XviD quant assignment in 2-pass mode

[cjv]

I have been experimenting with manually filtering (temporal & spatial) heavy action shots, pans, noisy scenes, etc. individually...in order to take advantage of the fact that the eye cannot notice loss of detail as much..And also sharpening character close-ups.
In theory this appears sound, and in constand quant mode it looks great, but now I am concerned that in 2-pass mode, XviD will see that these high action scenes compress better than the sharpened close-up scenes..and assign a higher (uglier) quant to those sharp scenes.

Please correct me if I misunderstand:
normal movie:
large frame size variation->poor compression->Xvid assumes high action->assigns high quant (4-5)
small frame size variation->compress well->Xvid assumes close-up->assigns low quant (2-3)

My filtered/sharpened movie:
large frame size variation->poor compression->Xvid assumes high action->assigns high quant (4-5) --->in reality this is a sharpened scene
small frame size variation->compress well->Xvid assumes close-up->assigns low quant (2-3) --->in reality this is my filtered motion scene

...or am I thinking XviD is really smarter than it is :)
(I have tried SMALL 2-passes and it appears fine, but over the course of an entire movie, this might change)
Thanks,

cjv

[Asmodian]

I don't think Xvid changes anything based on frame size variation in two pass mode. It just tries to make the video fit in the desired file size by increasing the quants for big frames (not rapidly changing in size frames) more then small frames. This is with curve compression on, with curve compression off I believe Xvid doesn't bias small frames over big ones. So without curve compression the two pass mode acts like 1 pass quality mode with the quality tuned to give the desired file size (maybe what you want?). Someone please correct me if I am wrong.

[cjv]

Originally posted by Asmodian
I don't think Xvid changes anything based on frame size variation

Bad choice of wording, I meant "variation from average frame size"..should have said "deviation".
I always use a linearly scaled curve, so the newly scaled curve should reflect the original. I guess where I'm confused is..why would XviD decide to use..say quant 5 instead of quant 2 for any particular frame? Does it just depend on how well a particular frame compresses, or does it have anything to do with motion and/or frames before and after it?

Ha, maybe I'm reading too much into this and should stick to the idea that "any type of filtering that doesen't negatively effect the film and helps compressibility is always a bonus". :)