New XviD settings - test encodes

[keenan]

I've prepared a bunch of test samples using the alpha XviD build. Settings are targeted to toward low bitrates, if anyone's interested in having a look at some shots and the settings I used: http://www.gotroot.ca/xvid/

I'd like some input on what I can do to improve quality, as well as your opinions on the different new options (quarterpel, b-frames, chroma motion).

There are 10 sample clips, 3 screenshots from each. The page contains about 3.5MB of screenshots, so be patient, and try not to kill my DSL line ;)

If anyone wants 1min samples, feel free to e-mail me and we'll arrange something.

[iago]

@keenan

Nice work! Actually, I was planning to do something similar but I just couldn't find the time.

I have a few humble suggestions for your future comparisons:
* You can use Linear Curve Scaling for XviD (Alt. Curve disabled, High/Low: 0/0 in the two pass tab) which I believe would give better results.
* You can use higher resolution (640*...)
* You can use GKnot defaults for Nandub SBC encodes.
* You can consider adding Divx502 with only b-frames to the comparison.
* You can use 3/150/100 for XviD b-frame encodes (going higher might result in considerable quality loss).

regards,
iago

edit: YV12 to YUY2 conversion may not be necessary in your script(?).

edit2: You may consider using the most recent XviD dev. build for the next comparison.

[neo_sapien]

I disagree with your findings on Qpel. Rather than detail, it seems to be just adding noise indiscriminately. Why not turn on noise in ffdshow and get yourself a higher compressibility test?

[wotef]

i disagree (agree?) with that, actually!

qpel definitely retains detail - it's very easy to see on a clean frame, like on arwen's face when crossing the ford

generally, i'm not sure what you mean by "reference xvid" + for a low, low bitrate comparison, you'd definitely want to be comparing with b-frames + perhaps even turn qpel off and don't use MPEG quants

otherwise, it's an interesting read + i also look forward to the high bitrate comparisons

[Houdini]

I think the idea of making comparisons between different encoding options is good but I think it's quite possible to make better encodes with each option.

On each fastmoving screenshot you can clearly see small blocks. Is it because of the jpeg compression or is it in the xvid encode?

Thanks anyway for the testing,

Houdini

[keenan]

There is no jpeg compression. Lossless PNG was used on all the shots. My intent for the reference encodes was just as sort of a baseline to compare against, to see which options improve quality the most. Thanks for the suggestions, I'll do up some more encodes hopefully tonight and see what happens.

Oh and the YV12toYUY2 is there because I was playing with the new avisynth/mpeg2dec, but I didn't want to use vdubmod (go figure).