Optimal resolution with Xvid

[kastro68]

Contrary to what others keep telling me, I always encode at the highest possible resolution, constrained only by keeping either the horizontal or vertical zoom at 100%.

Is it wise to do this with Xvid? I get sharper scenes when I do this...but I read on some guide that i should pick the resolution so that Bits/(frame*pixel) is approx 0.2

I believe that by keeping the resolution high the clip does not have to be enlarged as much, thus resulting in more detail. I have always used full resolution with 4.12 because it tends to smooth the image alot.

Should i keep the resolution approx 0.2 or should i just use the max resolution when using the Xvid codec?
:confused:

[Teegedeck]

Originally posted by kastro68
Contrary to what others keep telling me, I always encode at the highest possible resolution, constrained only by keeping either the horizontal or vertical zoom at 100%.
I have sympathy for this, I also try to keep the res as high as possible. :)

Should i keep the resolution approx 0.2 or should i just use the max resolution when using the Xvid codec?
:confused:
There is no such thing as an optimal resolution, it only depends on what looks better to you. To some extent, it is true that a quantizer does automatically what we try to reach with lower resolutions manually: It gets rid of details in order to make it possible to shrink the frame to the chosen size . But beginning at, say, quantizer 7, it starts to look really ugly, independent of personal taste. So you should have an interest to reduce the resolution enough that this quantizer or higher ones aren't used too regularly (you can check this with debug view in second pass).

As it differs from movie to movie, the best solution is to do a compressibility check with GKnot before you do the whole movie. Use about 10% of the movie, I'd say (check out the GKnot-guide and GKnot's internal help if neccessary).

[rui]

Originally posted by Teegedeck
As it differs from movie to movie, the best solution is to do a compressibility check with GKnot before you do the whole movie. Use about 10% of the movie, I'd say (check out the GKnot-guide and GKnot's internal help if neccessary).

Currently i am using Gnot's divx4 comp check. The results aren't bad, but i would like to know other people opinions about if this is correct.
I choose divx4 comp check because XviD, like divx4, is 1k=1000b, but i don't now if this has any influence in the comp check.

[Teegedeck]

Didn't know either, till -h told me it's all about framesizes, not kbps. So it shouldn't matter.

[kastro68]

Teeg,

Are you implying that Xvid and Nandub vids have the same compressibility?

I am aware that using a lower resolution may act as a noise filter, but doesn't using a lower resolution also result in loss of detaiL?

[Teegedeck]

No, they don't have the same compressibility; a compressibility-test with Nandub-produced .stats-files and XviD-produced .stats-files should yield very dissimilar results.

>I am aware that using a lower resolution may act as a noise filter, but doesn't using a lower resolution also result in loss of detaiL?

Yes, it results in a loss of detail. The thing is, it might look better to have a horizontal resolution of 608 and quantizers ranging from 3 to 7 than having a resolution of 720 and quantizers that start at 5 and go up to 10. A low resolution makes for a loss of detail but it just looks unsharp, soft; very high quantizers don't result in a 'softened' picture but in a downright ugly one. I'd suggest you try encoding some minutes of a movie at various fixed quantizers and compare them to find out where it starts looking bad for you.

If you still want to stay with the highest resolution and use comparably low bitrates, you better stick with the H.263 quantizer as it generally softens the picture. Or try Koepi's new modulated quantizer -versions of XviD.