Ridiculous Quality...

[borgosjoe]

I am looking for a MAX quality method. I have done my share of reading and also of trying different settings and sequences myself. What I am looking for is "Ridiculous Quality". I do not care how large the files get or how long it takes to encode. With that said here is my hardware and usual steps...

AIW 9700
P4 2.4Ghz
1GB SDRAM
1 - 80GB HD
1 - 175GB HD

1. I start off by capturing as an avi using MMC with Huffyuv.
2. Then open with Vdub, cut out commercials, and Frameserve.
3. Using Tmpgenc I then encode to MPG2 and burn to DVD-R with one of many different authoring programs.

So the questions are:

When frameserving with VDub do the filters still apply?
What is the best encdoing method for Tmpgenc?
Max Quality settings for VDub, Tmpgenc?
Any other tips or tricks?


I know that this is an open question and realize that there are probably a million ways to do it, but am just looking for some insight from the pro's...

Appreciate and help on this one...

BJ

[racer-x]

Here's some good info about high quality setting on TMPGenc.
http://dvd-hq.info/Compression.html

[DJRumpy]

When frameserving with VDub do the filters still apply?

Yes.

What is the best encdoing method for Tmpgenc?

This will vary by source. There is no 'magic' setting that will produce the best results for any input. This varies by source, source length, source type (progressive/interlaced/telecine), intended media, etc. Every input can use different settings. What type of output and media (if any) are you planning to utilize? What type of input are you working with?

Max Quality settings for VDub, Tmpgenc?

See above. Depends on your source, and what your converting to.

[borgosjoe]

Source: TV Tuner of a ATI AIW-9700 card

Destination: DVD-R (MPG2)

[DJRumpy]

Is your input source telecined, or true interlaced?.

[borgosjoe]

I think that it is true interlaced after doing some reading and some frame comparisons...

[DJRumpy]

You can verify here: www.lukesvideo.com

Look for progressive, and interlaced frames. A combination of the two indicates a telecined source. A true interlaced source will have no progressive frames.

[tomasine17]

Hey DJRumpy,

Do the filters that you apply in VD combine with the filters applied in TMPGenc when frameserving from VD to tmpg?

Thanks for the help.

[DJRumpy]

The filters in VD are applied to the input going INTO TMPGenc. The filters in TMPGenc are applied during the encode process, after the VD filters are applied.

[hitechjunkie]

The filters in VD are applied to the input going INTO TMPGenc. The filters in TMPGenc are applied during the encode process, after the VD filters are applied.

here's a question:

are there any filters in TMPGEnc that should be used over the ones already available in VD?

for example, i've always preferred VDs de-interlacer and resize smart filters. i do, however, like using TMPGEncs noise filter on toon material.

[DJRumpy]

There are a few I can think of off the top of my head. You shouldn't crop by odd values on the vertical range. It can change the field order. You should also avoid cropping or resizing before running an IVTC filter. Run IVTC first. You should always try to run IVTC before any filter that modifies the input. The IVTC filter needs to find duplicate frames (two frames that are identical). If you modify a frame in some way, it could confuse the filter, so it doen't find the correct 'match'. Running IVTC first helps to avoid this.

You should always test with a small sample, before committing TMPGenc to hours of conversion. You may find that your output looks bad, is complete garbage, or isn't what you expected in some way.

[Blkout]

My question is why do you capture with MMC and then frameserve with VD? I have an AIW 9700 also, I just capture directly with VD, then cut out commercials, etc. in TMPGEnc, author and burn. I guess my question is, why do you go through extra steps to get to the same point in the end? The MMC software sucks for the more experienced user. Its much better to capture with VD.

[DJRumpy]

I agree. VirtualDub is excellent for captures. It's what I use myself for all my captures. Your question is somewhat irrelevant though, as both produce an AVI. Most do not filter when capturing, so the means to get the AVI isn't really important.

I didn't see the commercials step. If you want to save time, use AVISynth, and the TRIM command. It saves you from having to save a bunch of smaller avi's. You just filter out the commercials, and drop it onto TMPGenc for encoding. It requires no temporary avi files of any sort. You can remove commericals from a few hours of captured material with about 10 minutes work finding the commercial's start/stop frame numbers, and your ready to encode.

[drewson99]

IMHO you can achieve basically as good as it gets like this:

Cablebox RCA/S-Video to Analog-to-DV passthrough (no tape)

AVI edit w VD, then highest quality 2pass w/ TMPGEnc.

A

*edit* - need a passthrough device of course, most camcorders will do it, or go the DV VCR route as in my computer specs.

[DJRumpy]

Cablebox RCA/S-Video to Analog-to-DV passthrough (no tape)

Just another method of getting an AVI file. It won't improve quality anymore than capturing through any other device, as they all convert the analog source direct to digital via the capture device.

[drewson99]

aren't some pass-throughs of higher quality than others?

https://www.videohelp.com/dvanalog

The Benefits of Using your Digital Camcorder for Analog Capture

Most Digital Camcorders (but not all) have hardware converters that take analog video and convert them to DV AVI (digital video AVI). Most people will agree that the quality of this conversion is very good. My personal experience was that the DV Cam's analog-to-digital conversion ranged from being somewhat better to much better than the three capture cards that I tried.

I concur with the above statement, having tried many methods.

[DJRumpy]

DV is not a lossless codec. It uses a compression ratio of about 5:1. Huffman is lossless, meaning no quality is lost during the capture process. Your method would produce an inferior capture in this case, all other connections being equal. Many cards capture via s-video, and the high end ones can even capture via component jacks, allowing very high resolution captures. I suppose it would depend on the hardware involved in the capture card, and the connector used. I would also wonder what the cam capture does to the IVTC pattern for NTSC captures. Does it interlace it?

[drewson99]

Many cards capture via s-video, and the high end ones can even capture via component jacks, allowing very high resolution captures.

I wasn't actually referring to connections, as my AIW caps s-vid as well as my d800, I usually recommend this method over Huffyuv because my experience with the two methods has been that pass-through created a higher quality result in the end.

In addition, I wasn't taking into consideration component capture, hardware I am not familiar with.

Do you have a model I could check out? I'd like to go that route, actually.

[DJRumpy]

Just search google with the keywords "capture" and "component" and "video". You should get quite a few hits. Be prepared. They weren't cheap the last time I checked..

[adam]

I would also wonder what the cam capture does to the IVTC pattern for NTSC captures. Does it interlace it?

When you use a pass through method it just keeps the source the same as it is. The result, as far as the pattern of interlaced frames, is the same whether you use pass through and convert to dv or capture through hardware.

[DJRumpy]

Have they come out with any Cam's yet that will capture at 23.976, or 25fps progressive?

[adam]

Yes but they are much more useful for PAL users than NTSC users.

The Panasonic AJ-EZ1 and AJ-D200/210/215 and the Canon XL1 and GL1 have a "frame mode" which does capture progressively but in kind of a hacked way. It supposedly works well, but its not as sharp as a true progressive camera. Also the NTSC versions of these camcorders only record at 30fps.

I think the cheapest dv camcorder that captures true progressive frames at 24fps is the $4,000 AG-DVX100P.
http://www.adamwilt.com/24p/index.html

There some others that cost alot more though.

[Xesdeeni]

The Panasonic AJ-EZ1 and AJ-D200/210/215 and the Canon XL1 and GL1 have a "frame mode" which does capture progressively but in kind of a hacked way. It supposedly works well, but its not as sharp as a true progressive camera. Also the NTSC versions of these camcorders only record at 30fps.

I have the Canon Optura Pi. Do you know if the progressive scan on this camera is also "hacked?"

Xesdeeni

[adam]

I'm not familiar with the Optura Pi, but unless you spent anything around the $3500-4000 range, than I'm guessing its not "true" progressive filming.

Hacked probably wasn't the right word. I'll just quote what this author has to say, because I don't totally understand it myself.

http://www.adamwilt.com/DV-FAQ-etc.html#filmlook
These cameras get their "proscan" images not by truly perfoming progressive readout on the chips, but rather by offsetting the green CCD's read timing by one scanline during readout -- vertical pixel-shift, if you will. In essence, an even field from R & B CCDs is blended with an odd field from the G CCD, giving you a frame that has the scanlines for both fields captured at the same instant in time. This gives a definite improvement over mere field-doubled "frames", but it's not as sharp vertically as true proscan. Each "scanline" is actually composed of two scanlines from each chip, so there is some softening vertically; also, the effective chroma resolution is halved vertically.

It sounds similar to blend deintelacing.

[Xesdeeni]

The Optura Pi is a 1 CCD camera, so I'm not sure how that translates.

For the 3 CCD cameras, it sounds like the CCDs are hard-wired to do interlaced capture, so they take the red and green from one field and the green from the other to get a YUV that is kind of a combination. I see why this would reduce the resolution. But I'd have kicked my CCD manufacturer in the shin for not allowing me them to capture and hold both fields simultaneously, even if I had to read them out as two fields (I know the CCDs have a bandwidth limit on how fast they can ship out data, so you probably couldn't read out both fields in 1/60 second).

Xesdeeni