Quote:
Originally Posted by poisondeathray
On modern computers, native HDV editing isn't very slow ; surely "original" sampling is better than any other manipulation ? Moreover, many NLE's can "smart render" HDV if you're performing cuts type editing - so why not edit HDV natively ?
|
I take your point entirely; native HDV editing is fine if one is happy with results of the hybrid (original/re-rendered) MPEG2 output, in terms of desired/uniform quality, color characteristics and playability. Native HDV editing 'Smart frame-accurate rendering’ has come on a long way, at least in the higher-end editing suites.
But me not, at least with the editing software I use (Corel Video Studio). I prefer to edit in an intra-frame format that is intrinsically ‘frame accurate’, ‘smart render-able’ and one which then can then serve as a master and intermediate format for AVISynth processing and archiving, preferably with some economy in file size. So, either lossless or as ‘losslessly-lossy’ as is possible, preferably in YV12 (for preferred AVISynth filters/functions), or in a YUV format that be efficiently interchanged with YV12 over several generations, at least. Oh yes….and ideally one that has some pedigree, developmental-stability and well-defined configurability (that is a word)….so not one ‘in-development/experimental’ that could leave me in several years searching for an older version so that I can access my video files.
At present HuffYuv-YV12 (FFDShow variant) best fits that bill, for me; I know there are other well established lossless variants, several of which I rigorously tested a few years back, but each to his own. Rather, I was looking to see if there are viable alternatives that might offer more economy in file size.
Hence, the consideration of DVCProHD on one hand, Cineform on the other, and just maybe x264 (I-frame only) some where in between.
I had used the Matrox DV codec quite extensively before Cedocida came along. Had, maybe still has, a Top Field First interlace flag, contrary to accepted DV convention. But I don’t recall there being an issue with luminance, as I have now observed with the DVCProHD and Mpeg-2 I-Frame codecs.
I'm not sure if by 'clipping' you are referring to what I would term 'clamping' i.e. truncation of Y values below 16 and above 235, as opposed to back-scaling? If, so that's not what I observed. From the histograms (AVISynth Histogram - mode "levels" - after reconversion to YV12) the codecs seemed to behave as prescribed with respect to optional 'PC' 0 - 255 and 'TV' 0 - 235 scales. What I observed was a flattening of theY (luminance) plot. I'm sure some RGB sampling nuance is involved, but I’m not particularly inclined to get the bottom of it or reinstall the codec pack as their presence makes accessing the VDub codecs library very slow.
So, without a DVCProHD codec, I resorted to some comparative testing of DVCPro50 and Cineform in SD, with the assumption that this might reflect a similar behavior at the HD level. Which is not unreasonable, as DVCProHD (100) was developed as an extension of DVCPro50 by doubling the number of DV codecs, working in parallel, from two to four. There was DVCPro Progressive (in 4:2:0) at one point which, like HDV-SD, was introduced as an intermediate format (for sports, news gathering etc) during the transition from SD to HD.
That said, whilst the comparative SSIM scores I mentioned yesterday were encouraging, I thought my query about the efficiency of the involved YV12/YUY2 conversions might be better answered by some ‘in-use’ generation tests i.e. successive encoding, with each cycling through YV12 (as would be used for AVISynth processing and final target output), to see how well the quality was preserved.
For Cedocida DVCPro50, I merely set the encoder input and decoder output options to ‘YV12 Progressive’ sampling, so relying on the codecs internal transformations to and from YUY2. For Cineform I let the encoder accept YV12 input and then returned the decode YUY2 output with ConvertToYV12 (interlaced=false). What I found rather disappointed me. I’ve yet to fill in all the dots, but by 4 generations the quality of the DVCPro50 encodes (as judged by the SSIM scores) started to deteriorate, dropping from 94.84 to 93.89, and by the 10th generation it was down to 91.80, approaching the quality of (one generation) DV, with a certain loss of definition and more mosquito-noise.
Cineform held up rather better and by the 5th cycle had only dropped from 94.85 to 94.25. Excluding the ConvertToYV12 step (i.e. letting the codec cycle in YUY2) it was a little better – 94.47 after 5 cycles, which is probably not even statistically significant. I’ve yet to complete the series, although up to 5 cycles would not be an unrealistic reflection of use as an edit/processing intermediate, unless you are a "single pipe-liner". Some people have reported edge 'pixelization' artifacts with Cineform, and others blockiness, possibly related to use of non-standard resolutions. But frankly the encodes looked very clean to me, for SD. I'll test some more at HD.
I did do a few SD-level tests with x264. An ‘Intra’ (keyinterval =1) profile (Medium preset, FastDecode, Auto-level, no Tune preset) at crf 7.0 was slightly better (SSIM 95.06) than (the first generation) DVCPro50 and Cineform encodes. At crf 2.0 it was up to 97.97 with a file size (342MB) similar to that of the Cineform encode. Dropping right down to crf 0 (I know ‘lossless’ mode is technically in 4:4:4 format) however file size was up (530MB) on par with Huff-YV12 and YUY2. I couldn’t run the SSIM as DGAVCINdex threw an error. So I don’t think there’s anything to be gained in using I-Frame only x264 as a ‘lossless’ alternative to HuffYuv. Clearly it has potential use as a ‘less-lossy’ intermediate, albeit less convenient unless I use the vfw version (forgive me for even thinking that). Maybe an option for more economic archiving though.....at least progressive video.
Needless to say, I've modified my opinion somewhat on the usefulness of DVCPro50 (and, by extension, DVCProHD) as an 'intermediate' format.It was after all developed as a 'capture' format.
That’s enough for one day.