Which is the best lossless video compressor?

[Stears555]

Hello guys!

Which is the best lossless video compressor? Which lossless encoder has the highest compression ratio in yuv 4:2:0?

Requisites: 1 support for Adobe Premiere Pro, 2. Its compresson is higher than FFv1

Thank you for your reply!

[Cauptain]

Try UT VIDEO CODEC. Very nice codec in nowsday.




Claudio

[jagabo]

I believe MSU Lossless has the best compression. But it's very slow.
http://www.compression.ru/video/ls-codec/index_en.html

YULS claims to be even better but I've never used it.
http://www.yuvsoft.com/2d-technologies/lossless-video-codec/

[Stears555]

Yuls was the best lossless compressor, however it works only with SD VIDEOS.

The Yuls project was chaceled.

The 4:2:0 (YV12) MSU lossless codec doesn't work in virtualdub. In the reality MSU offers only 24bit/32bit RGB, which significantly increases the filesize!

[sanlyn]

Requisites: 1 support for Adobe Premiere Pro, 2. Its compresson is higher than FFv1

Good question, but I think you have it the wrong way around. It's not the compressor that supports the application, it's the application that's supposed to be able to recognize and use external compressors in your system.

[jagabo]

I didn't mention it since Cauptain already did, but UT is a good compromise of speed and compression.

[Stears555]

The newest version of UTI codec has not better compression ratio and speed than Huffyuv (with huffman tables). Moreover it is suprisingly fascinating because Huffyuv was last developed in 2002

[jagabo]

The newest version of UTI codec has not better compression ratio and speed than Huffyuv (with huffman tables). Moreover it is suprisingly fascinating because Huffyuv was last developed in 2002

Why would that be surprising? There have been no big breakthroughs in lossless compression techniques (of anything) since 2002.

And I disagree with your assertion that UT doesn't compress better than Huffyuv and that it isn't faster.

[Stears555]

The newest version of UTI codec has not better compression ratio and speed than Huffyuv (with huffman tables). Moreover it is suprisingly fascinating because Huffyuv was last developed in 2002

Why would that be surprising? There have been no big breakthroughs in lossless compression techniques (of anything) since 2002.

And I disagree with your assertion that UT doesn't compress better than Huffyuv and that it isn't faster.

I tried more than 6 video files with 10min length. When I enabled the "adaptive huffman tables" option in Huffyuv, the size of the videos veree the same as the videos which were compressed by UTI video codec. Both encoder USED the YUV 4:2:0 color spaces.

[El Heggunte]

I tried more than 6 video files with 10min length. When I enabled the "adaptive huffman tables" option in Huffyuv, the size of the videos veree the same as the videos which were compressed by UTI video codec. Both encoder USED the YUV 4:2:0 color spaces.

What version of HuffYUV you have? Mine is the "CCE-patched" one, and definitely it does not show any "adaptive Huffman tables" in the configuration applet.

Also, the UT codec let's you choose between "optimize for compression ratio" and "optimize for decoding speed".

[Stears555]

I tried more than 6 video files with 10min length. When I enabled the "adaptive huffman tables" option in Huffyuv, the size of the videos veree the same as the videos which were compressed by UTI video codec. Both encoder USED the YUV 4:2:0 color spaces.

What version of HuffYUV you have? Mine is the "CCE-patched" one, and definitely it does not show any "adaptive Huffman tables" in the configuration applet.

Also, the UT codec let's you choose between "optimize for compression ratio" and "optimize for decoding speed".

See the FFDSHOW encoder panel, which was posted in my previous message.

I Used the better comression option in UTI, but UTI couldn't produce lesser file-size than Hufyuv with the usage of "Adaptive Huffman Tablets" option. Also Huffyuv has shorter encoding time than UTI codec. The Huffyuv decoding used less CPU power, than the decoding with UTI compressor

[jagabo]

The picture shows he's using ffshow's Huffyuv encoder, a variant of Huffyuv.

[Stears555]

Hey guys!

Just do a fast compression comparison in the Adobe Premiere. Download the 64bit FFDSHOW , and the 64bit Uti codec. OK?


I'm waiting for your results....

[jagabo]

Keep in mind that what CPU you're using and whether the codec is mulithreaded will make a difference in speed too. Does ffdshow have a multithreaded huffyuv encoder?

[Stears555]

Keep in mind that what CPU you're using and whether the codec is mulithreaded will make a difference too. Does ffdshow have a multithreaded huffyuv encoder?

Of course : All encoders are 100000% multithreaded in the FFDSHOW project.

Did you do the fast UTI VS. Huffyuv comparison test with the Adobe Premiere?

[jagabo]

I don't have Premiere. I just ran some test encodes using ffdshow huffyuv YV12 with adaptive huffman tables, prediction left, and UT 4:2:0 with compression optimization, 4 threads. I used VirtualDub. The source was a 62854 frame Xvid AVI file, 640x328 frame size. I'm using a quad core i5 2500K CPU.

encoder, time, size
ffdshow, 92 sec, 7.65 GB
ut, 78 sec, 6.13 GB

So for me, with this one video, UT was both faster and gave better compression. Both only used about 35 percent of the CPU though.

Ah, I changed ffdshow to "prediction plane" and got better compression, 6.13 GB. No change in encoding speed though.

And "predict median" gave an even smaller file, 6.08 GB. Same encoding time.

I verified all the encoded videos were identical.

Just for kicks I tried Lagarith: 98 seconds, 5.75 GB.

And the original single threaded Huffyuv 2.1.1 (YUY2): 67 seconds, 8.83 GB. Keep in mind that this was developed as a fast compressor for YUY2 video capture in the days of single core CPUs.

Doesn't Premiere work in RGB?

[Stears555]

I don't have Premiere. I just ran some test encodes using ffdshow huffyuv YV12 with adaptive huffman tables, prediction left, and UT 4:2:0 with compression optimization, 4 threads. I used VirtualDub. The source was a 62854 frame Xvid AVI file, 640x328 frame size. I'm using a quad core i5 2500K CPU.

encoder, time, size
ffdshow, 92 sec, 7.65 GB
ut, 78 sec, 6.13 GB

So for me, with this one video, UT was both faster and gave better compression. Both only used about 35 percent of the CPU though.

Ah, I changed ffdshow to "prediction plane" and got better compression, 6.13 GB. No change in encoding speed though.

And "predict median" gave an even smaller file, 6.08 GB. Same encoding time.

I verified all the encoded videos were identical.

Just for kicks I tried Lagarith: 98 seconds, 5.75 GB.

Doesn't Premiere work in RGB?

No, premiere work in many color spaces.

It is a shame that a freshly released compressor like UTI is not better than an old Huffyuv encoder.

[jagabo]

It is a shame that a freshly released compressor like UTI is not better than an old Huffyuv encoder.

Again, why would there be? There have been no breakthroughs in lossless compression algorithms in many years.

[Brad]

The newest version of UTI codec has not better compression ratio and speed than Huffyuv (with huffman tables). Moreover it is suprisingly fascinating because Huffyuv was last developed in 2002

You aren't using the Huffyuv encoder that was last developed in 2002. ffdshow's addition of Adaptive Huffman Tables was recent and I believe that's the same premise on which Ut Video was based.

Doesn't Lagarith still give the best ratios? But decode times are the bummer.

[poisondeathray]

FFV1 gives better compression ratio than lagarith in YV12 mode

UT is the "best" in the sense that it is actually usable at HD resolutions in editors (fast decoding performance) with typical quad core computers (you don't need a dual socket workstation) .

FFV1 is too sluggish but may be suitable for archival purposes

[Brad]

Yes I see, FFV1 does a little better with my test clip and at a compression speed that would work for SD capturing. I've always been scared away by the idea of using ffdshow in VFW.

Decode speed is 1/3 of Lagarith, of course, and something like 1/10 of Ut Video. But FFV1 does utilize less of my CPU during decoding (?!) so I guess that would leave more CPU time for scripting and encoding.