x264 Presets and archiving

[Virnec]

I'm trying to find the best settings for archiving my fraps footage for later use.

I've been comparing results from various settings, and found that a crf 10 and the "slower"-preset seems to do the job. It's almost not possible to separate the original from the new encode.
Yet, the new clips never exceed a size of 9% of the original -- which is very impressing. So impressing, in fact, that I'm worried that too much of the information is removed.
I'm not an expert in video compression, but I was adviced, in an earlier thread, not to use any b-frames at all - which lead to much larger files.
I know for a fact that b-frames can drastically reduced the filesize, hence the slower encoding time and smaller files, but is it necessery to cut down on them, to make perfectly good archive-grade footage? And is there any other factors I should investigate?

Thanks for advices!

[poisondeathray]

"archive grade" implies original quality. So I would keep original format.

If it's important, I would keep the original for archive purposes, not a lossy compressed format. HDD are cheap these days.

Your fraps is RGB 4:4:4, x264 uses YV12 4:2:0, so you are throwing away a lot of color information, even if you use lossless x264 mode.

The long GOP helps compression, but is more difficult to edit. Fraps is all intra (all I-frame) , much easier to edit.

It's up to you to decide how much quality loss you are willing to accept, and the trade offs you are willing to make between size/quality. But it's certainly not "archive grade" in the strict sense.

[Virnec]

There's definitely a difference between, say work/hobby footage and family/important footage - which I would store raw anyway.
My method of comparing the clips may be inadequate, for all I know, but I use ffmpeg to capture raw frames from the orignial and encoded clip. I then simply take each set and switch back and forth between them and survey for errors. I suppose PSNR and other methods of comparing does roughly the same thing, but if you can't spot any difference without really, really looking, then it's a win, right?

Regarding compitability with editors and such -- isn't it possible do decode the video back to a raw format and work with that?


I've been looking at the screencaps and comparing the clips, thinking that if I can't separate them quality-wise, it's a win. On the other hand, I'm one of those who prefer flac over mp3, except on portable players of course. Because something does indeed happen to the music when you compress it down to 10% of it's original size.


I've read through all the options at this neat site here, and this is what I've come up with thus far:

--bframes 0 \
--qp 10 \
--min-keyint 30 \
--keyint 150 \
--qpmax 30 \
--me umh \
--merange 32 \
--subme 9 \
--trellis 2 \
--ref 8 \
--partitions all \
--rc-lookahead 60 \

Still, the files doesn't grow larger. It seems like I'll be investing abit more time in this than I planned to do ^^.
I'll jump over to doom9, for more techincal surveying.
Thanks!

edit: oh, and I found an interesting article here about GOPs. This will perhaps answer some of my questions.
Thanks again

[creamyhorror]

--qp is mutually exclusive with --crf. CRF allocates bitrate by a mechanism aimed at human perception (by allocating less bits to high-motion frames), while --qp (constant quantizer) allocates bitrate strictly at true constant quality. You should only use one of them.

--crf 10 or --qp 10 are unnecessarily low, and you'll be served fine by 15-16. At 16, it takes a lot of pausing, zooming, and flipping back and forth in order to identify significant differences (there are always changes in background noise). In motion, you won't be able to tell the difference.

The 9% ratio you obtained should not be considered that impressive, since FRAPS is highly uncompressed and contains a lot more color data (as poisondeathray said).

Setting b-frames to 0 means that the video requires a significantly higher bitrate, but will deliver correspondingly better quality in every single frame (although efficiency is compromised somewhat because b-frames are bi-predictive). It's a good option for preserving footage at near-archival, visually-lossless quality for future editing and high-quality rendering. For preserving footage for playback only, b-frames are probably good to use.

[Virnec]

--qp is mutually exclusive with --crf

I've been using the qp setting earlier, but I mixed it with the crf, for some reason..
I remember using qp, because I wanted all of the video to be in thoroughly good condition and re-usable.

--crf 10 or --qp 10 are unnecessarily low, and you'll be served fine by 15-16. At 16

That may indeed be sufficient, and I suppose it all comes to how much storage you can spare.
The reults I've recieved thus far, suggests that changing from 10 to 16 qp actually yields around 50% smaller files.

The 9% ratio you obtained should not be considered that impressive,

I'm honestly not sure what to expect, but I'm rather impressed by seeing the dents on my vallet getting flattened out, by not having to store it in a raw format

Setting b-frames to 0 means that the video requires a significantly higher bitrate, but will deliver correspondingly better quality in every single frame (although efficiency is compromised somewhat because b-frames are bi-predictive).

Indeed, and you'd only want to do this if you still want to work on the footage. When encoding clips for watching only, it makes sense to pack these as small as possible, where AVC seems to do a beautiful job.


Do you reckon there is any other settings you'd add, or tune differently -- and why?

Thanks for an enlightening response

[jagabo]

Use shorter GOPs to make the files more easily editable. File size doesn't start blooming until GOPS get down to the low double digits. Although this varies depending on the nature of the video.