well first i have to say this-i was using divx for a long time during in which i was satisfied with this codec but when i used xvid 1.0.1 for the first i got hooked-excellent codec guys
now something that bugs me :I
i create xvids for some time now using the guide-because i dont mind the encoding time i use the most time consuming options in order to get the best quality-i always use high bitrates because i dont mind the final size either.specificaly concerning the xvid codec options i always disable the adaptive quant,the trellis quant and i always enable the qpixel,the gmc and the chroma motion.
the created xvids are very good BUT the dark scenes in all the videos tend to have a large number of macroblocks.
yesterday i created a xvid from the movie Signs-its a very dark movie about 1.42 minutes-i used 512* resolution and 700mb final size.
again i used the same setting i use for all my xvids.
well the final xvid was undersized 500mb which is justified because of the movie's high compressibility-as always the quality of the picture in the light areas is perfect but again the quality in the dark scenes is terrible-large number of big macroblocks :I
if u want i can upload some caps from the movie to view
i dont know what causes this decrease of quality in the dark areas of the xvids-i dont think its the bitrate because i always use high bitrates and in the example of the signs movie i cant do much..
adaptive quant is disable so it cant causes this..
any help concerning this matter? :I
There have been a lot of threads discussing how to (try to) resolve that issue, you'll find them making a search on 'dark blocks'
Or quot;dancing blocksquot;
the top link of that search is ongoing
A movie that compress so well that it comes out at under 700mb?
Sounds suspicious to me
You might wanna encode it at fixed quantizer 2, disable bframes and/or increase the resolution a bit
specificaly concerning the xvid codec options i always disable the adaptive quant,the trellis quant and i always enable the qpixel,the gmc and the chroma motion.
I also use high quality most of the time (2 cd's, but mostly 3+). I was wondering if it's any good enabling Trellis Quant, since I'm using very high bitrates and don't have to worry about squeezing quality onto 1 cd. Using the search function didn't come up with any appropriate answer to my question.
Thanks in advance,
Splashdriver
Originally posted by OMINUS
the movie Signs-its a very dark movie about 1.42 minutes-i used 512* resolution and 700mb final size.
Resolution. More resolution. Much more resolution.
Technical possibilities have sneakingly changed a lot. These days, I start whining if I have to use a lame 640ish resolution for some reason.
For Signs, I assume you can keep full DVD resolution.
- Didée
Originally posted by Splashdriver
I was wondering if it's any good enabling Trellis Quant, since I'm using very high bitrates and don't have to worry about squeezing quality onto 1 cd. Using the search function didn't come up with any appropriate answer to my question.
Trellis doesn't help squeezin' anything. It actually results in a sharper, more detailed picture. Keep it enabled.
Oh, and as addenda to Didée's post: Anybody with undersize-problems:
Don't disable B-frames,
don't disable Adaptive Quantization.
There are much more effective ways for upping quality. As mentioned above 1)raise your resolution to full-res, 2)use a good matrix like SixOfNine(-HVS), 3)add something like unfilter(5,4) to your script. Those three will do wonders if you compare it to the very dubious 'gain' you get by disabling some of XviD's best features.
Edit: Oh; and BTW, I have no idea why those dark scenes could be blocky if the encode is saturated. OMINUS, could you make sure that quant=2 was used on those scenes?
Definitely raise your res on this movie as high as you can. Use a nice sharp resizer like say lanczos4resize() which is in new avisynth versions or don't even resize and encode it anamorph. With lanczos4resize you won't notice the difference much from lanczos but on more flat scenes there will be a little bit more detail which will help you out.
I might even try one of the new builds from the head branch as they have a better skip detection which should help on these dark flat areas which tend to get blocky. I would not use any mpeg matrix just use H.263. Quite honestly I don't see the purpose of using them.
You can adjust your b-frame settings, you can play with the number of b-frames (probably 1 b-frame is good in this case) and the ratio.
Whatever, you do though DO NOT filter. Unfilter will not help in this movie. Unfilter adds noise. And it will add more noise to those dark scenes making more blocks and swimming effects. I know because I have encoded this movie myself about 1-1.5 years ago w/ XviD. Filtering is bad news on this one.
I also strongly disagree with Teegedeck about the AQ. In this case, what is the point? It isn't like you have to lower the bitrate like crazy. You have reached saturation conditions. AQ can only do harm then!
Originally posted by Teegedeck
Trellis doesn't help squeezin' anything. It actually results in a sharper, more detailed picture. Keep it enabled.
Oh...I didn't know. Thanks for the Reply.
Greetings,
Splashdriver
Originally posted by BoNz1
I also strongly disagree with Teegedeck about the AQ. In this case, what is the point? It isn't like you have to lower the bitrate like crazy. You have reached saturation conditions. AQ can only do harm then!
Hi BoNz1,
you didn't quite catch me here. Firstly, these were general recommendations, not only for this specific movie; secondly, if you raise resolution and use unfilter (ok, you said it won't do good on this movie) the movie won't be saturated, anymore, so using b-frames, AQ etc. does make sense.
Believe me, I always do several versions of the movie I encode (with/without b-frames, with/without AQ, with/without quarterpel, with/without various unfilter/levels settings...) and I am pretty positive about my recommendation.
Also I am quite dazzled that you don't use custom matrices. High bitrate matrices make for the biggest improvement in quality if you aim for transparency. H.263 is about the worst choice for a quantizer these days (provided you don't encode anime).
Edit: Perhaps a custom-matrix could actually be designed to help with dark movies until sysKin comes by making his HVS-plugin; see:
showthread.ph...880#post527880
thanx guys for all your answers - now that i have time i will try to experiment with your questions-one question though
which experimental build do u suggest to try from the main branch and where can i find theses builds?
thanx again
Originally posted by Teegedeck
Oh, and as addenda to Didée's post: Anybody with undersize-problems:
Don't disable B-frames,
don't disable Adaptive Quantization.Edit: Oh; and BTW, I have no idea why those dark scenes could be blocky if the encode is saturated. OMINUS, could you make sure that quant=2 was used on those scenes?
I have some animes with really dark scenes and panning during these scenes. whatever you do at quantizer 2 (I always keep full quality first pass in case of it'll be the size I want) with mpeg or h263 quantization matrice, you get blocks.
MSP6, treillis, VHQ4, Q-pel, chroma motion, chroma optimizer enabled
Xvid 1.01 performs a little better than Xvid 0.9 (latest quot;oldquot; build by Koepi)
Disabling B-frames helps in this particular case. A matrix like 6of9 helps too, but if you keep B-frames with it, you still get blocks.
I'm actually trying to encode this scene with VHQ for B-frames. Maybe this'll be the quot;magicalquot; thing that'll fix my problem. |