Log in

View Full Version : Questions about AutoRV9


baddbill
24th December 2002, 20:42
I just made my first two AutoRV9 encodes and love them. It is a wonderful codec. But I have several questions about the RV9 and AutoRV9 that I thought others might be able to answer.

1. Is the audio encoded in VBR by default?

2. Since AutoRV9 doesn't have a comp test I've been using the one in Gordian Knot to decide what resolution to use. Is this right? Is the comp test in Gordian Knot designed for divx compressability and therefore useless for RV9? Or is it pretty much universal and will tell you how compressible a movie is for any codec?

3. I've never used the Lanczos resize filter. How does it compare to neutral bicubic? How does it affect the compressibility of a movie compared to neutral bicubic?

4. Under the audio tab should I choose stereo or surround for channels mode? I notice that changing it doesn't change the size of the audio file. I figured surround would require more bits to encode.

Any advice or input would be appreciated.

Bill

karl_lillevold
25th December 2002, 14:05
I am happy to hear you like the visual quality you get with RV9.
Let me try to answer some of your questions:
1) RealAudio is always CBR
2) I am not very familiar with the GKnot compressibility test. I am guessing the test is pretty universal, but the numbers might be a little different for RV9.
3) I like Lanczos for visual quality, and have used it for a few encodes, but have not done any comparisons with regards to the resulting compressibility. If anyone has, I would be curious to know their results.
4) Generally you probably want to choose surround for multi-track source material, unless you never ever intend to play back your compressed file with more than 2 speakers. The bitrate is the same, but the codec parameters are chosen such that the down-mixed surround information is better preserved.

iago
25th December 2002, 14:43
Regarding LanczosResize(...,...):

It's very similar to BicubicResize(...,...,0,0.75) both in terms of compressibility and visual quality. However, the visual quality of Lanczos, personally, pleases me more than "sharp bicubic".

Btw, I have not used it with RV9 yet, I'm pretty happy with my XviD encodes ;).

regards,
iago

RadicalEd
25th December 2002, 19:22
Originally posted by karl_lillevold

RealAudio is always CBR


serious?? wow, thats pretty amazing quality for cbr encoding... anyway, the main question is why CBR?? Wouldn't VBR substantially increase the (already excellent) quality?

karl_lillevold
26th December 2002, 05:56
Originally posted by RadicalEd
serious?? wow, thats pretty amazing quality for cbr encoding... anyway, the main question is why CBR?? Wouldn't VBR substantially increase the (already excellent) quality?
I had to get the thoughts from the codec engineer who designed and developed RA8. So here's what he said about CBR and VBR audio:


: My RA8 is CBR due to roots as streamable codec
: ATRAC3 is also strict CBR to facilitate playback with low-memory devices with constant (low) read speed from storage
: VBR can probably make slight improvements on some content
: but can make the worse-case behavior even worse
: causing a large variation between best and worst case
: along with other undesirable properties of VBR
: i think the best is to support both modes. i dont think "pure VBR" is such a great solution, all things considered.

Dark-Cracker
26th December 2002, 14:14
@ baddbill

hi,

i will add a compressibilty check on the next version, the gknot comptest will only get u an idear of the result but the result will not be the same with the rv9 compressibilty check (because gknot use divx5 to make the test).

Bye.

baddbill
26th December 2002, 21:33
Thanks for the info guys.

@ karl_lillevold

I have to tell you that in my opinion RV9 is a far superior codec to DivX 5.02. You can read about my experience with it in your sticky in this forum. My hats off to you for such an incredible codec.

A couple more questions:

1. If I encode audio at 132 and stereo, what bit rate should I use for surround to get the same quality?

2. Is there a chance VBR audio might be available in the near future? It really is superior if you have the cpu power to run it (most computers).


@ Dark-Cracker

Thanks for such a wonderful program. I am very impressed with the results. If you need anyone to help test the comp test or any other features of the new AutoRV9 keep me in mind. I'd love to help. If I was able to help with the developement I would. Unfortunately I don't know how to program.

I noticed that AutoRV9 resizes before it uses a video filter. What is the reason for this? I thought that filtering and then resizing resulted in better quality. Can I just change the AVS script around without any ill effects?

thanks,

Bill

Dark-Cracker
27th December 2002, 05:27
@baddbill

if u want to help me wait the beta test for the v1.3 and make some tests to find the (potential :) ) bugs, this will help me to improve my tool.

i filtered the movie after the crop & resize because less the surface is important and much faster was the filtering. i have make some tests and there is no (visible) differences if u filtered before the resize, i only keep the deinterlace & ivtc filter before the crop and resize to be sure the filteres find the right field order.

Bye.

karl_lillevold
27th December 2002, 11:54
Originally posted by baddbill
1. If I encode audio at 132 and stereo, what bit rate should I use for surround to get the same quality?

2. Is there a chance VBR audio might be available in the near future? It really is superior if you have the cpu power to run it (most computers).

Thanks for your feedback, bill. Here is at least one answer to your questions:

1. At 132 kbps you should not need to increase the bitrate, since the Dolby surround information encoded in the phase difference between the two channels is well preserved.

2. I am afraid I can't answer that question ;)

baddbill
27th December 2002, 20:51
@ Dark-Cracker

Just send me a PM when you need some but testing. I can't wait to play with the comp test.

I wanted to let you know about this thread that talks about resizing and filters.

http://forum.doom9.org/showthread.php?s=&threadid=36344&highlight=c3d+and+resize

It appears that for the best quality you should use a noise filter and then resize. It may take longer but the results are better. Maybe you should put an option in AutoRV9? Or make the default to denoise after a soft bicubic or bilinear resize. But to denoise before a neutral bicubic, sharp bicubic or Lanczos size filter, so that the sharper resize filters don't amplify the noise.

I'm no expert this is just my opinion. I thought you might appreciate a little input on your wonderful program.

Bill

31 Flavas
27th December 2002, 23:08
Originally posted by amirm

For 5.1 btw, you can start at just 128kbps which gives you three times the channels of MP3 in the same space!I hope i'm right here, but three times the channels means one third the bitrate to each channel, right? or ~~21k/bit per channel?

Originally posted by karl_lillevold

1. At 132 kbps you should not need to increase the bitrate, since the Dolby surround information encoded in the phase difference between the two channels is well preserved.But how does it work with RealSurround audio? At 132kbps, when played back with pro-logic equipment, are the maxtrixed channels 66k? or 22k? or something in between?

baddbill
31st December 2002, 07:35
@ Dark-Cracker

In Autorv9 I edit and save the AVS script. AutoRV9 says that the AVS script was saved successfully. If I then open the AVS script it shows me the original unaltered AVS. This has me wondering if Autorv9 is using the original or my altered AVS script. Any ideas? I would expect that after I save my altered AVS script and then reopen the AVS I would see my altered AVS script.

I've also encoded a 4 1/2 minute clip three different times. One without a noise filter, once with light C3D and once with extreme C3D. I can't tell the difference between the three encodes!

When does AutoRV9 run the C3D filter? I thought the noise filter would have it's own step in the encode process but as I watched the encode process I didn't see this happen. Does C3D filter each frame right before each frame is encoded?

How can I be sure the C3D noise filter is actually being run?

Any info would be appreciated.

Thanks,
Bill

Dark-Cracker
31st December 2002, 20:02
Hi,

to be sure the .avs was modified go in your movie directory and open the XXX_Movie.avs with the notepad and verify if the changes have been saved. i think it surely the preview button who have failed.

for the c3d filter it work fine :) to be sure make a 1 pass (because 2 pass correct the curve to obtain the correct size).
in 1 pass mode test the different preset (light..extreme) and look the size, normaly the c3d filter remove noise (this will increase the compressibilty and your finale size will be smaller than without the filter.), else u can make some screen shot on some noised frame and look the difference.

PS: but it's hard to see the difference with the postprocessing of the rv9 .

Happy new year.
Bye.

baddbill
31st December 2002, 21:07
Thanks for the info Dark-Cracker. You are right. If I open the AVS with notepad I see the changes that I made.

I will try making some 1 pass movies and see if the file size changes.

One more question for you. If I change the AVS from this:

#
# RESIZING
Video=Lanczos3Resize(Video,704,384)
#
# CONVOLUTION 3D FILTER (Extreme)
Video=Convolution3D(Video,0,8,8,8,8,2.8,0)

to this:

#
# CONVOLUTION 3D FILTER (Extreme)
Video=Convolution3D(Video,0,6,10,6,8,2.8,0)
#
# RESIZING
Video=Lanczos3Resize(Video,704,384)

Will this work? Or do I need to remove the (Extreme) if I'm going to change the numeric values? Does leaving the (Extreme) in, default the C3D to 0,8,8,8,8,2.8,0?

Thanks again,
Bill

karl_lillevold
31st December 2002, 21:27
Originally posted by 31 Flavas
But how does it work with RealSurround audio? At 132kbps, when played back with pro-logic equipment, are the maxtrixed channels 66k? or 22k? or something in between?
I am not quite sure I understand the question. At 132 kbps, I believe RealAudio ATRAC-3 encodes the two channels separately, preserving the phase difference information flawlessly, and the Pro-logic equipment simply extracts the surround signal from the two decoded channels. At lower bitrates, the two channels are encoded in a combined manner, but still encoded such that the surround information can be extracted from the decoded signal. RealAudio Surround does not need any overhead or extra channel to encode the surround signal, it simply sets its parameters such that the surround signal is carried in the two stereo channels. This reduces the bitrate requirement, but it is admittedly not a true discreet 5.1 solution.

Emp3r0r
31st December 2002, 21:35
This is AVISynth basic scripting
#
# CONVOLUTION 3D FILTER (Extreme)
Video=Convolution3D(Video,0,6,10,6,8,2.8,0)
#
# RESIZING
Video=Lanczos3Resize(Video,704,384)is exact same as
Video=Convolution3D(Video,0,6,10,6,8,2.8,0)
Video=Lanczos3Resize(Video,704,384)Any line with '#' at the beginning is a comment and doesn't affect anything.

Dark-Cracker
2nd January 2003, 01:19
Hi,

yes emperor was right :)

baddbill
2nd January 2003, 01:44
Thanks for the advice guys.

I encoded a 4.5 minute section of The Waterboy at 512x272 with AutoRV9 in one pass mode with a target size of 45 megs. I used Lanczos Resize and set video mode to sharpest motion. I put c3d in the AVS script before resize. The results are:

With no c3d file size was 45.3 megs.
With c3d set to HQ film file size was 45.2 megs.
With c3d set to LQ film file size was 45.1 megs.

Does this sound right? I thought I would see better compressibility with c3d and RV9. Maybe RV9 has a noise filter built into it and that reduces the effect of c3d.

With Gknot I encode the same section of The Waterboy at 576x320 with sharp bicubic resize filter. The comp test results where as follows:

No filter = 65.7%
c3d at HQ film = 69.1%
c3d at LQ film = 73.5%

These results make sense to me, a 8% gain in compressibility with c3d set at LQ film.

Any ideas?

Bill

Dark-Cracker
2nd January 2003, 09:12
@baddbill

yes c3d increase compressibility for the rv9 i think u should wait the versio 1.3 of autorv9 with the support of the compressibility test this will surely give you a better idear of the powerfull of this filter :)

perhaps try to make your test with c3d after resize :) and try to encode in 1pass 100% quality with the rv9.

PS: the compressibilty test only use 5% of the movie and must be use to give u an idear of the compressibilty of the movie , try with gknot to encode in 1pss and compare the final size.

Bye.

The Belgain
4th January 2003, 11:51
So is there no discrete multichannel audio possible in RV9 then? Is the only surround option stereo with DPL?

Rash
4th January 2003, 15:22
Dark Cracker, just a question. Is there a way of producing "size-free" rv9 with AutoRV9? I mean, do we always have to aim for a final filesize or can I simply type the avg bitrate and encode?

Thank you ;)

Dark-Cracker
4th January 2003, 16:08
normaly u must enter the desired finale size and the avg bitrate was calculated (it use the audio bitrate & the overhead size value to find the right bitrate) but enter a bitrate manualy could be possible even if i don't see the interest :)

Bye.

karl_lillevold
4th January 2003, 17:15
Originally posted by The Belgain
So is there no discrete multichannel audio possible in RV9 then? Is the only surround option stereo with DPL?
You are right, DPL Surround is the only option for now.

The Belgain
4th January 2003, 19:38
Ok, that's what I thought. It's just that the comments on real's site are a little misleading (not incorrect, just misleading). Here's a quote taken straight off real's site:

With RealAudio Surround, you can enjoy 5 channel audio as well as a dedicated subwoofer channel.

Admittedly the previous paragraph does mentions DPL, but all the same...

I have to say that despite the fact Real have come up with a really good codec, they haven't been able to resist making overly optimistic claims about it (VHS-quality at 160kbps anyone?, same quality as MPEG4 at 1/2 the size) which is a shame.

Rash
5th January 2003, 05:49
Originally posted by Dark-Cracker
normaly u must enter the desired finale size and the avg bitrate was calculated (it use the audio bitrate & the overhead size value to find the right bitrate) but enter a bitrate manualy could be possible even if i don't see the interest :)

Bye.

Thanks for your answer. ;)
Well, it won't allow me to enter a bitrate manually (it is grey). But I won't disturb you with this anymore. :D

Sometimes I like encoding things by quality and not really aiming a filesize. So, as I still don't have enough knowledge on RV9 to guess a final filesize for a good quality just by looking at the movie duration and resolution, I use codec's bitrates to encode them. :)

karl_lillevold
5th January 2003, 15:33
@Dark-Cracker:
You mentioned you needed documentation for anamorphic encoding, so in case you want to take a look, over in this thread (http://forum.doom9.org/showthread.php?s=&threadid=40572) I have added instructions for the manual steps that are needed in AutoRV9 1.2 for anamorphic encoding with RV9.