Alt CC. test

[LotionBoy]

SO I figure I'm going to queue up a huge Alt. CC test and see how things go with that. Here are the settings I am planning on testing. If you have comments or suggestions that would be much appreciated.
I've cut three clips from CTHD, one low motion, one mixed, and one high motion. Each are 4k-6k frames. I think I'll use a fairly low bitrate so that errors are noticeable (maybe something like 500 or 600). I'd appreciate suggestions, comments, etc. some time soon before I queue all this stuff up. :-)

LotionBoy

yeah, it would have helped if I told you want the test was, wouldn't it. :-) These are the tests for each clip. 225 total clips to be encoded.

Tests:
Low
50/100 Str: 20
100/100 Str: 20
50/200 Str: 20
100/200 Str: 20
200/200 Str: 20
100/300 Str: 20
200/300 Str: 20
300/300 Str: 20
50/100 Str: 35
100/100 Str: 35
50/200 Str: 35
100/200 Str: 35
200/200 Str: 35
100/300 Str: 35
200/300 Str: 35
300/300 Str: 35
50/100 Str: 50
100/100 Str: 50
50/200 Str: 50
100/200 Str: 50
200/200 Str: 50
100/300 Str: 50
200/300 Str: 50
300/300 Str: 50


Medium
50/100 Str: 20
100/100 Str: 20
50/200 Str: 20
100/200 Str: 20
200/200 Str: 20
100/300 Str: 20
200/300 Str: 20
300/300 Str: 20
50/100 Str: 35
100/100 Str: 35
50/200 Str: 35
100/200 Str: 35
200/200 Str: 35
100/300 Str: 35
200/300 Str: 35
300/300 Str: 35
50/100 Str: 50
100/100 Str: 50
50/200 Str: 50
100/200 Str: 50
200/200 Str: 50
100/300 Str: 50
200/300 Str: 50
300/300 Str: 50

High
50/100 Str: 20
100/100 Str: 20
50/200 Str: 20
100/200 Str: 20
200/200 Str: 20
100/300 Str: 20
200/300 Str: 20
300/300 Str: 20
50/100 Str: 35
100/100 Str: 35
50/200 Str: 35
100/200 Str: 35
200/200 Str: 35
100/300 Str: 35
200/300 Str: 35
300/300 Str: 35
50/100 Str: 50
100/100 Str: 50
50/200 Str: 50
100/200 Str: 50
200/200 Str: 50
100/300 Str: 50
200/300 Str: 50
300/300 Str: 50

[Foxer]

LOL That's a long queue.

Are you planning on posting the quant distributions and thresholds for each one? heheh :)

Seriously though.. That is quite a bit to test visually.

[LotionBoy]

agreed. It's probably too much to test visually. So, suggestions on cutting it down? Where the real differences should be found? And also suggestions on how to analyze the clips mathematically. I know there is a DRF analyzer for Divx3. I don't know if it work on Xvid. If it does, that would definetely help. If it doesn't, I might be able to scratch something together in C++. I wish the FairUse code for comparing different video images and generating a quality report was availible, because that would be extrememly useful. :-) Again, any and all suggestions and comments are welcomed.

LotionBoy

[rui]

I remember that Sierrafoxtrot made some tests with the alt curve comp, and he used debugview, in conjuction with some software Foxer made (correct?), to analyse all the logs created by the debugview software.

[Foxer]

Quantizer distribution printout at the end of 2nd pass was added and of course there's quantizer threshold printout at the beginning of the 2nd pass.

Quantizer distribution is nice to know but it can sometimes be a little misleading due to frame compressibility and the influence of the root overflow which is generated by the difference between desired framesize and produced framesize.

For example, if there's a large overflow when entering a completely still scene such as a black screen or something, it will generate a string of higher quantizers than the rate control system truely desires for that sequence until the end because only then would the higher quantization actually produce a smaller framesize.

heh I guess one needs a program that will take the stats file (for original framesizes) and use that with analyzing the decoded video with a quality measure like PSNR or better to generate a timeline of somewhat reliable info.