View Full Version : general quality question
luminous
22nd December 2002, 03:46
i just have a general question about the final quality of movies that get encoded. i've been backing up movies for a few months now and i've always used the gknot divx 5 guide and followed it to the letter. for the most part all the movies i've done have turned out just fine. recently a friend had me switch to ogg/vorbis for the audio, which seems to save a little space and sounds better in 2 channels. anyway, the last 2-3 movies i've done don't seem to look very good at full screen (amelie, a beautiful mind, traffic) so i was wondering what the general steps would be to get better quality movies while still only using one cd. do you use more compression? less compression? like i said, i've been following gknot, so i tray to stay in the .17, 50% range.
thanks
bkam
22nd December 2002, 04:23
I'm a newbie so I'm not totally sure, but I think it has a lot to do with the fact that these are pretty long movies. Traffic is I think 146 minutes, Amelie is 129, and Beautiful Mind is 136. Those are all over 2 hours, so they would have to be pretty compressible in order for them to look good on one disc. Since you didn't specify whether you burned to 1 or 2 cds I'm assuming you did 1. Traffic is just a long movie to put on one cd, even if it's fairly compressible I'd think you'd need two cds. Amelie I'm not sure about, but I would say it's probably not too compressible since it has a lot of color as I recall (dark, slow-moving scenes compress the best, which is probably the opposite of Amelie). I don't recall too much of A Beautiful Mind, I would estimate the compressibility is decent but 136 minutes is long.
I know you mentioned that for each movie, you got above 50%. This number is probably about the minimum people would consider acceptable, I think it's all right but many people desire a higher %. If you're maintaining a 50% compressibility with a resolution above 512 (although this is debatable (http://forum.doom9.org/showthread.php?s=&threadid=29352&highlight=acceptable+resolution)) then you should get a pretty good image. The problem that I would see (given the limited info that you posted) is that when you get the compressibility over 50%, your resolution drops too low, so when you go to full screen the picture looks low quality. The solution is to use filters that will increase compressibility, like Convolution3d, or use resizefilters that are more compressible (since you said you followed the doom9 guide really closely, I'm assuming you always chose NeutralBicubic). That means Bilinear (most compressible) or SoftBicubic (second most, I believe). Both of these ideas will force you to take at least a slight quality hit in terms of crispness, but will allow you to resize to a bigger size (remember to do another compressibility test if you change resizefilters). However, if you want to keep (actually, improve) quality without doing either of these things, the best thing is to go for 2 CDs, which will give you a lot more bitrate to work with. Try compressibility testing again after switching to 2cds, and it shouldn't be a problem to get 60% compressibility with 640 resolution (maybe better) for most movies. Right now I would predict that you're working with ~50% compressibility and <512 resolution. You should see a significant increase in quality.
You might also find this thread (http://forum.doom9.org/showthread.php?s=&threadid=40474&highlight=acceptable+resolution) interesting.
I hope this helps, people please feel welcome to tell me how wrong I am :).
luminous
22nd December 2002, 08:22
okay, so i read the other threads, and i'm just trying to get this straight in my mind, so bear with me.
the idea is to get a highly compressed movie at a decent resolution. this would mean a compression % of 50 or greater, and a resolution of 500x__ or better. is that correct?
luminous
22nd December 2002, 08:27
i'm fiddling around with a beautiful mind, and i encoded it @ 35% and 55% for one cd. when i full screen them the 35% one looks better. any ideas why? i'm thinking that i'll just change the settings to 2 cds and see what happens. and yeah, i'm still following the gknot guide pretty closely...neutral bicubic and 'light' whatever the hell it is.
bkam
22nd December 2002, 08:32
Exactly. Higher resolution will give you a higher amount of detail, while higher compressibility % will give you greater quality. In other words, resolution tells you how big the image is and how much you can see in it, but the compressibility % tells you how good the image looks. (For example, a 320x___ video would be very good quality, because it's very easy to compress an image that small, but it would have very bad detail. On the other hand, if you go overboard with the resolution, say a 704x___ with a compressibility in the 20s, you'll have a large image with everything in it, but it will be distorted because it's such low quality).
The moral of the story is that you want the highest of both that you can possibly get. However, there are limits: If you try for over 80% compressibility, many times your file will be undersized (you've no doubt seen this thread (http://forum.doom9.org/showthread.php?s=&threadid=24584)). So in the neighborhood of 60-80% should give you a close to perfect rip. The upper limit of resolution, AFAIK, is 720 since that's the size that the DVD rip is normally. However, if you leave the w-modul at 32 (read: you don't mess around with GKnot, which you probably shouldn't anyway because I think I've read that 32 compresses best) then the max res is 704x___ since 720 isn't divisible by 32. I'm pretty sure that increasing the size of the DVD rip isn't recommended.
So basically, get both numbers as high as possible up to 80% compressibility and 704 resolution. As I said above, the easiest way to increase these numbers is to do a 2CD rip (just click 2 CDs button under the Bitrate tab). In the Gknot guide there are instructions for how to split the 1400 mb file you end up with. If you don't want to go for two cds, then you have some other options, but in general they are more advanced.
Hope that helps.
bkam
22nd December 2002, 08:33
Originally posted by luminous
i'm fiddling around with a beautiful mind, and i encoded it @ 35% and 55% for one cd. when i full screen them the 35% one looks better. any ideas why? i'm thinking that i'll just change the settings to 2 cds and see what happens. and yeah, i'm still following the gknot guide pretty closely...neutral bicubic and 'light' whatever the hell it is.
That doesn't make a whole lot of sense to me, but it might be because the 55% one makes the resolution just altogether too small to be viewable. Am I right in assuming that neither one looks really great?
luminous
22nd December 2002, 08:40
thanks, bryan.
the only 2 cd movie i've done is heat, and it looks great, but i knew going in that it was going to take 2 cds since it's so long. i hate to have the movie on 2 cds, but if the only alternative is to have either a large, ugly rip or a small, clear rip, i guess there's no choice :)
luminous
22nd December 2002, 08:42
yeah, neither the 35% or 55% avs' look great, just the artifacts are less noticeable in the 35%.
OvERaCiD23
22nd December 2002, 08:42
you have it right for the comptests, >50% is what you want. however, fullscreen movies usually don't fit well onto 1 CD (especially ones over 2 hours!). i would suggest encoding the widescreen versions (should fit nicely onto 1 CD), or going to 2 CDs. full screen movies have nearly double the resolution of a WS movie, thus will be much harder to compress. if you want to increase compressibility, use some light noise filtering ~~~> i suggest Convolution3d for this, as it will clean up the picture but not blur it. this can increase compressibility by ~5-10%. search in the AviSynth forum for "Convolution3d" and you'll get a few long threads about it and which settings to use.
bkam
22nd December 2002, 08:49
Originally posted by OvERaCiD23
if you want to increase compressibility, use some light noise filtering ~~~> i suggest Convolution3d for this, as it will clean up the picture but not blur it. this can increase compressibility by ~5-10%. search in the AviSynth forum for "Convolution3d" and you'll get a few long threads about it and which settings to use.
I wrote a basic introuction to c3d (http://forum.doom9.org/showthread.php?s=&threadid=40847) for someone else who was interested in using filters but only had experience with the most basic GKnot methods. It might be helpful if you're not familiar with avisynth scripts (it took me a while to learn from those long convolution3d threads, having never used filters before).
luminous
22nd December 2002, 08:52
overacid,
all the movies i've got are widescreen, so no worries there. as for getting my current project onto 1 cd, here is what gknot is telling me;
0.17 Bits/(pixel*frame)
38.2% of 0.444 compressibility test
488x264
or
0.271 Bits/(pixel*frame)
61% of 0.444 compressibility test
388x208
or if i go to 2 cds,
0.274 Bits/(pixel*frame)
61.7% of 0.444 compressibility test
596x320
bkam
22nd December 2002, 09:05
Originally posted by luminous
0.271 Bits/(pixel*frame)
61% of 0.444 compressibility test
388x208
or if i go to 2 cds,
0.274 Bits/(pixel*frame)
61.7% of 0.444 compressibility test
596x320
308x208 is much too small, you will lose a lot of details (maybe not be able to tell what's going on in certain scenes I would predict). In this case I would definitely go to 2CDs unless it responded really well to a little Convolution3d (which they sometimes do, for noisy movies especially). Definitely try B-Frames and maybe Psychovisuals (although some people hate them) because these will increase compressibility. I don't think you can include them in your GKnot compressibility tests though. The values you got for the 2cd look excellent IMHO.
EDIT: Also, you have included the bits/(pixel*frame). AFAIK you shouldn't worry about these values after you've done your compressibility tests. Just use them to estimate a resolution for your compressibility test, and after that rely on the percentage you obtained. I believe that the bits/(pixel*frame) are an estimate to use for resolution, while compressibility is a more concrete measurement.
luminous
22nd December 2002, 09:21
cool bryan. i think i'll have to go to 2 cds for this one since the values won't get in an acceptable range for 1 cd. also, do you know or have you heard of splitting an ogm video? i've used nandub to split avi's before, but it won't work with ogm.
bkam
22nd December 2002, 09:34
Sorry, no experience with ogm. Anyone else know how to do this?
EDIT: Actually I just saw this (http://www.doom9.org/Soft21/Editing/OggCut.zip) which might help you. It's a program called OggCut, which seems like it might do the job? I have no experience but I saw it on the Doom9 downloads page.
?¿öM¿?
22nd December 2002, 11:26
Use VirtualDubMod, it does OGM as well as VBR audio. There is a thread about it on the VirtualDub board.
If you really want to go 1CD, try getting the UnFilter.dll and STMedianFilter.dll for avisynth. They're available from Doom9's download page I think. I managed to get Traffic to look not great but alright on 1CD and these filters are what did it. convolution3d gave over 10% compressibility increase as well.
;)
JensG.
22nd December 2002, 12:50
I converted the German PAL version of "Amelie" to divx at 560x240 neutral bicubic with a comp test somewhere between 70 and 80 %. ( I do not remember exactly.) So from my experience the movie compresses quite well and the resolution is good enough to compete with video quality.
luminous
22nd December 2002, 20:10
i let ABM encode overnight and it turned out pretty well. i wound up going with 2 cds, 544x292 @ 61% compressibility. the file wound up being 1.14 gigs. considering it's for 2 cds, is there anything i should/could do to fill up the remaining 260 mbs? i could up the res, but then it wouldn't be as compressed, and vice-versa. any suggestions?
as for the other 1cd filtering methods, i'll probably give them a try when i'm not working 50+ hours a week :)
bkam
22nd December 2002, 20:26
Well, the fact that you got an undersized filter means that the codec is saturated at this point. Are you using B-Frames? I would guess that the B-Frames did more than expected, so increasing the resolution, even though it drops the reported compressibility test below 60%, should still look good (when you get an undersized file that means it is the best possible quality that the codec can make it at that resolution). So you should probably increase the resolution a couple notches.
leadman584
24th December 2002, 03:26
Here's another Idea that will work very nicely.
I recently posted looking for a method of changing RC Averaging period. What I got was a suggestion to use GKnot to get a .d2v and .avs files. After closing GKnot open the .avs in VDub and you can set all Divx parameters including RC Averaging. Setting this value to roughly half the number frames in your movie, is a very good one.
I wanted to see if this give decent results at lower bitrates. I did the compress test and adjusted bitrate in GKnot for a single CD. The resulting compress value was 42.3%, much too low for what I have been doing. I plugged the bitrate from GKnot into Divx on both passes and set RC Average to half frame value and encoded. Resuslts were amazing, but film had long stretches of both low and high motion. Basically Doing this will allow for more efficient bitrate allocation across the entire film. The Default setting for Divx averages bitrate across a span of @85 seconds at 23.976f/s. What the heck something else to try. Good Luck.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.