TOO_SMALL_LIMIT in latest dev builds

[AndyP]

Hi, sorry to disturb.

Can I just ask something. I am seeing some artifacts in Koepi's latest dev build (12/10), similar to the problems I had before. Is TOO_SMALL_LIMIT set to 1?? [I read in a previous thread that it was 2 in the CVS but -h was going to change it... I don't know if he had time in the end].

Kind Regards,
Andy

PS I am not complaining about artifacts otherwise, I know that they are experimental (but am greatly enjoying the oppertunity to play :D )

[Koepi]

Nope, I just replaced the modified files in my source tree, and /utils/mbtransquant[.h|.c] weren't touched, so it's still set to 1.

Setting it to higher values doesn't give artefacts, but a "softer", less detailed image.

Regards,
Koepi

[AndyP]

Many thanks for checking :)

Kind Regards,
Andy

[Nic]

@Andy:
Yup both me & Koepi set them to 1 (unless we forget ;) ). But it is still set to 2 in the CVS.

-Nic

[TheXung]

Why is everyone obsessed with the TOOSMALL_LIMIT being set to 1? I thought syskin tested it out and found 2 to give a better PSNR at a given bitrate. True, it gives a softer image but the saved bitrate would in theory get used in reducing noise. It is not like most of us are using quant 2 that much anyway.

[Tommy Carrot]

Originally posted by TheXung
Why is everyone obsessed with the TOOSMALL_LIMIT being set to 1? I thought syskin tested it out and found 2 to give a better PSNR at a given bitrate. True, it gives a softer image but the saved bitrate would in theory get used in reducing noise. It is not like most of us are using quant 2 that much anyway.

In my experiences TOOSMALL_LIMIT=1 gave better result at any given quantizer. The bitrate gain is not as great as the loss of detail.

[HarryM]

@Nic
@Koepi
@-h


And what compromise?
Can you make this TOO_SMALL_LIMIT user defined in vfw? I think, that this is good idea, right?

[Nic]

It is a good idea but it works best as a constant definition & adding to VFW means we have to set it using the XviD API (which we cant at present). I think its best we decide on a value (be it 1 or 2) & just leave it at that. When set to anything but 1 many complained about there encodes...

Cheers,
-Nic

[sysKin]

Originally posted by TheXung
Why is everyone obsessed with the TOOSMALL_LIMIT being set to 1? I thought syskin tested it out and found 2 to give a better PSNR at a given bitrate. True, it gives a softer image but the saved bitrate would in theory get used in reducing noise. It is not like most of us are using quant 2 that much anyway. If you asked me, I'd tell you that I changed my mind right after I found out that the image is less detailed. True, it does improve PSNR but PSNR is only good if you can't really tell which looks better.
In my humble opinion, sharper image looks better even if it has some noise.

[unplugged]

In my first impressions about encodes made with TOO_SMALL_LIMIT=2 I felt something like H.263 quantizer matrix in place of MPEG... or even a slight level under H.263 matrix.
Further, some comparisons with previous encode and the lower 1st-pass size has produced me more than a suspect.
Sorry, I'm pretty sensible to these things :D