Why does this sample look so bad

[renski]

I encoded this with Xvid 1.1, the source wasn't interlaced but yet it looks really mpegy/compressed on the edges when the guys move their heads.. I can't workout why. I tried encoding with Neutral Bicubic and Lanczos (Sharp) and they both looked as bad.

Field operations was set to "none".

source was PAL.

check here, http://users.on.net/~renski/sample2.avi

[sysKin]

There's a horribly horribly bad quantizer distribution in the clip. I highly recommend to delete 1st pass statistics file, load encoder's defaults, and encode again.
Something went very wrong, maybe overflow settings are screwey...

Just to be really really sure: you did not encode CBR did you.
sidenote: if you want generally better compression, crop black borders from top and bottom.

[Teegedeck]

Yes, it looks like CBR going horribly wrong; quantizers jumping from 1(!) to 8 to 12 etc. You've been using much too stong compresson. Two-pass at a reasonable aimed-at filesize would seem a good idea.

[renski]

yah i had my overflow treatment set to 20 20 20.. gonna put it back to 5 5 5

ps. it was cropped.. the picture is WS (source was 4:3 with the bars included so WS).

[sysKin]

yah i had my overflow treatment set to 20 20 20.. gonna put it back to 5 5 5

ps. it was cropped.. the picture is WS (source was 4:3 with the bars included so WS).

Put it to 5/5/5 (or default, 0/0/0, which is also 5/5/5 I believe) and change quantizer restrictions to 2-31 rather than 1-31. Even with good overflow this can happen if quant 1 is allowed, because ratecontrol is exceptionally bad at quants 1...2.

Still, are you sure you didn't mess up 1st pass statistics file? Teegedeck said it reaches 12 but in reality I even saw it reaching 29 (that was a bframe I admit), this just shouldn't be possible with only four times larger overflow...

You are right, picture was cropped. MPC was trying to fool me by adding some extra black orders (1-2 pixels wide) but vdub makes it clear that it's ok.