mpegable AVC

[flower1209]

I just found mpegable's AVC codec. Does anybody made any experiences? More information on www.mpegable.com

Flower

[Nic]

Thanks for the tip,

It has a VFW decoder interface which means you can play/encode H264 AVI Files in VDUB

Also has VFW Encoder...I'll give it a try

-Nic

UPDATE:
Ok, Im quite impressed....Just did a 78second clip at 781kbps. Was encoding at 8fps on my 2.4Ghz P4 so slow and very slightly oversized. But the quality is pretty darned good seeing it's a 720x576 video.

@all: If you haven't tested this, give it a go. Oh and it doesn't seem to like fast recompress in VDUB, (it moans about frame size), so use Full Compression if you get that error.

[bond]

great news!!!

does the encoder include b-frames?
also is it limited somehow? 30days, number of encodes or so?
did anyone test the dshow decoder, whether its able to decode avc?

i give up, vfw will never die

[Nic]

1) Nope it doesn't use B-Frames (I checked )
2) Doesn't seem limited (ive put my computer into 2005, still decodes/encodes)
3) Strangely it doesn't seem to use a DirectShow filter to decode the video. It uses the VFW Wrapper. Which seems to work fine (no logo on screen or anything)

-Nic

[Doom9]

previous mpegable products had a logo if I remember correctly... and their bitrate control was horrible (and the company wasn't very cooperative.. when I asked for a full version without limits and restrictions they went into mute mode).

[Nic]

I know, and their codecs were cr*p anyway. Hence when I went to try this I expected the same heavy branding and bad quality. But I was quite surprised....I'm really surprised they've just given it away.

-Nic

[bond]

hm doesnt seem to show up in virtualdub/vfw codecs list

maybe it doesnt work in win98/me and only in xp and 2000? can anyone confirm this/the contrary?

[chilledoutuk]

if you change the fourcc from DAVC to H264 you can play it back with the ffdshow H264 playback filter.

I am testing it now I am impressed so far with the quality will report back with more findings later.

and maybe a sample clip.

[skal]

Hi,


JFYI, we (Bobololo and I) tested this codec yesterday
on the 'tempete' sequence, 2Mb/s. It's 2dB+ under
jm81a reference software, set up to use same features
(p-frame only, filtering, 16x16->8x8 modes, qp=25, cavlc).
More tests on referenced sequences would be great.
Someone?

Skal

[Nic]

@skal:
Why test with such a high bitrate on such a low-res file? The perceptible differences must have been miniscule....

I'll try some of the sequences from here and put results:
http://www.stewe.org/vceg.org/sequences.htm

-Nic

ps
And how did you're own code score on that test?

[skal]

@Nic:

well, feature-wise, mpegable codec is just a little
above mpeg4-v2, you know (the bonuses being: filtering
and 16x8 + 8x16 modes). 2Mb/s is what you then get from
jm81a with qp=25, which i expect to be the equivalent
of mpeg4's q=4 for a good rip. Don't forget we're
talking about constant-quant encoding here, not CBR (which
jm81 doesn't support very well, if at all).

Moreover, 'tempete' is a very difficult sequence, especially
if you don't use b-frames: it's a zoom-out with lot of
erratic/fast motion.

Now, this being said, it was only a first test. Given
the result, it should be better to try another working
zone. Something like 1Mb/s, QP=30 (mpeg4's q=10), PSNR ~= 34/35dB.

Skal

[Nic]

Thanks for the info I know what you mean about the reference source not supporting CBR. It doesn't seem to do it at all well.

I'll try some tests too. Thanks again

-Nic

[mpegable]

Hi,

we at dicas already made a lot of test, also against the reference software.
The results obviously depend on the encoding speed set in the codec's configuration dialog. For best quality we recommend to always use the slowest mode ;-).
How do you use the reference software? With CABAC, B-frames, ..?

For common bitrates, mpegable AVC (slowest mode) is usually around 0.5db below reference sofware (using Base profile tools) due to our tests, but looks better or equal than any of the exisiting MPEG-4 codecs at the same bitrate. As you know PSNR does not always correspond 1:1 to subjective visual quality. Especially, the in-loop filtering of AVC enhances the subjective quality compared to MPEG-1/2/4.

This version of mpegable AVC does not include a scene cut detection, pre-processing, ... There is still a lot of work - and potential.

I am already curious about your test results ..

bond: The installer of the current version only supports NT, 2000, XP but we will try to fix this asap.

Nic: When I checked the the fast recompress in Virtualdub it worked. Can you give me some info about the actual color format of your input?

Thank,

mpegable

[Nic]

@mpegable: Welcome to the forum You've got the start of a good and promising codec there. When I tried it on a PAL 720x576 Aviscript Script from MPEG2Dec3.dll I got an error about the input size being incorrect. The actual error was: "Output compressor error: The source image is not acceptable. (error code -2)". The same occurs with both an old and most recent version of VirtualDub.

When I check the "File Information" in VDub the Decompressor for the YV12 is listed as DivX 5.1.1 Codec. Maybe you have a different decompressor acting on yours?

Hope that's some use.

-Nic

[mpegable]

Thanks Nic, we'll check this.

[bond]

Originally posted by mpegable
bond: The installer of the current version only supports NT, 2000, XP but we will try to fix this asap.

thanks!

[bobololo]

Originally posted by mpegable

we at dicas already made a lot of test, also against the reference software.
The results obviously depend on the encoding speed set in the codec's configuration dialog. For best quality we recommend to always use the slowest mode ;-).
How do you use the reference software? With CABAC, B-frames, ..?

For common bitrates, mpegable AVC (slowest mode) is usually around 0.5db below reference sofware (using Base profile tools) due to our tests, but looks better or equal than any of the exisiting MPEG-4 codecs at the same bitrate. As you know PSNR does not always correspond 1:1 to subjective visual quality. Especially, the in-loop filtering of AVC enhances the subjective quality compared to MPEG-1/2/4.

Obviously we tested with the same tools and conditions to be fair (baseline tools : cavlc, p slice only, 16x16 -> 8x8 MB partitions, no scene-cut detection, same gop, etc). We used the slowest encoding mode with MPEGABLE. The results on the sequence we tried (tempete_cif) were the following :

MPEGABLE
PSNR: Y/U/V 35.232 35.527 37.317 (2034 kb/s) (target 2000 kb/s)

JM81a
PSNR: Y/U/V 37.28 38.83 40.25 (2160 kb/s) (Qp 25)

As you can see the difference is already huge with the same tools. When enabling all the main profile tools of the JM (cabac, bframe, multi-ref, etc) we had more than 3 dB of difference.

Anyway this test is not very significant, it was only on 1 single sequence and for one (PSNR,RATE) point. The complete R-D curve on several sequence would be more interesting.

btw: the inloop deblocking filter gets higher PSNR in most cases. till now I've never seen any sequences where the deblocking doesn't increase the PSNR. However it tends to smooth the picture in a way that sharpness lovers could consider as a subjective annoyance

Originally posted by mpegable


This version of mpegable AVC does not include a scene cut detection, pre-processing, ... There is still a lot of work - and potential.

I am already curious about your test results ..

Nic: When I checked the the fast recompress in Virtualdub it worked. Can you give me some info about the actual color format of your input?

I got the same problem here, the input was YV12.

-- bobololo

[Sagittaire]

As you know PSNR does not always correspond 1:1 to subjective visual quality

Perhaps with 0.3 dB but not with 3 dB.

[kilg0r3]

for me it only works with rgb input (full processing mode)

[mpegable]

Indeed, the version 0.7 only works with RGB input. An internal version already works with different YUV formats (YUY2, IYUV, YV12, UYVY). Does anybody has an idea what other formats might be important? thanks. Probably there will be a minor update next week.

We are still surprised about the large PSNR difference to JM but color conversion might be another cause - certainly not 2db. If somebody has more results with different sequences please let me know that we can compare it.