hard work... how to convert j-motion to xvid with good quality and acceptable size

[xklyx]

my camera produces the video, format:
j-motion 640*480 30fps,bitrate is very HIGH:about 13000-22000kbps
audio format ignored(no process,just frame copy)

so I need to compress it,xvid may be a good choice :confused:

disasspointed,i've tried to use virtualdub,and choose xvid, set single pass,quantizer 4,the result is the bitrate of the target file is very HIGH(4000-6000kbps),but very smooth.i use DRFAnalyzer,it shows the xvid file Average DRF/quantizer is about 6,some files may down to 8!:mad:

it is obvious that the j-motion is low-quailty,but i don't know how to discribe,you may see the picture below.

Is there anyway to compress my files with bitrate about 2000kbps and Average DRF/quantizer less than 5?

I've tried to upload a sample of my j-mtion video,it's too big,i just cut a little segment

I'm just a starter,please give me some advices,thanks a lot!

PS:forgive my English

[xklyx]

the last 2 video segment
thank you for your help

[Teegedeck]

Attachments approved - but please use some kind of online-filestorage service next time.

[foxyshadis]

Garbage in, garbage out. You're going to want to fix that up with some filters (presumably with avisynth) first; it needs a good derainbowing, stabilization, and some distortion removal. Doing that will significantly lower the required bitrate on its own.

Then you need to actually configure the codec, instead of using the default fast, low-quality speeds. If you have a rough target bitrate, you should probably also use 2-pass, which gives you total size prediction without occasional severe quality drops.

[xklyx]

Garbage in, garbage out. You're going to want to fix that up with some filters (presumably with avisynth)...

could you tell me the details,about which filter will be better?i've tried avisynth but I don't know how to configure the filter.
I've tried the VDM's blur filter,but not effective,and it also reduce the quailty.

i want constant quailty,may 2-pass help?

[foxyshadis]

I'd start with Deblock or Deblock_QED or blocking, then TComb or GuavaComb for the rainbowing, or fft3dfilter(plane=3). MVDepan/Depan by (read the documentations carefully, especially the examples) to stabilize it. Then either more fft3dfilter, vaguedenoiser, mvdegrain, degrainmedian, frfun7 to clean up all that ringing... there are lots of possibilities, depending on your patience and needs.

Note that there's also an fft3d variant called ft3dgpu that works with your video card, if you have a fast enough one.

You might also want to go back to the original source so you can do a better deinterlace, but it's hard to gauge how good or bad this is based on such a short sample. I suspect the water patterns are an artifact of a strange deinterlacing algorithm, though.

2-pass will give a generally constant quality with a well-defined file size. You should also see the xvid presets thread if you really want to dig into setting it up.

[xklyx]

:thanks:
I have found a video,it's taken by the same camera model I use
you may use some download tools

http://pub.pconline.com.cn:8001/cms_pconline/vfs/0000288/000028802/0610/other/061028_r5_movi.part1.rar

http://www.pconline.com.cn/pingce/pingce2006/digital/dcpc06/0610/other/061028_r5_movi.part2.rar

I find out that if i use code Xvid Mpeg-4 Codec(see picture 'xvid'), the set q4,the filesize will be smaller,but DRFAnalyze shows the real DRF belows 4,may be 5 or worse,compare with useing ffshow,set as Xvid q4,the filesize will be much bigger,and the DRFAnalyze shows the real DRF is 4 exactly .
So I suspect if the Xvid's Target Quantizer have precision problem?

I do care about DRF value

I think to compression this video is a great challenge :cool:

[foxyshadis]

Like I told the other guy who cared about DRF: You're caring about the wrong thing. You're trying to guess how the engine runs by looking at how much the paint is peeling. You need to learn how the codec works, for instance, you don't even seem to know that b-frames automatically have a higher quantizer than other frames, which DRF analyser doesn't explicitly adknowledge. (It's an important part of how xvid works!) If you don't have a pretty deep understanding of how the codec works, DRF is a nearly useless statistic for you.

Here's one quick example why: By changing the quant matrix, you can encode the same source with the same settings and the same quantizer, and there can be a 3 to 1 size difference between them (ehr and ulr), and DRF analyser will report exactly the same quality, even though it's painfully obvious that it isn't if you look at them.

I already recommended the xvid presets thread (http://forum.doom9.org/showthread.php?t=119399) because it will give you quick information on the most important xvid options to change - leaving them all default is usually a bad idea - to figure out how compressible your source is, so you can achieve the quality you want in the time you want. That way you're almost certain to get good quality and a good DRF, not just one or the other.

[xklyx]

I learned so much from you,thank you for your post.I'll keep learning,but i think it will take me times to understand so many things :p