Lumagen deinterlaces 1080i - finally - Page 8  

Clarence

A little confusion on my part. VIDPP should only be used for video 60 frames source correct or does that yield results with movie material 24frames as well. Thanks!

But doesn't almost all video have motion? And if so, that means that video is basically BOB all the time?

-- Rich

(copied from another post)
The plateau for high quality non-film de-interlacing uses bob for those parts of the picture judged to contain moving subject matter and weave for those parts of the picture judged to contain stationary subject matter. By "per pixel" we mean switching between bob and weave many times within each scan line if needed. This is contrasted with bobbing the entire picture if there is a certain amount of motion detected whcih is an inferior method.

The Lumagens have reached this plateau.

Judging whether material is stationary or moving is not perfect, and within the plateau there are some de-interlacers that judge better than others.

There do exist better methods but these are sufficiently expensive so as not to be commonplace. One is called "motion compensation" and it gets better results using more weave and less bob.
(copied from another post)
Use AUTO for film source 1080i. This mode uses 3-2 pulldown recognition falling back on motion adaptive per-pixel for the best results. (If 3-2 pulldown recognition did not fall back on something else, bad edits would result in pure weave with the attendant combing)

Allan Jayne posted the same explanations I'd found...
"Use AUTO for film source 1080i. This mode uses 3-2 pulldown recognition falling back on motion adaptive per-pixel for the best results."

So it looks like Film should use AUTO, which incurs the 54 mS delay with audio. :(

When I turn on the V/F indicator flag, AUTO accurately identifies video (V) and film (F). The User Manual says "the Lumagen processor allows the user to force deinterlacing to assume the source material is video". When I force DEINT to "VIDPP" or "VIDNC", I always get the "V" flag, even for Film sources. So, yes, VIDPP and VIDNC are for Video.

Jim's advice:
MEMA: genlock on, deint Auto (for Film)
MEMB: genlock off, deint VIDPP (for general video)

But after viewing HD concert clips tonight, I know the delay from AUTO is too much for me. And my audio system doesn't have any audio delay controls.

I'm trying to decide if I should keep trying to figure out these deinterlacing modes or try to return to my blissful ignorance of a few weeks ago when all I knew was that simple 1080p looks much better than 1080i.

Now that I've re-confused myself, I'm going to re-watch some film clips in the various modes.

Itching to try this 1080i deinterlacing on my Pro HDP over in UK, but need HD broadcasts to start. :( Soon we hope!

Question: when in AUTO mode (i.e. scaler decided whether to be in Film or Video mode) which type of Video deinterlacing does it use: VIDPP or VIDNC mode? Can this be selected?

If there anywhere I can read about the differences between the VIDPP and VIDNC modes?

StooMonster

Clarence, would something like this maybe help you?

http://www.felston.com/dd540/index.htm

You could use it prior to the video processor, adding a pre-delay to the digital audio stream.

I ran through my test clips again last night with deint=Auto and genlock=on. There were only a couple of instances where the lip sync was off. And I really had to look for it.

Concert clips seem the best (worst?)... especially with closeups of the vocalists with closed-lip consonants... p, b, m. I have a 1080i test clip of "Barenaked Ladies" singing "One Week". Good staccato vocals with close-ups. It's probably more of a negative-placebo effect... I only think I see it when I'm looking hard for it.

Stuart: In Auto is defaults to VIDPP mode when not in film weave mode. I don't think that they are going to allow a choice of NC or PP as fallback when it's not film.

Gordon

Reading the manual ... PP is 'per pixel' which is what you'd want (I guess), and NC appears to be a 'bob' (full screen scale of single 1080i field). Is that right? And what does NC mean?

StooMonster (fingers crossed for Sky HD in the near future)

VIDNC = "no combing"

How does the Lumagen do on the Gary Merson tests for seeing if 1080i is treated as 540p and then scaled to NR or does the 1080i undergo true deinterlacing to 1080p, regardless if video or film based sources? About HD-Leeza?

I remember reading that Gary Merson article (link) ...
I'll use dr1394's .ts test patterns with my HDP and HDLeeza then report back.

I remember reading that Gary Merson article at the barbershop...

Related thread (link) ...
I'll use dr1394's .ts test patterns with my HDP and HDLeeza then report back.

These are not the same patterns used by Gary Merson. That said, they fail with the Lumagen. The top bar in the vertical motion pattern flashes gray adn all white. It is displayed properly by my Panasonic 657UY plasma.

-- Rich

Hi RichB,
The Merson tests, valid for displays, should also be valid for seeing if the processors can truly denterlace 1080i and not reducing the 1080i to 540p prior to scaling to 1080p, correct? Do you have the latest firmware of a lumagen HDP or HDP/Pro?
Thanks
mark

OK, I'll see if Mark Rubin can post the real test patterns...
http://www.avsforum.com/avs-vb/showt...&&#post6609407

I just tested vertrezmotion.ts with the latest firmware 031806. Test failed: the top bar stays mostly white. There is some flickering with the grid lines. I am running auto mode. Also the lines are not as defined as they are in using the Panny scaler. I am using a MyHD 130 card with DVI out that is sent though a Gefen HDMI splitter to the Lumagen and directly to the screen so it is easy for me to compare the de-interlacing performance.

I am not sure if there is a threshold or screen region used to the Per-Pixel method, but this test seems to be causing the scaler to BOB.

-- Rich

I tried to test it, too. My XBox360 wouldn't play it as a .ts file, but I've had that happen on a few other files... usually I can just click the trigger to ffwd 30 seconds and it'll start playing, but this test clip is too short to fast forward.

On my HTPC, I was suprised with how much the outvaried with different players, particularly
VLC (link), MPC (link), and VideoRedo (link).

VLC had the various deinterlacing modes that you could select.

But only VideoRedo played vertrezmotion.ts as a single rotor... the other players seemed to produce two slightly offset rotors.

Another good player to use is the fusionhdtv demo player from dvico.com
http://www.fusionhdtv.co.kr/ENG/Download/Demo.aspx
http://demod.dvico.com/down/FusionHDTVDemo2.60.exe
It also provides Configuration... Display.. Deinterlacing: Auto, None, Bob, or Software.

Also, when I used VideoRedo to resave the .ts as a .mpg, the resulting smaller mpg played fine on the xbox, but with the artifacts described above. I'll also try simply renaming the ts as mpg.

And I'll also try this on my HDLeeza.

there is one thing I don't understand:

does the VisionDVI finally support inverse telecine of 1080i50 (PAL) or ist this reserved for the HDP versions?

Inverse telecine should be much easier when the source is PAL (2:2 pulldown) instead of NTSC (3:2 pulldown).

In 3:2 you have a recognizable cadence, in 2:2 you don't have any easy method of cadence detection... When you have four fileds It's as easy to merge the two fields that actually are a match, as it is to merge the ones that don't!?

cheers

/Par

I'm suspicious that the motion adaptiveness could need another round of changes.

Given the six milliscond video delay (lip sync error) that was mentioned for Per-Pixel, that is not enough time to size up three fields' worth of video to determine sameness (do correlation) and accurately distinguish between combing and fine vertical detail. With three fields (call them A, B, and current field C) on hand (in buffers) you can always do an odd-odd or even-even correlation (with the field before the previous) and for those (ever changing irregular shaped which defines Per-Pixel) areas on the screen that are the same across those fields (A and C) the chances of doing weave (of B and C) for best resolution and getting it right are greatly improved. With just two fields on hand the more difficult guessing doing odd-even or even-odd correlation (sizing up just the current and previous field) has to be done 100% of the time. (This is the same correlation complexity that makes 2-2 pulldown recognition difficult) A test pattern of alternating black and white scan lines is an extreme test that can cause less than ideal de-interlacers to treat a stationary display of same as a combing motion artifact that motion adaptiveness uses bob to deal with.

Since the Lumagen 3-2 pulldown recognition uses (I think) at least three fields given the video delay mentioned for that, much of the groundwork for using at least 3 fields for Per-Pixel appears to be in place.

I don't think Vision DVI processes HD, does it?

oferlaor,
according to the website the VisionDVI does, but the output res is limited to max. 720p (i.e. no 1080p). 1080i is available though.

only the VisionHDP and Vision HDP Pro will output 1080p.

http://www.lumagen.com/testindex.php?module=products

The DVI does HD-Deinterlacing. My question was, if it also does inverse telecine of PAL or just motion-adaptive deinterlacing. I want this Processor only for scaling of film based HD material (1080i -> 720p). I want to use the Lumagen with a 2300$ PJ so I do not want to spend another 1800 for the HDP (thats the price for the unit here in Europe).

Any suggestions?

The DVI does not do 2:2 for 1080i.

Gordon

From what I am told the 'DVI, when updated with the latest software, does have the film mode (3-2 pulldown recognition) as well as the motion adaptive de-interlacing, the only difference is the final output is scaled to no more than 768p (x1280 horizontally I believe).

Correction -- Just checked Lumagen's support and download site and just the 'HDP and 'HDP Pro are listed for the 3-2 pulldown recognition film mode part.

If Gordon says the DVI doesn't do 1080i film mode, I would believe him. He distributes Lumagen in the UK, and runs the Lumagen Support Forum.

The HDP models do both 1080i video and film DI, the DVI just does 1080i video mode / motion-adaptive DI. (All with the latest FW).

Nick

Allan: You had me doubting myself for a moment there;)

Gordon