Quote
Jude Cotter
Don't go to H.264. Banding is common in lower res formats, especially in gradients.
H.264 isn't a low res codec. H.264, like the ProRes and the uncompressed codecs, can be used for a range of resolutions from sub-SD to supra-HD. Nor is H.264 an unsharp codec when it is used with a sufficient data rate for the resolution. H.264 is low bit-depth codec: 8-bits rather than 10-bits, making it more prone to banding. But whose monitor is 10-bits? Also H.264 is an interframe compressed codec, so one might suspect that it could mishandle fades. This needs to be checked before slamming H.264 as the source of Mike M.'s problem.
What exactly is Mike M.'s problem?
banding/artifacting/CRAP! Banding is when a smooth
spatial gradient shows steps. Distinct tones appear where there should be none. Banding is a phenomenon within a single frame. But in a fade the bands move, making them more noticeable.
8-bit coding alone can't produce serious banding, but image processing of 8-bit images can exaggerate the steps. For example the original 8-bit image might have a step from code 200 to code 201 that would be barely visible. Suppose an effect changes the two bands to codes ~200½ and ~201½. Suppose the first code rounds down and the second code rounds up. Now the step goes from code 200 to code 202 and is definitely visible. The "rounding errors" have reduced the 8-bit coding to 7-bit coding. And then the next effect is applied, etc. This is why it is good to work in a 10-bit codec even when release will be in 8-bits. Mike M. seems to have only outputted in H.264.
It should be studied whether H.264 bands more than an 8-bit codec should. H.264 might skimp at low data rates, but at high data rates does it look any worse than 8-bit uncompressed?
I verified that FCP7 fades are 10-bit by making a 956 frame long cross dissolve between superwhite code 1019 and black 64. The FCP7 dissolve is perfectly linear in code value. Each value between 64 and 1019 appears in the dissolve once. I've my own grief with FCP7 fades. They often appear
temporally unsmooth. This experiment seems to absolve FCP7. Can my playback be that bad?
Dennis Couzin
Berlin, Germany