Errors in Avisynth. Wondering if my DV capt is OK?

[RathO]

I capted from miniDV Sony DCR TRV-25 to Studio 8. Then i exported to avi file using the Mainconcept Codec 2.04.

Im using AVIsynth to create *.avs.
I wanna use the plugin: Convolution3DYV12.dll to de-noise.
But i get the error: Convolution3D supports YV12 color format only.

So my question is: Is my DV.avi the problem?

_______________
Im workin on P4@2.8gHz w/512Ram under WinXP

[OvERaCiD23]

nothing wrong with your capture file.

sounds like you're using Convolution3dYV12 in AviSynth v2.0 (not 2.5). Use the regular Convolution3d.dll, and all should be fine.

if you're planning on putting the footage back onto tape, you will need to filter each field separatley or you will lose quality due to losing the fields through filtering (search the forum, the script is posted multiple times on how to separate the fields and weave them back together after using C3D). if you're planning on keeping it in a format that doesn't support interlaced material (i.e. DivX), you should deinterlace using Decomb and then use C3D.

[RathO]

Originally posted by OvERaCiD23
nothing wrong with your capture file.

sounds like you're using Convolution3dYV12 in AviSynth v2.0 (not 2.5). Use the regular Convolution3d.dll, and all should be fine.

if you're planning on putting the footage back onto tape, you will need to filter each field separatley or you will lose quality due to losing the fields through filtering (search the forum, the script is posted multiple times on how to separate the fields and weave them back together after using C3D). if you're planning on keeping it in a format that doesn't support interlaced material (i.e. DivX), you should deinterlace using Decomb and then use C3D.

Im Using Convolution3dYV12 in AVIsynth 2.07 (that i installed after i setup quick2pass 0.71 / for XviD encoding...
AVIsynth 2.5 is alpha .. should i install it? or i may keep my 2.07?

Im also not planning to get it back on tape, ill do a 2 pass XviD.

About deinterlace, should i use Bob or Decomb?

I think i should also walk to the AVIsynth forum since we*re not talkin anymore about DV since my DV output is ok. (theres alot of noise but its ok... as long as i can denoise!)

[bb]

Either use Convolution3dYV12 in conjunction with AviSynth 2.5, or use Convolution3D 1.01 with AVIsynth 2.07. As far as I know DV is specified for YUV2, i.e. it doesn't support YV12 anyway.

You didn't say what your target format is, so I assume DivX (or XviD, ffvfw, etc.). Use Decomb for deinterlacing. You can get good results with Bob, if you want to double your framerate, else drop it.

bb

[sh0dan]

DV PAL is almost similar to I420 (which is actually almost the same as YV12, except the chroma planes are swapped). NTSC DV is different, but more like YUY2 (but only has chroma for every 4 pixels horizontally).

See http://www.adamwilt.com/pix-sampling.html


... and yes, you are using the wrong filter version - download the plugin for AviSynth 2.0

[RathO]

thx dudes for helping. Ok bb, i just got convolution3d-1.01 using avisynth 2.07 i have and ill give it a try.

PS. As said previously, i wanna get XviD 2pass.

Ill be back with results!

[RathO]

Seems to work with:

LoadPlugin(<path>\SmoothDeinterlacer.dll")
LoadPlugin(<path>\Convolution3D.dll")

AVIsource(<path>\myfile.avi")
Smoothdeinterlace(doublerate=true).trim(1,0)
Crop(8,0,704,480)
BilinearResize(640,480)
Convolution3D(preset="movieLQ")

Doing XviD 2pass using VirtualDubMod.. @ 11 fps!!.. is this normal?

[bb]

I'm afraid yes, that's normal. The filters eat up the CPU. For better quality you should place Convolution3D before the resize.

By the way: Do you get the correct aspect ratio using your crop / resize settings? I'm still looking for somebody having an NTSC camcorder to verify those settings to make sure that circles don't come out as ovals.

Just take a short clip of a square sheet of paper, crop / resize and watch the result on a computer monitor (TFT preferred, but CRT is ok as well, use a 4:3 resolution like 640x480, 800x600, 1024x768, etc.). Measure the width and height of the paper square on your monitor and see if they are the same. I'd be really happy to know, so that I could add this information to the guide.

How does your doublerate video compare to the standard deinterlacing way using Decomb's FieldDeinterlace?

bb

[onesoul]

Originally posted by bb
How does your doublerate video compare to the standard deinterlacing way using Decomb's FieldDeinterlace? Well there's no way to compare this since doublerate won't blend or drop fields that fielddeinterlace does. You'll get smoother motion at fast rotation of the camera. What's really a shame is that you can't play such video at tv.

[RathO]

bb, OK ill do the NTSC test with piece of paper this weekend. But i have a LCD Screen TFT with native 1280x1024... still ok?

BTW, the convolution3d thing was a big crash for me. Im very disappointed.

I got to admit that ive been trying soooo much settings to get so so results. OK, my source is DV and the DV Video Encoder used by Studio 8 to export to avi file makes noise (im not sure if its Mainconcept 2.04 or a DV encoder from Studio that is used here. For sure i select DV Video Encoder in Studio8....), so heres what i got from gknot:

------------
# PLUGINS
LoadPlugin("<path>\mpeg2dec.dll")
LoadPlugin("<path>\decomb.dll")
LoadPlugin("<path>\SimpleResize.dll")
#
# SOURCE
avisource("<path>\myfile.avi")
#
# DEINTERLACING (1)
FieldDeinterlace()
#
# CROPPING
crop(8,40,704,400)
#
# RESIZING
SimpleResize(640,400)
#
# DENOISING: choose one combination (or none)
# 1) little noise (fast)
TemporalSmoother(2,1)
------------

Then i followed the Doom9 guide fot DV to DivX for the 2 pass process.
I use DivX 5.02.
I encode to have like 60 minutes on a 700 MB CD.
Results are ok..
Bitrate is ~1560kbps

Do u guys have better settings than that?
or does it always depends on lots of factor... :)

[OvERaCiD23]

I got to admit that ive been trying soooo much settings to get so so results

What is your complaint with the output? You can't expect the DivX video to be an exact replica of the DV video. DV is very noisy, thus very hard to compress, and even noise filters as good as C3D can't remove all the noise while maintaining a sharp picture.

C3D will only remove so much noise, and you'll have to deal with the encoded results. I could never achieve quality DV to DivX transfers (quality in my eyes, I'm sure they'd look great to the casual observor though), so I simply use DV tapes as my medium and have no quality loss. Seems like most of my shots are low-light (casual footage, I'm not a movie-maker), and that is near impossible to fit 60 minutes of quality video onto 1 CD. I'd say stick to the DV format, or accept the results for your encodes.

Combinations of filters work well too. TemporalSmoother(2,1) followed by Convolution3d(preset="movieLQ") can yield nice results, with an acceptable amount of blurring. Also, there are other spatial-temporal smoothers out there ~~~> NoMoSmooth, FluxSmooth are also very good and are much faster than C3D. But like I said before, I could never get my DivX-DV videos to look great, so I just stuck with the DV format. I have little need to put them on CD anyways, so think about your need for MPEG4 conversion and go from there.

[RathO]

Originally posted by OvERaCiD23
DV is very noisy, thus very hard to compress, and even noise filters as good as C3D can't remove all the noise while maintaining a sharp picture.
Thx OvERaCiD23 for this reply. Thats the kind of reply I needed to read, cause im going crazy with quality thing... rhaaaa (my last breath). I didnt know that all DV were noisy.. I thought that my DV Encoder was doing all that sh*t.

is near impossible to fit 60 minutes of quality video onto 1 CD. I'd say stick to the DV format, or accept the results for your encodes.

I absolutely cant afford to keep all my footage on DV tapes, so ill just encode my stuff using 2 pass DivX5 and accept the results of my encodes.

Thanx! :D

[bb]

Originally posted by RathO
bb, OK ill do the NTSC test with piece of paper this weekend. But i have a LCD Screen TFT with native 1280x1024... still ok?

Unfortunately 1280:1024 is not 4:3, but 5:4, so you have to take that into account. I'd say just take the measures and post the results. I'd like to know the dimensions of the viewing area, too, so I can verify if the display has a 1:1 pixel aspect ratio.

bb

[bb]

Originally posted by RathO
I didnt know that all DV were noisy.. I thought that my DV Encoder was doing all that sh*t.
It's not true that all DV is noisy. Low-light shots are noisy, sometimes more, sometimes less, depending on the camcorder. If there's enough light, there's no noise in DV - except for the quantization "noise", which is present in all DCT codecs (DV, MJPEG, MPEG-1, MPEG-2, MPEG-4, ...). But bright video is also harder to compress than dark video. And DV shots are often very sharp, which makes it harder to compress, too.

RathO, for DivX 5.0.2 you can set pre-processing to "light". This is a compromise between denoising quality and encoding time, at least worth a try.

bb

[OvERaCiD23]

It's not true that all DV is noisy. Low-light shots are noisy, sometimes more, sometimes less, depending on the camcorder.

Low light shots are incredibly noisy, and I have never thought of trying to compress them with high quality. By the time the shot has enough light to be "noise-free," it's brightness will be hard to compress (not quite the same level as the low-light shots, but still...).

A suggestion for RathO, do CompressionTests before you encode to ensure you're getting acceptable quality. Download the DivX5Enc from here: http://jonny.leffe.dnsalias.com/. Create your final avs (with all the filters you wish to use and such), and do a CompTest (3%-5%, use 'SelectEvery'). Then, open GordianKnot and set all the parameters to your DV clip (# of frames, audio size, desired file size, resolution), and then load the log file created by DivX5Enc into GKnot. This will return a %, which indicates how compressible your clip is (sorry if you know all of this from encoding DVDs, but I don't know so I'll say it all anyways). For DVDs above 60% is acceptable, I might drop down to 50% for DV footage. You can then see where you need to tweak settings to obtain good results (lower the audio size, more filtering, lower resolution, etc). Then with your new settings discovered from the CompTest, encode as you normally would (actually, DivX5Enc can automate the encoding process as well, but only if you like). This will help you to prevent encoding the whole clip only to find out the results are not acceptable.

[RathO]

Wow, thx for the info, ill do some DV dource encodings this week-end.
I got 2 christmas clips. Hopefully, one was taped during the day, so video is sharp, and the other one was shot during the night, so more noisy as i can see.

Ill let u know the results..

[RathO]

@OvERaCiD23

I must admit that DivX 5 Enc does a real good job for tha CompTest.

I did my first clip that was shot during the night, inside house, with ~enough lights. Here is what i got for my 21min53sec clip:

Job 1 start: Sun Jan 19 01:46:26 2003
Job 1 end: Sun Jan 19 01:47:38 2003 (00:01:12)
Job 1 status: [Ok] 56.66%, target size: 269,379,435, predicted size: 475,443,660 (1356 / 39358 frames used, 29.970 fps, 1313.25 seconds, 640x336)

So i reached the 50% u said, and bit under the 60%.
Dunno if im missing some info to make it clearer or what.

The other clip ill do is real dark... well see for compressability check!

[RathO]

@OvERaCiD23

I must admit that DivX 5 Enc does a real good job for tha CompTest.

I did my first clip that was shot during the night, inside house, with ~enough lights. Here is what i got for my 21min53sec clip:

Job 1 start: Sun Jan 19 01:46:26 2003
Job 1 end: Sun Jan 19 01:47:38 2003 (00:01:12)
Job 1 status: [Ok] 56.66%, target size: 269,379,435, predicted size: 475,443,660 (1356 / 39358 frames used, 29.970 fps, 1313.25 seconds, 640x336)

So i reached the 50% u said, and bit under the 60%.
Dunno if im missing some info to make it clearer or what.

The other clip ill do is real dark... well see for compressability check!

[OvERaCiD23]

The thing is, there is no hard-set limit for the CompTest. But from my experience, below 50% WILL induce very visible macroblocks into your encodes. I'd say your results are good. I'd say stick with those settings as long as the video is pleasing to your eyes.

PS....You never did say if you were using any DivX5 Pro options (B-frames, Q-pel, GMC). Are you? B-frames is the only one you should be using, as noone reports any bugs with. Q-pel and GMC constantly have bugs reported, and are not commonly used. B-frames should increase compressibility by ~10%.

[RathO]

Yes im enabling the b-frames.
I do not enable other "GMC" or "Quarter Pixel" stuff..

Im thinking of putting on my web page the way i encode my DV sources now.
Of course a big link will re-direct to doom9 web site since im not creating anything new, but just to show the way i do things from now on..

Im finally satisfied with my settings and, as you say, my encodings are now pleasing to my eyes!

Thanx dude for the help!

PS. im planning to change the program im using noww for DV capture (Studio 8), cause i lost the sound at the end of a 32 minutes capture... rhhaaaaa... im thinking of ScLive or DVIO. Ill let u know if it works great.

[OvERaCiD23]

Glad I could help. As for a new capture program, without a doubt go with scLive. It's well worth the money, as it makes capturing so simple (two clicks to capture and have all your scenes ready in real-time). DVIO is good combined with the freeware version of Scenalyzer (still downloadable from their website), except it's a two step process to capture into one big file and then use Scenalyzer to break it up into scenes (same end result, but a problem for people with smaller drives).