice on how much these quants are allowed to differ. If I'd choose 'max. big frame quant=max. small frame quant+(0)', I'd have perfect VBR, all other positive values would allow me to treat small frames more benignly in a way that I would still call intuitive. And it would be easier to find values good enough for a possible default. And maybe all this is nonsense and wouldn't work at all. ;) Just let me know what you think.
Please forgive me if the post is a bit incoherent; it's late.