Can compression affect conversion speed?

[Guernsey]

I usually like to use the Slower compression for my x264 mp4 videos. I always wondered if it could affect how fast video converters like HandBrake convert and compress video? Or am I wrong in that assumption? If not, what does affect conversion speed? I mostly used Virtualdub2 to convert videos to mp4 lossless after upscaling them.

[johns0]

https://forum.videohelp.com/threads/404643-does-the-conversion-speed-really-affect-the-quality

[rrats]

yes.

[hdfills]

slower compression , plus more compression will always slow down the process

[Guernsey]

So it does? That explains why it takes Virtualdub2 a long time to convert the videos. I also learned in that forum thread that CRF also plays a role as well. I usually use lossless but it looks like I will have to stick to making CRF 16.

[jagabo]

Of course slower compression takes longer. There's a 100 fold difference in encoding time speed between x264's ultrafast (it's fastest) and placebo (it's slowest) presets. Slower compression delivers better image quality.

[Alwyn]

Hmm, what's the relationship between "Placebo" on the Preset menu and the CRF? "Placebo" indicates no "change from original" but obviously, with any CRF greater than 1, I assume you won't get a Placebo copy. or is the term "placebo" a misnomer, Very very slow being a more accurate term?

[jagabo]

I think the name placebo is meant more as a joke than a real description. It takes much longer than veryslow with very very little real benefit. You'll think it's better than veryslow but it's more of a placebo effect.

[Alwyn]

Thanks Jagabo, I wonder how much the quality is actually improved going from say Fast to Slow?

[Sharc]

I wonder how much the quality is actually improved going from say Fast to Slow?

How would we quantify "how much"? One would have to agree on the method and the metric for the quality assessment.
Going from Fast to Slow will affect the file size as well assuming that you keep the CRF value fixed. For a fair comparison ('apples with apples') one would have to compare the quality at equal file sizes. Means one will have to adjust the CRF between tests accordingly (by trial and error, good luck and much patience) to achive very similar file sizes, or do a 2-pass encode with a fixed file size.
It also depends on the video (anime, natural, noise and grain, details, action or static scenes ....).
The 'quality' of encodes can be 'measured' with PSNR or VMAF metrics for example, or better use your critical and trained eyes, because perception of 'quality' is also subjective, and personal expectations on quality (how much quality is good enough) vary widely.

[Guernsey]

Looks like I will have to play the waiting game or use the Slow preset. It is slightly faster than Slower and Veryslow.

[davexnet]

Looks like I will have to play the waiting game or use the Slow preset. It is slightly faster than Slower and Veryslow.

One would almost infer that from the preset names

[jagabo]

Here's an example encoded at crf=15 at both the slow and veryfast presets. The bitrate of the two encodings is almost the same. Watch them full screen in a dark room. The difference is very obvious.

[Alwyn]

Thanks Jagabo, very interesting. Stills don't look a lot different:

https://imgsli.com/MTU4NjU3

[Alwyn]

I watched both videos, full screen on my monitor in my darkened office and to be honest, I don't think I could pick the two apart in a blind test. Should I go to Specsavers?

[jagabo]

I watched both videos, full screen on my monitor in my darkened office and to be honest, I don't think I could pick the two apart in a blind test.

You don't see the horrendous posterization (banding) in the dark backgrounds of the veryfast video? The slow video has much less posterization.

[Alwyn]

Not at all, Jagabo. Not saying you're wrong, I just can't see it. Can you post a frame number and area to look for it?

[Sharc]

Thanks Jagabo, very interesting. Stills don't look a lot different:

... to be honest, I don't think I could pick the two apart in a blind test.

You don't see the horrendous posterization (banding) in the dark backgrounds of the veryfast video?

Nice example how subjective quality is, what matters to the viewer, individual tolerance level, and how difficult it is to give a general answer to the question of "how much quality improvement" will result.

Edit: And there is a difference regarding quality claims when a reencode is made for casual viewing or for archiving.

[Sharc]

FWIW here a script which I sometimes use for comparing variants on the PC, by stepping through the frames:

original=LWLibavVideoSource("original.mkv").subtitle("original")
v1=LWLibavVideoSource("very fast.mkv").subtitle("                     very fast")
v2=LWLibavVideoSource("slow.mkv").subtitle("                                                        slow")

#outputs for comparison:
out01=interleave(original,v1)
out02=interleave(original,v2)
out012=interleave(original,v1,v2)
out12=interleave(v1,v2)
outdelta01=subtract(original,v1).Levels(127, 1, 129, 0, 255)  #plain gray means no difference
outdelta02=subtract(original,v2).Levels(127, 1, 129, 0, 255)  #plain gray means no difference
outdelta0102=interleave(outdelta01,outdelta02) 

#select the output from the out* list above:
return out12

.... I just can't see it. Can you post a frame number and area to look for it?

All flat dark backgrounds. Using the (modified) script above for highlighting the differences between the 2 encodes, the banding contours pop up, e.g. around frame 380:


[Attachment 69520 - Click to enlarge]

P.S. If you can't see the banding of the "Very Fast" encode at all, you might have to readjust brightness, contrast, saturation .... of your TV. Maybe the banding is just masked (drowned in black).

[jagabo]

Can you post a frame number and area to look for it?

It's in every single frame (except the all black frames at the start).

[Alwyn]

Fair enough. I can't see it.

[Sharc]

Fair enough. I can't see it.

So the range is somewhere between "horrible posterization/banding" and "can't see the difference". Fair enough.

[Guernsey]

I did some encoding from my library and it turns out that Slow is faster while at crf 16 or lossless. I also wonder if memory can affect compression as well as the presets? Also, what is the difference between lossless and crf in regards to compression?

[davexnet]

https://m.youtube.com/watch?v=3nym3fkHqyk