Ateme H.264 HP Beta Test : Bugs & Issues

[bobololo]

Hi there,

So here we are, the new round of our new encoder is about to start and we all hope it will have as much success as the previous one.

We're currently testing internally the latest builds in order to smash out the latest obvious bugs and just before unveilling the beta to the testers. We should put it online on tomorrow night or at most on wednesday night. By that time, testers will receive by email the instructions and the url to get the beta package. They will need their personal login and password they should have received by postal mail.

We'll keep the same scheme for collecting your feedback as previous. We'll have this current thread which is intended to discuss about the bugs and issues you may meet (encoder crash, filters that don't work, etc.) while another thread "Quality feedback" will deal with quality feedback and discussions.

We're very excited to show you our latest improvements and we hope you'll enjoy them !

-- bobo

[LigH]

That's fine. "Fun and success!"
__

Can you tell us a bit about the range of systems you already own for tests? Which are the most/least powerful, or "most exotic" systems?

[acidsex]

Did my login information arrive in postal or email? I have yet to receive either as I have been away on a business trip.

Thanks

[chilledoutuk]

it will be snail mail well mine was.
Wating patiently for the beta test to start.

[Zarxrax]

I just recieved my login info in the mail today, so it might still be a little while before you get yours.

[bobololo]

A little high profile clip encoded with the latest pre-beta build:

ftp://mood.ateme.com/beta/cinderella-ateme-3000k-hp.mp4

with 8x8 transform, custom scale matrix (flat) and 4 slices (encoded on a bi-xeon).

Enjoy :)

[peteag]

enjoy? i'm on your knees guys. this codec is a world-wonder!

[jfehr]

Tease!

Very nice though. Can't wait for the beta to start. :)

A little high profile clip encoded with the latest pre-beta build:

ftp://mood.ateme.com/beta/cinderella-ateme-3000k-hp.mp4

with 8x8 transform, custom scale matrix (flat) and 4 slices (encoded on a bi-xeon).

Enjoy :)

[Sergejack]

Is that codec the property of ahead ?

EDIT : I'm wondering how much MB does that sample do.

[jfehr]

Its a 1280x720 HP encoded video stream, encoded at 3000Kbits/second. The file is 53 MB for 2 minutes, 28 seconds. The codec is Ateme/Ahead's (I imagine its joint, anyway)...

I can't play it without dropped frames on my P4-3Ghz though. I hope the DVD players that'll be capable of playing these kinds of streams come out soon. :)

[ChronoCross]

A little high profile clip encoded with the latest pre-beta build:

ftp://mood.ateme.com/beta/cinderella-ateme-3000k-hp.mp4

with 8x8 transform, custom scale matrix (flat) and 4 slices (encoded on a bi-xeon).

Enjoy :)

too many people seem to be requesting this file and/or other files.....any chance of getting the FTP to accept more than 10 connections total?

[R] 530 Sorry, the maximum number of allowed clients (10) are already connected.
[R] Connection failed

[plonk420]

patience, young padawan ;)

[eb]

A little high profile clip encoded with the latest pre-beta build:

ftp://mood.ateme.com/beta/cinderella-ateme-3000k-hp.mp4

with 8x8 transform, custom scale matrix (flat) and 4 slices (encoded on a bi-xeon).

Enjoy :)
It could be interesting comparison with CinderellaMan_HD trailer from DivX, resolution 1280x720, (but with mp3 128kb/s audio)

eb

[LigH]

Mplayer and ffdshow can not yet play this file:
[h264 @ 00A3A468]custom scaling matrix not implemented
[h264 @ 00A3A468]non existing SPS referenced
Error while decoding frame!
But this is not your fault - seems that your encoder already uses features which are not yet implemented in libavcodec.

It plays well with the Nero Video Decoder in MPC (through Haalis splitter).

[Sharktooth]

It could be interesting comparison with CinderellaMan_HD trailer from DivX, resolution 1280x720, (but with mp3 128kb/s audio)

eb
I think the source is the divx trailer.

[LigH]

Indeed, looks a bit "upscaled", somehow...

[hpn]

ftp://mood.ateme.com/beta/cinderella-ateme-3000k-hp.mp4

Pretty detailed and sharp-looking clip. I want to take some "full resolution" screenshots. How can I make Nero Showtime show the real 1280 pixels by cropping the extra 256 that do not fit in my 1024 LCD screen? I could force such HD playback with some players, but couldn't find a way to do it with Nero Showtime.

Here is the 4'000k DivX version of the same trailer so people could compare: http://trailers.divx.com/Universal/CinderellaMan_HD.zip
Is this the sourse or Bobololo has used some higher quality version?

[bobololo]

Dear testers,

I've just sent to you all the instructions email to retrieve the beta package. You should receive it shortly. Don't forget to use your personal login and password (sent by postal mail) when prompted.

The new beta round officially starts now ! Gentlemen, please start your CPU ;)

-- bobo

[bobololo]

Here is the 4'000k DivX version of the same trailer so people could compare: http://trailers.divx.com/Universal/CinderellaMan_HD.zip
Is this the sourse or Bobololo has used some higher quality version?

I confirm, this clip was encoded using DivX's one as source.

[LigH]

(Pushing my thumbs that it will work on my old soap-box...)

Received, downloaded, starting...
__

First problem:

It was a little hard to get the H.264 decoder filter used for playback. Although I disabled ffdshow's decoder for H.264, Media Player Classic 6.4.8.4 reported that the Ateme Splitter 1.2.5.2 is used, but as decoder it reports the "AVI Decompressor". Reason: ffdshow's VfW interface was still enabled for H.264. After disabling, the previous Nero decoder filter still had a higher merit (0x00600000) than the new Ateme H.264 decoder (0x005fffff). So I used GSpot 2.52 to raise its merit to "preferred" (0x00800000).

[acidsex]

I cant download it yet because I have not received my login information in the mail yet ( I am in the USA). When were they sent out and how long does it take to get international mail?

Bobolo, is there any way to get my login info or a temp login until mine arrive in the mail?

edit: I did receive my email with the url link just not the postal login information.

[bobololo]

(Pushing my thumbs that it will work on my old soap-box...)

Received, downloaded, starting...
__

First problem:

It was a little hard to get the H.264 decoder filter used for playback. Although I disabled ffdshow's decoder for H.264, Media Player Classic 6.4.8.4 reported that the Ateme Splitter 1.2.5.2 is used, but as decoder it reports the "AVI Decompressor". Reason: ffdshow's VfW interface was still enabled for H.264. After disabling, the previous Nero decoder filter still had a higher merit (0x00600000) than the new Ateme H.264 decoder (0x005fffff). So I used GSpot 2.52 to raise its merit to "preferred" (0x00800000).

I don't know which player you're using, but mpc has a very convenient feature for dealing with such issues : in the filters override options, put can ateme's filters to the top and set them as prefered. If this isn't enough you can ultimately block nero and ffdshow filter.

[bobololo]

I cant download it yet because I have not received my login information in the mail yet ( I am in the USA). When were they sent out and how long does it take to get international mail?

Bobolo, is there any way to get my login info or a temp login until mine arrive in the mail?

edit: I did receive my email with the url link just not the postal login information.

You should receive it very soon (probably tomorrow). Many of US residents as well as Australian testers just received it.

[acidsex]

Okay, Ill hold off until tomorrow. Ill let you know when/if it arrives and then we can go from there. :) Much thanks and tons of excitement to get started testing.

[Sagittaire]

encavc is "little more complex" than for first round beta test ... lol
Well we can have little more detail for this test ... please

- CBR rate control mode (HRD compliant)
HDR compliant?

- psychovisual improvement
find best psy mode test? find best adaptative deblocking matrix? find best custom scaling matrix?

- full interlaced support (field only, paff, mbaff)
paff? mbaff? interlacing with HDTV resolution?

- 8x8 transform
quality with/without 8x8?

- 4x4 and 8x8 custom scale matrix
find best custom scaling matrix?

- monochrome encoding


- lossless encoding
resolution or compatibility with other function (interlaced+lossless for example)?

- multi-threaded encoding support
Not for my Sempron 1.75@2.10 ...

- sse2 and sse3 instructions support
Not for my Sempron 1.75@2.10 ...

- quality and performance improvement


- source denoiser / film grain preservation
Film grain is Post-Process artificial grain? Find best setting for fgboost?

http://jfl1974.free.fr/forum/smileys/F9.gif ... more details ... please

[babayaga]

encavc is "little more complex" than for first round beta test ... lol
Well we can have little more detail for this test ... please

- CBR rate control mode (HRD compliant)
HDR compliant?
...
- full interlaced support (field only, paff, mbaff)
paff? mbaff? interlacing with HDTV resolution?
http://jfl1974.free.fr/forum/smileys/F9.gif ... more details ... please

The encoder is not designed only for movie backups. It is a full featured encoder that can be used for various applications. That's why for instance it can output a TS file that can be fed to a standard set-top-box.

Just for 2 points out of those many features:

The (CBR) Constant BitRate mode is compliant with what is expected out of an encoder for pro streaming applications (Digital Terrestrial TV for instance). This rate-control mode could be surprising to some since the quality is much lower for the same overall size than the 2 pass mode for instance. But on the other hand, the output bitrate is strictly constant over a buffer (called the CPB or Coded Picture Buffer set by default at 224kB). This means that the generated stream can be broadcasted over a fixed bitrate link without risks of overflow or underflow of the receive buffer located on the decoder's side.
It is HRD (Hypothetical Reference Decoder) compliant which means that it produces the kind of streams a decoder shall expect regarding the bitrate variations over time. Furthermore, the state of the decoding buffer the encoder expects is signalled in the stream. This enables low buffering delays or fast channel switching.


AVC/H.264 proposes several encoding modes for interlaced contents :
fixed field encoding: the video is assumed to be fully interlaced all the times
PAFF (Picture Adpative Frame Field): the video contains progressive and interlaced frames. The encoder chooses the best mode for each frame
MBAFF (Macroblock Adaptive Frame Field): the video contains progressive and interlaced frames but the interlaced or progressive parts can be limited to several macroblocks in each frame. This is the most efficient way of dealing with interlaced content. As with PAFF, the encoder chooses the best mode, but this time at the macroblock level
Regarding the resolution, there is a limitation but its very high: at least 36864 macroblocks per frame (for instance 4096x2304)

[bobololo]

@Sagittaire: you anticipated my test guideline post ;) Actually, for you particularly, you should test this :

-rcmode 2pass -br xxx -qual full -ref 8 -bref 4 -enhchrp -deblock 0 -setef xf8x8,csm -psy 3

If you're patient enough, that should give you the highest ssim score you've ever seen ;)

@Testers: here are a few hints helping you to getting started with this test.

First it would be interesting to check the stability of the encoder, ie. does it crash under some conditions ? (clips, hardware config, tools used, etc).

Then we have added many encoding tools : interlaced support, 8x8 transform, custom scale matrix, etc. Does those tools bring good improvement ? Are there some cases where they are inefficient or perform poorly ? Which matrix are optimal for different type of contents ?

New features were added : CBR rate control mode, lossless & monochrome encoding. Do those new features working fine and make you happy with them ?

Others were improved : how do you find our new psy3 model ? And what about the performance ? Does it run fast enough compared to other codecs ? Check out multi-thread support for HT/dual core CPU owners and evaluate the speed boost it gives !

We have also new stuff : the film grain modeling. Even it's still very preliminary, we wanted to show the abilities to filter out some film grain making the clip easier to encode and restituting this grain on the decoder side. It's very slow and under huge development but your feedback are welcome to help us improving it.

Finally and not the least, what about the overall quality, which settings give you the best results according your eyes :)

-- bobo

[Razorblade2000]

I've got some problems getting the decoder filters to work...
If I change in Register.bat:

from
FOR %%x IN (*.dll *.ax) DO regsvr32 /s "%%x"
to
FOR %%x IN (*.dll *.ax) DO regsvr32 "%%x"


he displays an error when executing the batch file...
"LoadLibrary ("Ateme H264 Decoder.dll") fehlgeschlagen - Das angegebene Modul wurde nicht gefunden"

--> LoadLibrary ("Ateme H264 Decoder.dll") failed - The chosen module wasn't found

same for the parser


maybe I forgot something... just reinstalled XP on my Laptop and it's time to go to bed now.. zZzZzZzzzZZZzz

[bobololo]

--> LoadLibrary ("Ateme H264 Decoder.dll") failed - The chosen module wasn't found

It's certainly due to a dll dependency problem. Can you check with depends.exe if all required dll are installed on your system ?

you can find it here : http://www.dependencywalker.com/

[Zarxrax]

I just made an encode with only the -lossless option... and my output certainly does not look lossless. The output looks ok on some frames but is blocky as hell on other frames. I did not experience anything like this on encodes where I did not use the -lossless parameter.

Edit: Upon further inspection, it seems the old nero decoder was being used... if i set mpc to block that decoder though, there is nothing to decode these files. It seems the ateme decoder isnt being registered, although the ateme splitter is working just fine.

[Tommy Carrot]

You can use mplayer to play it back too, it can decode everything if you don't use custom matrices.

BTW, i'm a bit disappointed in the lossless mode, so far i couldn't get it close to the efficiency of snow and msu codecs. I'm still experimenting with the settings to get the best result, so far it seems to me that disabling b-frames can help a bit.

[Zarxrax]

Well mplayer isnt playing it back properly either. (edit: using MPlayer-mingw32-dev-CVS-050514.zip) Lots of blocking artifacts there too. It makes me wonder if its an error in the encode itself, because the compression on it looks too good to be true. Its 71mb, compressed from a 120mb yv12 lagarith file.

[Tommy Carrot]

Use the latest build (http://www.aziendeassociate.it/cd.asp?dir=/mplayer) from mplayer.

[Zarxrax]

I'm still getting blocks with the build you linked.

[CyberGuy]

This is slightly off topic, but after I registered and verified with GSpot that I’m using the beta Ateme DirectShow filter, Ateme H264 Decoder.dll, I can't play my Reference Encoder clips. I re-registered the previous Nero filter, NeVideo.ax, and it plays correctly, except for the CABAC Initialization Table value error which causes it to freeze after about five seconds. The beta Ateme DirectShow filter does not seem to play a JM 9.6 clip with the correct or incorrect CABAC Initialization Table values. You can download the clip that I’m testing here:

http://rapidshare.de/files/2412410/HPII.MP4.html

BTW, the newest NeVideo.ax, 06/09/2005, Nero decoder still does not seem to have the correct CABAC Initialization Table values.

[Tommy Carrot]

I'm still getting blocks with the build you linked.
Well, it works flawlessly with the videos i encoded...

[Zarxrax]

Well, like i said, i think it might be an error in the encoding itself. I'll be able to verify once I get this darned ateme decoder to register @_@

[*.mp4 guy]

I just made a lossless encode from a huffyuv file, I got nearly a 50% reduction in size, and everything looks to be in order as far as blocks are concerned. The one problem I had is that the lossless encode looks slightly darker then the source file, its prolly just my overlay mixer, but I'm going to look into it.

[edit] Confirmed as overlay mixer discrepency. I'm going to make some other lossless encodes with snow et all to see how good the compression is.

[jfehr]

Did a couple quick tests with lossless tonight:

Source was test.avi, 30,726 KB, DV interlaced, pretty shaky. Here's what I got with a few different lossless settings:

-qual fastest -setef interlaced -lossless:
00:34, 32,739 KB

-qual full -setef interlaced -lossless:
00:37, 30,956 KB

-qual full -setef interlaced -lossless -rcmode 2pass
01:12, 31,646 KB

-qual fastest -lossless:
00:46, 56,015 KB

-qual full -lossless:
00:50, 53,639 KB

-qual full -lossless -rcmode 2pass
01:38 54,728 KB


Since the quality should be the same (since its lossless) I'm surprised the 2pass encodes both increase the size of the output file.

Also, while it looks like the lossless encoder did ok with -qual full and interlaced, it seems the filter is actually deinterlacing? (When I view the video there is no interlacing, and it seems blocky, like occurs with some deinterlacers.)

[babayaga]

Did a couple quick tests with lossless tonight:
Source was test.avi, 30,726 KB, DV interlaced, pretty shaky. Here's what I got with a few different lossless settings:
....
-qual full -setef interlaced -lossless -rcmode 2pass
01:12, 31,646 KB
...
Since the quality should be the same (since its lossless) I'm surprised the 2pass encodes both increase the size of the output file.

Also, while it looks like the lossless encoder did ok with -qual full and interlaced, it seems the filter is actually deinterlacing? (When I view the video there is no interlacing, and it seems blocky, like occurs with some deinterlacers.)
We did not think one second at using lossless in 2 pass mode since to achieve lossless, the quantiser is fixed at 0 (lossless is in fact VBR at q=0) :o There might eventually be a bug in the application that should not authorize this.

Regarding the aspect of interlaced content, it looks like an issue with the video renderer that de-interlaces (badly). The decoder filter does not de-interlace the video, but signals to the renderer that the content is indeed interlaced. It's then up to the renderer to act correctly.
There is an option in the decoder filter to force signalling the interlace content for the MBAFF case.

[Sharktooth]

Just downloaded. I'll start testing tomorrow.

[jfehr]

I didn't think it would make a difference either, but that's what beta testing is about, right?

I put up the video in question at http://www.altbits.com/test.avi if you want a look. This was my avs script (that resulted in incorrect looking lossless video):

# start of script
AviSource("d:\test.avi")
ConvertToYV12()
# end of script

and this was my command line:
encavc.exe -i d:\test_avi.avs -o d:\test.mp4 -qual fastest -setef interlaced

Let me know if you need anything else.


We did not think one second at using lossless in 2 pass mode since to achieve lossless, the quantiser is fixed at 0 (lossless is in fact VBR at q=0) :o There might eventually be a bug in the application that should not authorize this.

Regarding the aspect of interlaced content, it looks like an issue with the video renderer that de-interlaces (badly). The decoder filter does not de-interlace the video, but signals to the renderer that the content is indeed interlaced. It's then up to the renderer to act correctly.
There is an option in the decoder filter to force signalling the interlace content for the MBAFF case.

[*.mp4 guy]

FFV1 compressed the same sample from my previous post 24% better then Ateme avc HP and was 5 times faster. So I suppose that indicates avc probably wont be able to beat codecs specifically designed to be lossless. More complex sources might change things a bit.

[CyberGuy]

@Sagittaire: you anticipated my test guideline post ;) Actually, for you particularly, you should test this :

-rcmode 2pass -br xxx -qual full -ref 8 -bref 4 -enhchrp -deblock 0 -setef xf8x8,csm -psy 3

If you're patient enough, that should give you the highest ssim score you've ever seen ;)

@Testers: here are a few hints helping you to getting started with this test.

First it would be interesting to check the stability of the encoder, ie. does it crash under some conditions ? (clips, hardware config, tools used, etc).

Then we have added many encoding tools : interlaced support, 8x8 transform, custom scale matrix, etc. Does those tools bring good improvement ? Are there some cases where they are inefficient or perform poorly ? Which matrix are optimal for different type of contents ?

New features were added : CBR rate control mode, lossless & monochrome encoding. Do those new features working fine and make you happy with them ?

Others were improved : how do you find our new psy3 model ? And what about the performance ? Does it run fast enough compared to other codecs ? Check out multi-thread support for HT/dual core CPU owners and evaluate the speed boost it gives !

We have also new stuff : the film grain modeling. Even it's still very preliminary, we wanted to show the abilities to filter out some film grain making the clip easier to encode and restituting this grain on the decoder side. It's very slow and under huge development but your feedback are welcome to help us improving it.

Finally and not the least, what about the overall quality, which settings give you the best results according your eyes :)

-- bobo
These settings only gave me a SSIM 1 value of 69.43 compared to the 71.68 achieved from hp2-ateme-450k-hp.mp4 in post http://forum.doom9.org/showthread.php?p=659098#post659098. What settings did you use there? Reply should probably be posed in http://forum.doom9.org/showthread.php?t=95890.

[Razorblade2000]

It's certainly due to a dll dependency problem. Can you check with depends.exe if all required dll are installed on your system ?

you can find it here : http://www.dependencywalker.com/


You were right... The laptop I'm testing stuff on was a pretty "clean" Installation of WinXP --> msvcr71.dll, MPR.dll, msvcp71.dll and efsadu.dll were missing

[Manao]

Guys, don't forget that there's a quality feedback thread, and that this one is dedicated to bugs & issues.

[Selur]

seems like thread support is not working atm,..
(on WinXPpro, 512MB, Dual Athlon MP 1800+)
I set the thread option to two, but the text during 1st pass only mentions 1 thread and CPU usage also suggests that only one thread is used.
(around 70-80%, x264 thread support uses 88-98%)

Cu Selur

[Manao]

Selur : what's your commandline ?

[Selur]

encavc.exe -i "atemeTest.avs" -o atemeTest.mp4 -qual full -rcmode 2pass -br 2553000 -psy 3 -deblock -2 -adaptdbk -maxb 3 -enhchrp -thread 2 -setef ipred,ppred,bpred,wpred,cabac,deblock,part,hpel,qpel,xf8x8

Cu Selur

ps.: CPU usage in second pass drops to 50% :(

[bobololo]

Well, like i said, i think it might be an error in the encoding itself. I'll be able to verify once I get this darned ateme decoder to register @_@

Maybe you've the same issue as RazorBlade2000 ? Can you check your dll dependencies ?