Log in

View Full Version : Ultra low bitrate encodes using free encoders


manolito
17th December 2004, 15:26
During the last two weeks I did a couple of test encodes using QuEnc 0.56 Alpha and FreeENC 0.31.

I have seen reports that QuEnc 0.56 freezes sometimes when used with DVD Rebuilder, but with DVD2SVCD it never gave me any problems.

I normally use D2S with CCE 2.67 Trial, and frankly I did not expect that a free encoder would even come close in quality, but man was I wrong!

After some "normal" DVD to SVCD conversions with an average bitrate of around 1400 kbps looked just as good with QuEnc as with CCE, I decided to take my tests to the extreme. Here is what I did:

I used the movie "13 Days". This movie is almost 2 and a half hours long, 1.85:1 AR, and it has a very bad compressability. With CCE it is impossible to squeeze it on 1 CD with acceptable quality. The video bitrate for 1 CD is 707 kbps. Even with all settings tweaked for low Q, strong filtering and CVD target frame size I get a Q of around 60, and the OPV encoding result is painful to watch. Very strong mosquito noise (even at Image Quality Priority setting of 16), and also a lot of blocking. I repeated the encode in 2 pass VBR mode, but the quality did not improve significantly (if at all).
(My CCE settings were: DC precision 8, GOP 15, QC 16 (also tried 28), Jawor 1CD matrix)

Next I used QuEnc and FreeENC with the same filters. For FreeENC I used the "Incredible_KDVD" template because the default template had a tendency to undersize. The other settings were: 2 pass VBR, High quality, Trellis off, Scene detection on, GOP 15, DC precision 8, Extreme settings off, Jawor 1CD matrix.

Both encodes looked a lot better than CCE. No mosquitos at all, no visible blocks, just not very sharp due to the strong filters and the 352 horizontal resolution. FreeENC consistently was a little faster than QuEnc.

Since I really wanted to see what kind of artefacts the two free encoders produce, I did another encode. This time I used the full SVCD frame size at 480 x 576, no filters at all, Bach1 matrix. Just for fun I let D2SRoBa determine the resulting Q for CCE, it came out at 123.

Both QuEnc and FreeENC did produce "watchable" encodes. Still almost no mosquito noise, the only annoying flaws were changing pixels in grey background surfaces. Overall FreeENC seemed to have a little less artefacts, and since it also is a little faster, I prefer it over the current QuENC 0.56 Alpha.

Conclusion:
For those real ultra low bitrate encodes where CCE throws in the towel, both QuEnc and FreeENC deliver superior quality. Before reducing the target frame size or filtering the source to death, it is definitely worth to give those two a try. I would have liked to also test Peter Cheat's NuEnc, but it has a different commandline interface and is not compatible with DVD2SVCD.

One drawback for libavcodec based encoders is that there is no quality based OPV mode, and some posts suggest that implementing it will be very difficult because the codec has severe issues with one pass encodes. But even as it is now, I do not see much reason for anyone to spend big bucks on CCE SP any more. You can get real high quality encodes using only freeware, and who would have believed that this was possible only a short time ago....

Cheers
manolito

Inc
17th December 2004, 17:03
One drawback for libavcodec based encoders is that there is no quality based OPV mode, and some posts suggest that implementing it will be very difficult because the codec has severe issues with one pass encodes.

Did you had a look at PeterCheats NuEnc? And its experimantal OPV Mode Qaulity based encoding capabilities? ;)

manolito
17th December 2004, 18:00
Originally posted by incredible

In libavcodec its not possible to do a prediction like when using CQmatic. I know u used the slicer() function in DRPenc appl. to do a prediction based 1pass VBR but many and many tests on libavcodec based binaries did show that a "known" prediction is not possible. You cant set minrate and 2pass is obligatory if results should look good. The quantizer curve at 1pass behaves crazy and will only be set optimal finally at the 2nd pass.

This old post from you seems to indicate that this is not possible. Do you think Peter Cheat fixed it?

I could not test NuEnc with DVD2SVCD because a message "Bad Parameter" popped up and NuEnc did not start.

Cheers
manolito