HCEnc: Hcbatch 016 Test Release
 

Hi all,

Well, that's been a while since I was here...
And surely I missed a lot of useful information :cry:

But, here's a new release, a test version of HCbatch016 only

link: http://hank315.dyndns.org/HCbatch_01...08-09-2005.zip

Changes:

- improved auto gop option
- fixed a bug in Motion Estimation which could cause blocky playback on some HW players
- for fixed GOP 15-1, GOP length sometimes was set to 16-1, fixed
- added bias and dc precision as parameters for HCbatch
- small GUI changes
- as usual some other bugs fixed

Recommended avisynth version: 2.5.6 RC1
And I still encountered some problems with DirectShowSource :evil:

Would you recommend setting bias to 0 or is there any reason for keeping it at 20-30 as it is in CCE?

I'll run a test encode with the new version on the troublesome The Great Dictator I bitched about at D9. HC_015 was able to encode the segment in which CCE got hiccups.

Normally I use values 0 - 25 for bias.
If a bias of 100 is used the birate tends to CBR but it is not really CBR, the bitrate prediction just tries to set it at CBR.

I also tested the bias settings with the PSNR/SSIM2 test sequence from Sagittaire, bias 0 was the best for metric values...

Re: HCbatch016 test release
 

Hi Hank :).
Avisynth 2.5.6 is still in Release Candidate status.
Any special reason why we we would need to use avs 2.5.6RC1 instead of good old 2.5.5 with the new HCBatch 0.16?
BTW do we need to specify a Bias at all, or can we avoid it?
Cheers

The last RC1 has a new DirectShowSource.dll, which seem to solve some problems but not all :cry:
If older versions work good for you there's no reason to switch to RC1.

If you don't specify a bias it will be set to zero.
A small bias may help to avoid blockyness in dark scenes.
Bias settings are not very agressive, 25 is still small.

Thanks for clearing up my doubts Hank ;-).
And thanks for sharing HC news with us :D
Cheers mate

Thanks for the update Hank of your excellent Mpeg2 encodeer, great work. :)

A new test release: http://hank315.dyndns.org/HCbatch_01...16-09-2005.zip

Most issues are solved now:
- hangs while running minimized
- wrong GOP size for some fixed GOPs
- artifacts at the right side of some frames

The artifacts were caused because motion vectors could point out of the frame by half a pixel, some decoders don't like that...

Still one issue left, using Avisynth DirectShowSource with the WMV decoder sometimes won't read the stream correct during the second pass, it can skip frames or even jump back.

Hi Hank :)
Thanks for the updates on HC :D.
I guess you're pretty close to have a final 0.16 final, right?
I always use MPEG2Source, so I've never came across the DirectShowSource bug.
Could you tell us what you have in mind for 0.17?
Cheers m8

Hi Rui,

For HC017, there are still alot of things I want to do:
- 1 pass VBR
- Trellis quantization
- HQ mode
- code cleaning and optimization
- better GUI
- etc

For the moment I'm working on 1 pass VBR and Trellis.

Hi Hank :D.
I see, you mean OPV prediction, right?
BTW, what do you mean by HQ mode?
High Quality? If so what can we expect from it?
Cheers

Hi,

Just to say that 016 version give me a better quality that good old 015 one: in the Mulan backup I'm doing, I get a lot less mosquitos (it's a 2000 average bitrate encoding, at 704x576).

Thanks again! :-)

Fabrice

Thanks for the test update Hank. :)