Dramatic Inconsistencies

[Shibblet]

I just did a comparison on a DVD Encoding and came up with some inconsistencies. Obviously you normally exchange file-size for time when it comes to encoding. Most of us want the best picture quality (Same as source), with the smallest file-size, along with a quick encoding time.

So, I decided to encode the living heck out of this DVD to see which one gave the best file size. The CQ meter is at 20, All filters off, constant framerate, basically, I left "Normal Profile" the way it was.

Check out the results.

Princess Bride - DVD

Ultrafast
Size - 2,375,224,991
FPS - 755.381042

Super Fast
Size - 1,421,447,444
FPS - 599.621521

Very Fast
Size - 827,256,545
FPS - 427.930603

Faster
Size - 874,089,089
FPS - 284.734222

Fast
Size - 927,826,111
FPS - 207.975296

Medium
Size - 919,937,074
FPS - 168.014359

Slow
Size - 907,121,842
FPS - 108.336792

Slower
Size - 908,035,839
FPS - 62.949097

Very Slow
Size - 823,583,675
FPS - 35.665752

I am a bit stunned how "Very Fast" produced results almost as good as "Very Slow" but in 1/12 the amount of time.
Can anyone explain why this is? Are all of my movies going to be just as good when encoded at "Very Fast" ?
What's the Dealio-Yo?

[Shibblet]

1080p Test Video (2nd Attempt)

Stranger results this time, but with "Very Fast" still coming out on top for quality and this time, smaller file size!

Ultrafast
Size - 265,527,555
FPS - 22.275557

Super Fast
Size - 230,733,239
FPS - 26.797058

Very Fast
Size - 84,173,827
FPS - 24.327824

Faster
Size - 94,324,671
FPS - 17.469124

Fast
Size - 103,345,249
FPS - 22.331667

Medium
Size - 103,588,937
FPS - 20.050056

Slow
Size - 99,919,727
FPS - 22.109766

Slower
Size - 101,883,207
FPS - 18.932636

Very Slow
Size - 84,525,026
FPS - 14.846879

[mduell]

Constant quality isn't quite constant quality across different speed presets. Use tune SSIM and compare the SSIM values for the two very presets and you'll see the difference.

Also I wouldn't put much stock in your fps numbers; are you doing something else that uses the CPU while you're running these "tests"?

[Shibblet]

Constant quality isn't quite constant quality across different speed presets. Use tune SSIM and compare the SSIM values for the two very presets and you'll see the difference.

Not really sure what that means. Can you explain?

Also I wouldn't put much stock in your fps numbers; are you doing something else that uses the CPU while you're running these "tests"?

Nothing. Ran Handbrake and got the Average FPS results from the log files.

[mduell]

Constant quality isn't quite constant quality across different speed presets. Use tune SSIM and compare the SSIM values for the two very presets and you'll see the difference.

Not really sure what that means. Can you explain?

Which part?

[Shibblet]

Constant quality isn't quite constant quality across different speed presets. Use tune SSIM and compare the SSIM values for the two very presets and you'll see the difference.

Not really sure what that means. Can you explain?

Which part?

I understand there will be some inconsistencies with a CQ slider, and speeds. What is SSIM?

[mduell]

Structural similarity, a measure of video quality.

Tune SSIM disables the psycho-visual optimizations that x264 has which make the video look better but lower the objective quality (as measured by SSIM). With that done you'll see the quality improvement for the slower presets that produce the same file size as the faster presets.

[Shibblet]

Structural similarity, a measure of video quality.

Tune SSIM disables the psycho-visual optimizations that x264 has which make the video look better but lower the objective quality (as measured by SSIM). With that done you'll see the quality improvement for the slower presets that produce the same file size as the faster presets.

Then what is the purpose of the slower presets?

[Deleted User 13735]

Slower presets try to give better compression. YMMV
The RF slider balances quality with everything else.
That's why you would need a quality metric (SSIM), otherwise nothing is valid.
Very Fast is quite popular, because it is efficient, as you've already noted.

Most of us want the best picture quality (Same as source), with the smallest file-size, along with a quick encoding time.

You are asking for a free lunch. That's not going to happen. Pick two.

[Shibblet]

You are asking for a free lunch. That's not going to happen. Pick two.

Yep, the age old addage. Pick two of the three. That's been a staple statement in the encoding world since KVCD's (Nobody remembers?)

So, then I should ask the question... What's the problem with "Very Fast" ? Or is there one?

[mduell]

So, then I should ask the question... What's the problem with "Very Fast" ? Or is there one?

Poor compression efficiency.

[Shibblet]

Poor compression efficiency.

Not to beat a dead horse, but sometimes you get a smaller file size with VeryFast. And the times that you don't, you are in the same ballpark with VerySlow. It's never as dramatic of size difference as UltraFast or SuperFast, or the next step down Faster. It's a bit of a conundrum, that's why I am wondering what the differences between them are?

[JohnAStebbins]

"Compression efficiency" doesn't equate higher efficiency to smaller files. Compression efficiency is a function of both file size and quality. For the faster presets, if you see the file size get smaller, you will also see the quality decrease. If you were to adjust your settings so that the slower and the faster preset resulted in the same quality (typically through adjusting the RF value), then the slower preset will result in a smaller file. If you were to adjust your settings so that the slower and faster preset resulted in the same file size (through selection of the same ABR bitrate), then the slower preset will result in better quality.

Note that a given RF value does *not* result in the same output quality for two different presets. I.e. RF 20 with preset "Very Fast" *will* result in different output quality than RF 20 with preset "Medium". RF is a relative scale. This scale changes every time you change any other x264 setting.

[Shibblet]

"Compression efficiency" doesn't equate higher efficiency to smaller files. Compression efficiency is a function of both file size and quality. For the faster presets, if you see the file size get smaller, you will also see the quality decrease. If you were to adjust your settings so that the slower and the faster preset resulted in the same quality (typically through adjusting the RF value), then the slower preset will result in a smaller file. If you were to adjust your settings so that the slower and faster preset resulted in the same file size (through selection of the same ABR bitrate), then the slower preset will result in better quality.

Note that a given RF value does *not* result in the same output quality for two different presets. I.e. RF 20 with preset "Very Fast" *will* result in different output quality than RF 20 with preset "Medium". RF is a relative scale. This scale changes every time you change any other x264 setting.

Yep, I getcha on that one. So, I should see a difference between the video quality if I encode at RF 20 with VeryFast and VerySlow. Hmm... if there is a difference It's next to impossible to see. And size of the file doesn't matter too much for me, like I said before, they're in the same ball park.

However, with VeryFast, I can encode an entire DVD Series in under a day. Blu-Ray rips that used to take upward of 10 hours, seem to come out just as good, within a couple hundred MB of the same file size, but now in about 2 hours. I may just have to stick with this setting.

Gotta love HandBrake and all of the features it has!

[s55]

Quality is subjective. Some people see a big difference, some don't. Eyesight, what type of display you have, display size, etc, etc, etc all play a part.

Even then, some sources will show more drastic differences than others.