Test encode at high bitrates with and without TS results

[markrb]

I did the entire movie (Ghostworld) with and without TS(2,1) at high bitrates(3250 MAX, 2750 AVG) using 4 pass VBR and then compared the results. Based on the post DDogg made I was expecting the TS encode to be worse with such a high bitrate, but there was no noticable difference. I did not see an increase in mosqiuto noise or any blurring. No difference in file size either.
I watched both in full screen on my 19" monitor as it shows defects in my encodes better then my projector does.

Here are some figures.

Movie: Ghostworld
Resize: Bicubic .60
TS: 2,1
Bitrate:
600 Min
2750 AVG
3250 MAX

Mark

[DDogg]

I wish you had the time to do the exact same thing at 2200

[markrb]

I will the moment I get a chance. I have enough disc space now that I generally keep the Vob files for awhile. At least until I am sure about the discs playing all the way through on my DVD player.

Now when you say 2200 are you talking AVG or MAX?

Mark

[DDogg]

Just let D2S calculate the appropriate bitrate for a 2 hour movie. Don't manually bump up the rate. In effect, a standard encode.

[markrb]

I started an encode with TS last night with the following settings:

Max 2500
Avg 2000
Min 300

A pretty standard encode. I could drop the AVG bitrate and start over if you think I should. Maybe to 1900 or so?

Edit Computer crashed(Promise HD controller failed). Starting encode over and dropping the AVG to 2000 as I think this is closer to the norm of an encode.

Mark

[Trahald]

2,1 is pretty tame... besides adding length to the encode i wouldnt expect it to do much - negative or positive

try 3,9 and see what it does ;)

[markrb]

After I finish with (2,1) and no TS. I will try 3,9 and compare.

Can anyone point me to a description of what the strengths actually mean and an explanation of how much stronger 3,9 is over 2,1?

Yes I am being lazy and I haven't looked myself yet, but right now it's 4:13am and I am about to pass out.

Mark

[markrb]

@w00kiee I have been reading around the forum and all the Avisynth info I can find and only two sets of strengths have been recomended the first is the default of (2,1) the second a little stronger at (3,5) nowhere does anyone suggest using as strong a setting as you suggest. Where do you get your info from? Even Dividee doesn't suggest to use anywhere near that strength.
In fact it is often posted in the Avisynth forum that a high setting with TS will cause extreme blurring.

Since it takes way to long to do test encodes of a whole movie I have broke down and decided that for the rest of my testing I will do no more then 3 chapters. It's about 20 minutes. Still with an average bitrate of 2000 and 4 pass VBR. I have already done the TS(2,1) encode, now I am doing the no TS encode and next I will do a test at TS(3,5). If I have time I will try (3,9) as well.

Started the (3,5) encode a little while ago. Wow this is really slowing my encode down. With TS(2,1) I got just under 1.0 with CCE. Now with TS (3,5) it's only .576. I get around 1.49 with no TS and Bicubic of .60

Mark

[markrb]

This was tougher then I expected.
For the background and non moving objects the greater TS 3,5 was the best at getting rid of noise and behind that was 2,1 and in last place was no TS. They all still produced noise around moving objects and foreground objects. The 3,5 encode took almost twice as long to complete then the 2,1 did.
None of the encodes seemed to blur or reduce picture quality at all. The written signs in the background were all just as sharp in each encode.
In each case I left the CCE noise reduction on as well at the default setting of 2.
At this point I can't say if the increase reduction of noise with both TS settings was enough to compensate for the loss of speed.
Further testing is required.

Edit: Did another test to compare CCE's noise reduction vs. TS. I did a quick encode with no TS and a noise reduction setting of 10 and I saw no difference to it set at 2. I am currently doing an encode with CCE noise at 10 and TS at 2,1.

The following are tests I plan to do all at 2000 AVG and 4 Pass VBR:
TS 2,1 CCE noise 10 and 2 Bicubic at .50
TS 2,1 CCE noise 15, 10 and 2 Simple resize
CCE noise 15 Simple

I will find the best settings if it kills me. I rule out TS 3,5 because it takes way to long although it has proven to me to give better background results.

After I find the best results at 2000 I plan to do some of these tests again at 3000 since my new DVD player handles that bitrate fine now.

2nd Edit: Finished TS 2,1 CCE noise 15 Simple resize.
So far it has proven to have good results regarding noise, although the image is slightly blurred there is much less noise around moving objects and even the background is almost as good as TS 3,5 while still getting a CCE speed of over 1.0. I am now trying the same setting, but dropping CCE noise from 15 to 10 to hopefully reduce the blurring. The image is less sharp then Bicubic and that is more important I think.

3rd edit: I am not sure of anything anymore. I have made about 9 encodes with different settings and they are all starting to look the same to me. Maybe because it is light out now and I can't be as critical with the video or the fact that I can see noise in every encode no matter what settings I use. The difference in sharpness does not appear as great as it did last night. They all seem very close now.

4th edit: In every case Bicubic .60 with no TS and a noise setting of 2 was better to my eyes then every test I did with Simpleresize. The image was sharper and the amount of noise was only slightly less if at all. I am starting a new set of encodes with Bicubic at .75 and raising the CCE noise filter. I am hoping that the extra sharpness I get from .75 will offset the blur/smoothing that a high CCE noise filter number gives to give a good picture with less noise. Only time will tell.

Mark

[mfrancis123]

I'm impressed with the testing you have been doing. Keep up the good work. I've finally started getting results with DVD2SVCD that I can play on my standalone but I keep seeing contrary opinions in the guides and in the forum. What are your thoughts on:

Simple Resize vs. BiCubic
4 passes vs. 5
Bicubic .5 vs. 75
Temporal Smoothing Yes/No

Most of my encodes are 2500 max 2250 avg max

I'm not concerned with encoding time since they usually run while I sleep and when I'm at work.

Currently I'm doing AI with 2500 max 2200 max avg bicubic .75 5 passes, I'm trying to get the best quality and 2500 is my max bitrate.

Most of my encodes have noise when things are moving around which filter would help correct this?

Thanks

[gerti67]

Hi Mark,

very good testing indeed! - I just want to add something from my experiences as I've got quite an arsenal of little chapters from my SimpleResize testings with all possible settings. ;) (But did not had enough time to test TS yet)

In my opinion the CCE noise filter only really kicks in when using SimpleResize - seems the resize algorithm which really only resizes as Tom always said - leaves some "good food" for CCE's noise filter and so the softening (blurring) of the picture is much stronger as it is with the BicubicResize - but from my experience with any value higher than 15 with it you won't see any difference anymore to those done with 15.

In fact, I found using SimpleResize with CCE noise filter of 15 or above renders the movie like the BilinearResize, leaves more mosquito noise than the bilinear one but it still has sharper edges than the bilinear one. And SimpleResize with CCE noise filtering still is 10% faster because the speed decrease with the CCE noise filter tends to zero when using it all over the range from 1-32 with my testings.

And I can really understand your confusion after watching all the samples - I had to rest for one day and then get back to see some differences again. ;)

Greetings,
Gerti

[DDogg]

In every case Bicubic .60 with no TS and a noise setting of 2 was better to my eyes then every test I did with Simpleresize.
That's kinda what I ended up thinking too. :) (after a lot of head banging).IMO, those should be the new default settings.
I am hoping that the extra sharpness I get from .75 will offset the blur/smoothing that a high CCE noise filter number gives to give a good picture with less noise. Only time will tell.
That should be interesting. I don't really trust the cce noise filter at those levels but that is more of a gut reaction than one based upon data.

[markrb]

I have done many encodes over the last couple of days and so far the only one that even touches Bicubic .60 No TS CCE noise filter 2 is
Bicubic .60 TS 3,2 NO CCE noise filter. The second is actually slightly better, but IMHO not enough to make it worth the extra encode time. This is for overall picture quality. In some encodes there was less noise, but the image blurred some. In others the image was very sharp, but there was a lot of noise. These two have proven so far to be the best comprimise.

I have completed the following tests all at VBR 4 pass on the same chapter of Ghostworld. This is a very film based movie with muted colors and not much action.

Bicubic .50 No TS CCE noise 2
Bicubic .60 No TS CCE noise 2
Bicubic .60 No TS CCE noise 10
Bicubic .60 TS 2,1 CCE noise 2
Bicubic .60 TS 3,9 No CCE noise
Bicubic .60 TS 3,5 CCE noise 2
Bicubic .60 TS 3,2 No CCE noise
Bicubic .75 NO TS CCE noise 10
Bicubic .75 NO TS CCE noise 15
Bicubic .75 TS 3,2 No CCE noise
Simple NO TS CCE noise 2
Simple No TS CCE noise 10
Simple No TS CCE noise 15
Simple TS 2,1 CCE noise 10
Simple TS 2,1 CCE noise 15
Simple TS 3,3 CCE noise 10

Testing continues.

@mfrancis123 As I have said so far I prefer .60 NO TS with a CCE noise setting of 2. I prefer 4 pass, but if I feel silly will do a 5 pass as anything over 4 pass really doesn't give much improvement.
This is all at an average of 2000. If I were to use anything higher or lower these results would not hold up. I will be repeating many of these tests at much higher rates as my DVD player supports up to 3500 kbit video.

Mark

[DDogg]

Markrb, can you take one of the encodes and run it with a min of 755 or 900?

[markrb]

Pick which one and I will do it.
If I were to pick one I would take the one that already had the best results and see if they got better, but if your intent is to see if a poor encode can become better I can do that as well.

I just finished another encode. This time it was with much higher bitrates. The settings were Max 3250, AVG 3000 and Min 600 with TS at 2,3 and NO CCE noise. As expected it was much better then any of the 2000 AVG encodes although there was still a good deal of noise, but less then the lower rate encodes.

Mark

[sandchar99]

@Markrb

When you find the best high bitrate settings, will you do one more test?--as if your plate wasn't already full.

Your 4 Pass VBR vs CBR and 1 Pass VBR at whatever you find to be the optimal settings regarding TS vs no TS, bilinear (whatever c value) vs simple, and anti-noise settings.

I might likely to do the same as I have an ugly "hybrid" encode tonight (first half is Film and the rest is interlaced) that I'm going to do as a high bitrate encode and spread it out over 3 cds. Most likely, I will do it as CBR.

[markrb]

I did a dangerous thing and I started comparing the encodes to the original VOB file. In all cases the image was not nearly as sharp as the original. I guess this is to be expected, because I have dropped the bitrate quite a bit. I was surprised to see the .75 encode is still not as sharp though. I always thought that this setting sharpened the image artificially.
I did notice that some of the noise I was being critical on in all my encodes was there in the original as well. The mosquito noise and the larger blocks around people was not there, but the small background pixel noise was there.

I did try some outside filters like the smart smoother by Donald Graft, but these made the colors artificially deeper and brought them out more.

The quest continues for the best settings. I am pretty sure I will be working on this for a long time. I am too much a tweaker at heart.

Mark

[markrb]

Found the best way to encode so far, but it is complex and very time consuming for video encoding.
What I did was to edit the avisynth file and add two new filters.
The first was Donald Graft's Xsharp filter and the second was his tweak filter. The tweak filter was needed because Xsharp needs the video to be converted to RGB to work and this caused the contrast and color to change slightly. The tweak filter returned it to normal.
It came out really nice once I also added TS and a CCE noise filter of 10. Problem is my CCE speed with all of this is a very low .25-.3 when I can normally get a speed of almost 1.5 for the same video not using these filters. I am hoping once Donald Graft ports the Xsharp filter to work natively with Avisynth that I can drop the Tweak filter and not have to port through Vdub, which is the way it works now and this would speed it up.

Until then I need to find the best way to encode without using these filters as they take way to long to work.

Mark