DiAVC decoder, yet another fast H.264 high profile decoder - Page 17

Keiyakusha · 21st December 2009, 02:27

This comparison by yesgrey3 looks interesting. Does DiAVC has (S)SSE3 optimizations or they still planned for beta?

schweinsz · 21st December 2009, 17:49

Quote:

Originally Posted by Keiyakusha

This comparison by yesgrey3 looks interesting. Does DiAVC has (S)SSE3 optimizations or they still planned for beta?

All ssse3 code were finished last month. It include luma and chroma motion compensation and buffer copying. It is never included in current version but will be included in the upcoming beta version.

schweinsz · 21st December 2009, 17:53

Quote:

Originally Posted by vwpassion

Does DiAVC support out of spec motion vectors? Am looking for an alternative to CoreAVC since it produces glitches with out of spec MV's. FFDshow works fine, though it's not nearly as fast as CoreAVC or DiAVC.

Forgive me if this question has been answered already.

Yes, it is supported.

the_corona · 21st December 2009, 18:00

Just wanted to say great job so far!

I have tested it for the first time just now on my HTPC and for almost all samples I tested it's usually the fastest, if only by a few percent really.

Only thing a bit annoying is slighly longer initialization time when opening a video.

Can't wait for the beta which I guess will be even faster? Can you estimate how much?

Also, I understand DiAVC will be payware, is that correct? I ask because on your homepage there is a link to the source.

Thnx

LoRd_MuldeR · 21st December 2009, 18:05

There only are the sources of "DiAVCSettings.exe", a small configuration utility. They were published, because some AntiVirus programs triggered FALSE ALARMS with that file

I don't think that the sources of DiAVC itself will be published. And the "BUY" button on their web-site clearly indicates that DiAVC is supposed to be payware...

schweinsz · 21st December 2009, 18:15

Quote:

Originally Posted by the_corona

Can't wait for the beta which I guess will be even faster? Can you estimate how much?

Also, I understand DiAVC will be payware, is that correct? I ask because on your homepage there is a link to the source.
Thnx

The beta version is definitely faster than the alpha. But the gain depends on the cpu. The cpu with ssse3 will be more positive.
The DiAVC will be a payware. But it will be cheap and all following improvements such as super-resolution, 64-bit and new features are free for customers.

the_corona · 21st December 2009, 18:23

I see, thnx for that.

Luckily my Core Duo (not Core 2 Duo) in the HTPC seems to support SSE3, so that's good.

SuperResolution you say? So...Fancy Resizing? Stuff like FFDShow has?

Looking forward to the Beta...if it's about the price of other decoders it's fine with me :-)

LoRd_MuldeR · 21st December 2009, 18:32

Usually "SuperResolution" means that instead of one expensive "high resolution" camera we use several cheap "low resolution" cameras. Then we combine several "low resolution" shots to one "high resolution" picture by using a fancy SuperResolution algorithm. The result will (hopefully) be much better than just upscaling a single "low resolution" shot. However in the case of DiAVC I think the "SuperResolution" simply refers to a fancy upscaling method, because we only have one single "shot" of each frame. Maybe something like Edge-Directed Interpolation or tritical's NNEDI filter.

schweinsz · 21st December 2009, 18:34

Quote:

Originally Posted by the_corona

I see, thnx for that.

Luckily my Core Duo (not Core 2 Duo) in the HTPC seems to support SSE3, so that's good.

SuperResolution you say? So...Fancy Resizing? Stuff like FFDShow has?

Looking forward to the Beta...if it's about the price of other decoders it's fine with me :-)

It is SSSE3 not SSE3. Core Duo not supports SSSE3. SuperResolution is enlarging the current picture to display and motion-compensation-based SuperResolution can improve the picture quality greatly.
I just google it and get the following link, there is some pictures at the end of that paper.
http://graphicon.ru/oldgr/en/publica...d%20Fusion.pdf

LoRd_MuldeR · 21st December 2009, 18:42

Quote:

Originally Posted by schweinsz

I just google it and get the following link, there is some pictures at the end of that paper.
http://graphicon.ru/oldgr/en/publica...d%20Fusion.pdf

Ah, interesting. So in this scenario several consecutive "low resolution" images are combined to a "high resolution" image by using Motion Compensation.

Do you have any sample clips on how this looks with "real life" footage and in motion?

schweinsz · 21st December 2009, 18:49

Quote:

Originally Posted by LoRd_MuldeR

Ah, interesting. So in this scenario several consecutive "low resolution" images are combined to a "high resolution" image by using Motion Compensation.

Do you have any sample clips on how this looks with "real life" footage and in motion?

No, I designed most of the algorithms last month but I have not implemented it.

canTsTop · 21st December 2009, 18:58

How about DiAVC decoder support for DGAVCDec?

the_corona · 21st December 2009, 19:09

Quote:

Originally Posted by schweinsz

It is SSSE3 not SSE3. Core Duo not supports SSSE3. SuperResolution is enlarging the current picture to display and motion-compensation-based SuperResolution can improve the picture quality greatly.
I just google it and get the following link, there is some pictures at the end of that paper.
http://graphicon.ru/oldgr/en/publica...d%20Fusion.pdf

Ah, sorry I misread you. In that case....too bad :-( Still, looking forward to it.

Quote:

Originally Posted by LoRd_MuldeR

Ah, interesting. So in this scenario several consecutive "low resolution" images are combined to a "high resolution" image by using Motion Compensation.

Do you have any sample clips on how this looks with "real life" footage and in motion?

That is highly interesting....but the pessimist in me thinks it wont work nearly as good as the samples for most cases and it'll be just too slow. I'm really hoping u prove otherwise of course! :-)

Edit: BTW, while super resultion might be useful, many people watch mainly HD stuff anyways and since their panel can't display anything higher it's perhaps not that useful. Now Motion-Compensated-Frame-Interpolation on the other hand.....

I know, I'm just dreaming....but I would def. pay a premium for that one :-)

Blue_MiSfit · 22nd December 2009, 02:41

@CanTsTop:

Quote:

How about DiAVC decoder support for DGAVCDec?

Erm.. I don't think so. DGAVCDec is neuron2's baby, and it uses libavcodec for H.264 decoding. I don't think he would want to use DiAVC, considering it's not open-source, and uses DirectShow.

You're a little confused

~MiSfit

LoRd_MuldeR · 22nd December 2009, 02:47

Well, it seems Neuron2 isn't very happy with libavcodec!

The latest version of DGAVCDec still used an old outdated version libavcodec (one that is broken with various samples), because all newer versions of libavcodec have frame-accuracy problems.

If I remember correctly, there even were plans to make DGAVCDec use CoreAVC, as soon as CoreAVC exposes a suitable API. But now it seems that this won't happen, as CoreAVC won't offer such an API.

Therefore CanTsTop's question about DiAVC Decoder and DGAVCDec isn't completely absurd

squid_80 · 22nd December 2009, 03:05

Quote:

Originally Posted by LoRd_MuldeR

If I remember correctly, there even were plans to make DGAVCDec use CoreAVC, as soon as CoreAVC exposes a suitable API. But now it seems that this won't happen, as CoreAVC won't offer such an API.

It won't be offered to a closed source project that violates the GPL, no.

Astrophizz · 22nd December 2009, 07:32

squid_80, no need to troll and dig up an old (and resolved) issue. It's OT anyways.

schweinsz · 22nd December 2009, 07:41

Quote:

Originally Posted by Cyber-Mav

this is what people may be interested in the most, results done on my laptops pentium M740 cpu which is a single core 1.73ghz cpu that supports up to sse2. windows 7 professional 32bit 2gig ram.

The possible reason that DiAVC is slow for single-core cpu is due to the use of multi-threading. The current design of DiAVC simplily use threadNum=CoreNum+2, It hurts performance of single-core cpu probabily. I believe that the old cpu has a bad support for the multi-threading. As threadNum=CoreNum+2, I simplily set it and never test if is most efficient.
Besides, the DiAVC never uses any mmx code, it only uses sse2 code. as Dark Shikari said, it hurts performance for old cpu.

22nd December 2009, 08:13

Quote:

Originally Posted by squid_80

It won't be offered to a closed source project that violates the GPL, no.

Sales numbers of CoreAVC must be really bad.

I for one welcome competition. It benefits the consumer. Period.

Dark Shikari · 22nd December 2009, 08:17

Quote:

Originally Posted by honai

Sales numbers of CoreAVC must be really bad.

I for one welcome competition. It benefits the consumer. Period.

How is DGAVCDec competition to CoreAVC?

They don't even do the same thing.

21st December 2009, 18:05	#325 \| Link
LoRd_MuldeR Software Developer Join Date: Jun 2005 Location: Last House on Slunk Street Posts: 13,248	There only are the sources of "DiAVCSettings.exe", a small configuration utility. They were published, because some AntiVirus programs triggered FALSE ALARMS with that file I don't think that the sources of DiAVC itself will be published. And the "BUY" button on their web-site clearly indicates that DiAVC is supposed to be payware... __________________ Go to https://standforukraine.com/ to find legitimate Ukrainian Charities 🇺🇦✊

21st December 2009, 18:32	#328 \| Link
LoRd_MuldeR Software Developer Join Date: Jun 2005 Location: Last House on Slunk Street Posts: 13,248	Usually "SuperResolution" means that instead of one expensive "high resolution" camera we use several cheap "low resolution" cameras. Then we combine several "low resolution" shots to one "high resolution" picture by using a fancy SuperResolution algorithm. The result will (hopefully) be much better than just upscaling a single "low resolution" shot. However in the case of DiAVC I think the "SuperResolution" simply refers to a fancy upscaling method, because we only have one single "shot" of each frame. Maybe something like Edge-Directed Interpolation or tritical's NNEDI filter. __________________ Go to https://standforukraine.com/ to find legitimate Ukrainian Charities 🇺🇦✊ Last edited by LoRd_MuldeR; 21st December 2009 at 18:37.

22nd December 2009, 02:47	#335 \| Link
LoRd_MuldeR Software Developer Join Date: Jun 2005 Location: Last House on Slunk Street Posts: 13,248	Well, it seems Neuron2 isn't very happy with libavcodec! The latest version of DGAVCDec still used an old outdated version libavcodec (one that is broken with various samples), because all newer versions of libavcodec have frame-accuracy problems. If I remember correctly, there even were plans to make DGAVCDec use CoreAVC, as soon as CoreAVC exposes a suitable API. But now it seems that this won't happen, as CoreAVC won't offer such an API. Therefore CanTsTop's question about DiAVC Decoder and DGAVCDec isn't completely absurd __________________ Go to https://standforukraine.com/ to find legitimate Ukrainian Charities 🇺🇦✊ Last edited by LoRd_MuldeR; 22nd December 2009 at 02:51.

22nd December 2009, 07:32	#337 \| Link
Astrophizz Registered User Join Date: Jul 2008 Posts: 184	squid_80, no need to troll and dig up an old (and resolved) issue. It's OT anyways. Last edited by Astrophizz; 22nd December 2009 at 07:35.

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

21st December 2009, 02:27	#321 \| Link
Keiyakusha 契約者 Join Date: Jun 2008 Posts: 1,576	This comparison by yesgrey3 looks interesting. Does DiAVC has (S)SSE3 optimizations or they still planned for beta?

21st December 2009, 18:00	#324 \| Link
the_corona Registered User Join Date: Mar 2006 Posts: 55	Just wanted to say great job so far! I have tested it for the first time just now on my HTPC and for almost all samples I tested it's usually the fastest, if only by a few percent really. Only thing a bit annoying is slighly longer initialization time when opening a video. Can't wait for the beta which I guess will be even faster? Can you estimate how much? Also, I understand DiAVC will be payware, is that correct? I ask because on your homepage there is a link to the source. Thnx

21st December 2009, 18:23	#327 \| Link
the_corona Registered User Join Date: Mar 2006 Posts: 55	I see, thnx for that. Luckily my Core Duo (not Core 2 Duo) in the HTPC seems to support SSE3, so that's good. SuperResolution you say? So...Fancy Resizing? Stuff like FFDShow has? Looking forward to the Beta...if it's about the price of other decoders it's fine with me :-)

21st December 2009, 18:58	#332 \| Link
canTsTop Registered User Join Date: Apr 2007 Location: Lithuania Posts: 120	How about DiAVC decoder support for DGAVCDec?