Doing something wrong - custom profile terrible results

[Copnm]

Hello,

I too have been looking forward to VSO implementing 2 options: Variable bitrate by reading the average bitrate of the source, and Constant Quantizer. CQ is hands-down THE BEST way to convert a file if there is no size restriction, and 2-pass variable bitrate is THE BEST way to convert if there IS a size restriction. I'm honestly not sure if there are plans for these, but I would love to see these in future iterations.

I also would love to see more than just the "Use best video quality settings". I'd like to see either custom options (like Handbrake offers), or presets (like placebo, fast, slow, fastest, etc. similar to Handbrake's presets).

Adding all of these will improve end quality as well implement all the best encoding options into a program that is easy to use with an intuitive UI, which VSO is known for. But I digress...

Thanks for reading,
Neil M

[cedric]

Hi all,

Please, take a look on the last beta version (1.4.0.14) : http://forums.vso-software.fr/post90200.html#p90200

Regards,

[MyMojo]

Hi eon...

Was wondering if (when you get time) you could do your testing again and post your results like you did before the new beta .14 upgrade. It would be nice to see those same pictures and settings against HandBrake and DVDFab again, compared with VSO's new CQ settings. Thanks!

I've been playing with it most of the afternoon, and so far everything looks nice. I'm new to this stuff, so would be nice to see someone with more experience put it through the paces.

[eon_designs]

I'll do some testing and post screen shots at the weekend. Going to be away the next couple of days.

[cribber]

I've had a quick look at the settings and if I choose a preset (slow, slower, veryslow etc.) the converted clip's framerate stutters.

If I manually choose the number of reference frames, the converted clip (in MediaInfo) always says 2 no matter how many I choose (Is this because of CRF ?)

Constant Quality (CRF) of 20 still doesn't look as good as DVDFab's Constant Quality 20. Is the slider fully working ? I ask this because the filesize is also a lot lot bigger than DVDFab.

It's still very slow at converting, I think you need to add support for Intel QuickSync.

I have no desktop icon.

This converter is definitely moving in the right direction and I appreciate your work. I can't test any more at the moment because i'm going out now for a couple of days. Thanks.

[eon_designs]

Hello.

Managed a few test and so far quite impressed. The ref frames setting does not seem to work, it stays on 2 no matter what I select.

However, the new CF and CQ settings are looking good.

First up, constant quality CF, agree that using the same setting as DVD fab or handbrake does not look quite as good although the size is comparable. On all encodings done with a CF setting (tried 20 down to 16) the beginning of the clip looks terrible, but then it starts to improve. Bus still some noticeable blockynes.

Then trying the constant quantizer setting (CQ) gave a completely different result. Tried a CQ value of 20 and the result was fantastic, with quite a small size. 1.45gb in comparison to the same setting with CF which got a size of 1.7gb. Increasing the quality to CQ18 gave a file size of 2.1 gb and again excellent results. These were all encoding to 1280x720 and all took about 25 mins. Leaving the video at 1920x1024 took slightly less time 17 mins but became obviously much larger in size, 6.1 gb.

A final test of CQ19 at 1920x1080 gave a size of 4.9 gb mad again was a fantastic quality. As always the quality is always subjective, but I certainly think things are going the right way.

I'll try and post some comparison images between DVD Fab, VSO CF and VSO CQ settings at the same value of 20.

Looking forward to future updates and hopefully include a remove black bars option.

[cribber]

I read this somewhere.

RF=Rate Factor
QP=Quantizer Parameter

When you use CQP (constant quantizer parameter ), you compress every frame the same amount, regardless of the content.

When you use CRF (constant rate factor), it varies the QP slightly. When a scene has a lot of action and motion, it will raise the QP (compressing more). This is because your eye will be distracted by everything going on, and won't have time to see the heavier compression. When a frame doesn't have a lot of motion, it will lower the QP, compressing it less. This is because your eye will have more time to look at the image, so you want it to be as much like the source as possible.

If you were a computer, you would look at a CRF encoding and say it was lower quality than the CQP copy. And it would be. But if you're a human being, subjectively, the CRF copy will look better. It least compresses the parts you see the most, and most compresses the parts you see the least.

Both CRF and CQP are for single-pass encoding.

When you do a CRF or CQP encoding, you don't have any control over the output's size or bitrate. All you control is how heavily compressed each frame is.

If you need to get a particular size, you can do a 2-pass ABR encoding.


Personally, I always use CRF single-pass encoding. I don't see any reason to use CQP or 2-pass encoding. The latter is much longer, but still can give the result that is worse than CRF encoding. Anyway it might be a good idea to take a video sample and use different encoding methods to compare the results.

I don't know why DVDFab CRF 20 looks better than VSO CRF 20. It should look the same. Something is not right here. Also on the VSO version you said "the beginning of the clip looks terrible, but then it starts to improve". It should be good throughout. Maybe these new VSO settings they added are just placebo and is still using their original encoder with a few tweaks.

You can get rid of the black bars by changing the 'Video aspect ratio' setting in the custom settings. Unfortunately it doesn't work anymore in v1.4.0.15, another bug.

And VSO takes literally twice as long to convert (compared to DVDFab) probably due to the lack of Intel QuickSync.

Also, at the bottom of the main window is the quality advisor. It says the 'Estimated converted size (Not accurate with CRF)'. But it also says 'Quality factor used: VSO'. This shouldn't even be an option when using CRF.

[eon_designs]

Apologies for the delay, however here are some screen comparisons.

Encoded the same episode, all at 1920x1080 full AC3 5.1 soundtrack.

DVDFab version Constant Quality setting 20 = 3.1GB (7mins to encode using Quick Sync)- I got a similar size from Handbrake so just used DVDFab as the example.
VSO version Constant Quantizer setting 20 = 1.32GB (25mins to encode using CUDA)
VSO version Constant Quality setting 20 = 2.71GB (25mins to encode using CUDA)

All images on the left are from DVDFab (CRF20)
All images in the middle are from VSO (CQ20)
All images on the right are from VSO (CF20)

Looking at these the best are obviously DVDFab (CRF20) and VOS (CQ20), however I would say that DVDFab is just slightly sharper with more detail retained. BUT there is a considerable size difference between the 2. The VSO version is half the size of DVDfab . I am sure that if I lowered the CQ value then the detail would remain and the file size similar to that of DVDfab.

Can't wait to test the next version when Quick Sync is implemented.

[cribber]

That clearly shows that VSO CRF Quality is definitely not working properly, the top two sets of images (right image) look horrible. Hopefully Cedric is in the process of fixing this as it should look the same as the DVDFab ones.

[eon_designs]

Hopefully yes

But I think the CQ option is the way to go - file size to quality ratio is excellent (IMHO - it's all very subjective when it comes to visual quality)

[MyMojo]

I did my tests, and with CQ at 19, my movies are nearly half the size and look identical to the same video frames encoded with VC before CQ was implemented...I have since started re-converting my discs to the newer settings and love it. I have about 1800 DVDs and Blu-rays, and reducing them to near 50% smaller than the old settings, this is a HUGE savings in space! I'm very happy I bought this software, and when they get 8+ channel audio support built in, along with Quick Sync, it's going to be awesome.

[eon_designs]

I totally agree - the new implementation of CQ is fantastic. I'm considering doing the same with my collection. Thankfully I don't have quite as many discs as you to convert - mine are just in the few 100's not the 1000's like you. What a task that's going to be.

However, the benefits are worth it, especially when streaming from a NAS to my XBMC

Fantastic work by the VSO team - keep it up

[cribber]

The new version 1.4.0.18 is looking nice and tidy. I recently did a few quick tests and I think i've found the problem to the poor CRF quality.

In the x264 custom profile, CRF quality set to 20 (for example)
profile set to main :

x264 Main.jpg (9.27 KiB) Viewed 10415 times

The resulting encode

If you set it to high :

x264 High Problem.jpg (9.29 KiB) Viewed 10415 times

The resulting encode looks like a complete mess

I used this clip to test (very good for testing)
http://www.auby.no/files/video_tests/h2 ... _birds.mkv

P.S. Why do the thumbnails in this forum only show up when you're NOT logged in ?

[MyMojo]

cribber, is that a blown-up view of a small area in the frame, or the whole picture? I am using CRF @ 19 with High 4.1 and mine look awesome using .18.

However, I played a couple of movies on the plasma to test entirely, and I'm getting audio problems.... sound blips, and even complete audio dropping out. For the drop-out problem, if I fast forward the movie, then reverse to a few seconds before it dropped out, it will play the audio again. But if I let it play, it stays off. So the audio is there, but something in the file is causing it to 'mute' for some reason. I haven't had time to narrow it down yet, but when I do, I'll make some posts with logs and samples.

[cribber]

cribber, is that a blown-up view of a small area in the frame, or the whole picture? I am using CRF @ 19 with High 4.1 and mine look awesome using .18.

It's just a section I cut out from the 1080p clip so it wouldn't take long to load/upload. You notice the poor quality on very busy scenes, that's why I posted the link to the clip I used as it's very good for testing (it's a large group of flying birds). Try the test yourself, it doesn't happen on Handbrake.

[MyMojo]

interesting. thanks

[cedric]

Hello cribber,

Please, can you post your log file?

Thank you!

[cribber]

Hello cribber,

Please, can you post your log file?

Thank you!

[cribber]

Just did another test and it's caused by CUDA again.

My Laptop

Intel Core i5-4200M
HD Graphics 4600 (Haswell)
nVidia GT 745M (2 GB) CUDA v3.0 (Driver version 340.52)

[cedric]

Hello cribber,

"Try the test yourself, it doesn't happen on Handbrake."
Do you used CUDA when encoding with handbrake? (I don't think because I can't find hardware encoding option in handbrake...)
So, if you want to do some comparison tests, you need to use same encoder mode in both programs (software H264 encoding).

I will check if I can found something wrong in CUDA high profile, but keep in mind that choosing CUDA encoding will give higher speed BUT lower video quality, you need to choose...
CUDA encoder video bitrate is often limited to 12Mbits/s (depending on graphic card), when software encoding is not limited.
I done same test as your, and handbrake result gives 24Mbits/s video bitrate when CUDA gives 12 Mbits/s, so we can't really compare these 2 files.

Regards,