what´s the optimal resolution?

[((( atom )))]

hi,

i recently encoded 8mm at a vertical resolution of of 624 with divx5. since this movie compresses very good i went for a 1cd-rip.

since i normally don´t go that high with my resolution on 1cd-rips(the comp-test was too good though, so i did this time), i wondered, what the movie would look like at a res of 608.

i reencoded it and compared the same frame of both movies in acdsee by letting acdsee scale the two images to the same sice. in acdsee you can go back and forward using the mouse-wheel and the image frame stays put but the file actually changes. this way the difference jumped into my face: the smaller image /looked/ like it would be of a higher resolution, since much more detail was present.

i checked this with another movie - same result. on the third movie though, the bigger image had more details, so there must be a magic point. gknot compressablitiy-check seems to give a help here, but there must be an exact optimum.

my question now is of course: does anybody have an idea how to spot out that point?

i´ll attach a zip-file with bitmaps of "Amelie". i suggest to view them as described.
especially the little spots on her umbrella show the effect very good.
idiot me deleted the samples of 8mm as well as the high-res-movie, so i cant´t attach these. 2bad, since they were of much higher quality as amelie, but well...

with regards

/jc

p.s. ...just figured, i can´t attach. might be my router here, it has some firewalling on. i´ll work something out...

[dragoman]

Hi,

I usually encode at 640xXXX for all my encodes. Yes, yes, I know most of you are recoiling in horror and saying "You HAVE to change your resolution based on the comp check!!!"

Well, no you don't. I leave the resolution as it is (even, *gasp*, skip the comp check!) and encode. Haven't had a problem with quality yet....

Also, I usually leave it on "neutral bicubic" no matter the length of the movie...

Yes, I am probably not getting the optimum quality I could by adjusting my resolution and resize filters....but I don't care. I get great looking encodes (DivX 5 Pro is very forgiving) that I can watch on my TV without anyone knowing it's not a DVD they are watching....that's all I care about.

My recommendation - encode at 640xXXX and you'll get good results.

dragoman

[((( atom )))]

well, i am one of those who go for the highest quality. if one stops and is satisfied, there would be no progress. in other words, there probably would not be all this software, that enables us to really store complete movies on a single cd-rom.

also, what about tomorrow? we /will/ have better tv´s, projectors will become cheaper and better as well, etc. and our future equipment will not be as forgiving as todays tvs, so i think it makes big times sence to get the best out movies. also i like to do things reasonable, if i already spend time doing them.

the sample-pictures have to be validated by an administrator i heard, so you can´t download them atm, but if you want, i can send them to you via email. i think, they are worth to be taken a look at.

/jc

[theReal]

compared the same frame of both movies in acdsee by letting acdsee scale the two images to the same sice.I think the scaling in any pictureviewer can have a bad effect on picture quality. Usually, pictures look much worse when looked at in some size like 172% of the original compared to when you look at them in 200% of the original size.

That would of course also mean a 640xNNN movie looks better when watched with a monitor resolution of 640x480, while a 512xNNN movie might look better when you watch it at a 1024x768 resolution.

Maybe also the resizing works better for 608 compared to 624 because 608 is divisible by 32 ? I always keep the width divisible by 32, it can't hurt.

[Acaila]

I'm actually up there with Dragoman. I use either 640 or 704 (depending on number of CD's and movie length/type) for my encodes no matter what the compressability check says. I always watch them on my pc monitor at 1024x768 and they look real great to me.
Going under 640 is just insane in my opinion, but I know some people actually do that a lot. But I don't.

[((( atom )))]

@ theReal:

the resizer in acdsee seem s to quite a good job, because i compared
these pictures:

a, 576X240 extracted from final avi
b, 576x240 extracted from avs-file
c, 608x254 extracted from final avi

just to make sure, i checked out after your post

d, 608x254 extracted from avs-file

both the divx-files of course are samewhat blurry compared to the
avs-ones, but c a lot more than a.
a comes closest to the original-files while while c is blurry and
mising details. as i said, it looks like a lower resolution-image.
b and d look the same exept for a slightly different aspect-ratio.

i really think this is worth being discussed. since people spend
time and thoughts on resolution, this should be taken into these
thoughts, since nobody wants to loose detail.

@ Acaila:
why is it insane, if you would by trusting your eye mistake a lower-
resolution movie for a higher one? I think, your rather high res,
especially for 1cd-movies could be beaten by a lower resolution.

what counts is what your eye tells you, not the numbers. also i
didn´t say, 576xsomething was ultimately the solution to take,
i was actually asking, if anybody had an idea, of how to find the
ultimate resolution for a given source.

@ all:
is there anywhere i can send the zip-file to? you´d be convinced
right away, if you saw the images.

greets

/jc

[theReal]

Sounds very interesting, I think I'm going to compare resolutions more closely next time I'm doing a rip.

I also like resolutions of 640xNNN and up for my movies, but before using filters on a one-CD rip, I rather make it 576xNNN (or even 512xNNN, however anything below that is an absolute no no).

[((( atom )))]

finally ma attachment made it to the post. just check it out and we´ll talk ;)

ok, for me... i´m going out to party now.

[poopity poop]

I try to shoot for around .2-.3 bits/(pixel*frame). If I'm making an SVCD I try to go for .35-.45 bits/(pixel*frame). Which usually comes out to around 640xXXX.
If you are encoding a 4:3 AR 23min. episode I made up a little guide that shows the optimal resolutions for the file size you want on my web page: http://web.syr.edu/~tjmyers/avisynth.html
just look down to number 5. Summuraized here:

320x240 115Mb/ep (6/CD)
448x336 175mb/ep (4/CD)
512x384 220mb/ep (3/CD)
512x384
or 350mb/ep (2/CD)
640x480

but most of you are making real movies which 640xXXX works fine...

[((( atom )))]

@ poop:

checked your quide. -why is there no resolution between 512xXXX and 640xXXX? my research takes place right in that area. did you relate them to usual monitor-resolutions? i am not that afraid of resizing when watching. the ar might get a little out of range, but i could never tell, if i didn´t have a direct compare. following your guide i´d also go for 640 because 512 starts showing stairs even on a tv sometimes.

once i find the time, i´ll just sit and encode ovr and over until i can nail it down to any parameter or value.

regards

/jc

[Acaila]

Another thing you could do is try XviD on that movie. Using DebugView you can actually see what quantizers are being used to encode and you can later restrict them to the quantizer range you like. Scale to a lower/higher resolution and see what happens to the quantizers. It's quite informing.
I usually run one 2nd pass to see the normal distribution and a 2nd 2nd pass with restricted quantizers. I usually come out at a P-frame range of 2-4 or 2-5 for the average movie, and 2-3 for the extreme compressable ones. You can imagine a movie with only frames of 2 and 3 will look superb.

Anyway it's just a thought, but XviD is quite a bit more complicated than DivX5, so it could take some time getting into. But it's well worth it in my opinion.

[Teegedeck]

You can't tell until you see the result. Some movies do actually look good at pretty high quantizers, most don't. I do a compressibility test to get a general idea of the range of quantizers that I'll get (I use XviD and don't let quantizers fluctuate by more than 3 or 4 quants) and then encode some minutes at the suspected average quantizer.

I don't like post-processing, so blocks are a no-no in my encodes, but I also can't stand lo-res. So most of the time, I end up at around 608x for 1 CD. For 2 CDs it really depends. 708x, most of the time, but sometimes an unresized (anamorphous) movie at an average of quant 4,5 looks better to me than a resized one at quant 3.