[MPlayer-users] Re: divx 6

[Corey Hickey]

Mathieu Monnier wrote:
>  - apart from Rich, which seems allergic to anything coming out of this 
> test, others seem to have acknowledge the results : vb_strategy=2 is 
> commited two day after the results, Corey is working on XviD's RC 
> integration. Nobody seems to criticize lavc raw coding efficiency, which 
> indeed seems a bit higher than XviD's.

Just to clarify a bit where credit is due:
Michael wrote the XviD ratecontrol wrapper. I tested, gave feedback, 
Michael fixed bugs, and I tested some more.


My eventual results were that:
1. xvidrc only works right when the first pass is vqscale=2
2. With vb_strategy=2, lavc doesn't use nearly as many b-frames (16% vs. 
55%) when the first pass is vqscale=2.
3. Putting the above together, I couldn't get xvidrc to have quite as 
high PSNR or look quite as good as lavc with my own preferred 
ratecontrol equation.

Having XviD's frame-type decision as well would probably improve the 
results for xvidrc.

-Corey