Quote:
Originally Posted by Stereodude
I don't know that I need lossless, but I'd like to avoid unnecessary quality losses. Uncompressed UHD 10-bit files are very large hence my interest to avoid them.
I tried an x264 encode with a CRF of 1 and that opened in Resolve. I wasn't sure how the quality of that would compare to Prores or DNxHR.
|
--qp 1 will give you larger filesizes /less loss than --crf 1
Here is some sample data from a native UHD 10bit422 source (it's very clean, more easily compressible, you can tell from the results because ProresHQ typically scores ~ 45-55db on slightly noisy sources )
ProresHQ (ffmpeg)
591Mb/s
psnr_avg:68.73 psnr_y:67.46 psnr_u:69.74 psnr_v:71.49
x264cli --crf1 --keyint 1
356Mb/s
psnr_avg:73.20 psnr_y:70.83 psnr_u:78.08 psnr_v:79.68
x264cli --qp1 --keyint 1
512Mb/s
psnr_avg:82.08 psnr_y:79.72 psnr_u:86.44 psnr_v:89.33
x264 Intra will give you higher quality at a given bitrate, at the expense of higher seek latency compared to Prores or DNxHR or Cineform (--qp 1 will yield higher quality than even at cineform filmscan2)
But with x264 you have the option to use short GOP's to significantly reduce the filesize, while still keeping higher quality than Prores , DNxHR, Cineform .
Or you can tweak it to perform faster using fast decode options at the expense of quality, or tune it in other ways, etc..
But the resolve export side is quite limited - you don' t have access to x264 or various options. You're right - uncompressed 10bit UHD is no fun, but that's almost your only option for getting lossless material back out
A fairly common workflow people use is AAF/EDL/XML import into resolve and export that back into their editor. It's basically a type of frameserving. But avisynth/vapoursynth do not support it
Maybe one day...
Another lossless import/export option that people use with resolve is EXR . It's usually used for higher end workflows. Virtually all formats (YUV, subsampling, any bitdepth, RGB) can be converted back/forth losslessly with EXR float . But you need vapoursynth because it can import/export float formats and do lossless colorspace transforms properly (avs+ currently cannot import/export float) . 16bit half float is all that is required for most sources or consumer formats. If using EXR piz compression, they are only slightly larger than typical 16bit PNG, but some types of sources are actually smaller than if using 16bit PNG (content dependent). But that's still going to be much larger than a --qp 1 encode, but smaller than uncompressed UHD 10bit422 or 420. And some people don't like using image sequences