Lumagen deinterlaces 1080i - finally - Page 2  

Stopdog

Im a little confused. This is not for movie material yet correct. Are you talking about improvement when watching 1080I video?

Alan

I think its for both Film and video 1080i. From Lumagen website :

Q: What do the diiferent deinterlacing mode settings do?
A: As of the 110705 sw release the deinterlacing mode command was updated to add some HD deinterlacing modes. The command is found in the menu under IN->CONFIG->CNTRL->DEINT->MODE . If you're on an HD input you can set it for SD or HD sources while if you're using an SD input you'll only have a setting available for SD deinterlacing. The 2 settings for SD are "AUTO" and "VIDEO". The "AUTO" setting automatically changes between film mode and video mode based on what type of source material is detected. The "VIDEO" mode forces the SD deinterlacer into a video deinterlacing algorithm no matter what type of source material is detected and may be preferable for some video sources.
The HD deinterlacer currently has 3 settings. "AUTO", "VIDPP" and "VIDNC". The "AUTO" mode automatically chooses the optimal algorithm based on what source material is detected. The "VIDPP" mode forces the use of a per-pixel motion adaptive video deinterlacing algorithm which eliminates/minimizes combing and provides for very good detail. The "VIDNC" setting forces a video deinterlacing algorithm which completely eliminates any combing but is does not give as detailed a picture as the VIDPP mode.

I don't think the inverse telecine is in effect yet, but I could be wrong. For 1080i film I set the Lumagen to "VIDPP" and Genlock ON. Looks fabulous, maybe that 1080p digital will have to wait a year or so still. Now that I have it set up so well, it will give good chance to compare to 1080p DLPs and Rubys when I go to CES.

Oops, now that I read it again, you may be right, this one for HD video deinterlacing only. Film to come soon then hopefully. Another upgrade coming then, yipee.

guys I run an 8" barco crt at 720p and 1080i for hd. would 1080i output as 720p look better with the latest hd deinterlacing?

Alan, can you or someone else please explain to me what Genlock does as I have not been a ble to find a description which makes sense to me.

TIA,

Joel

Genlock causes there to be a lock step of one output frame for every input field. Without genlock, there may be an occasional extra output frame or an occasional skipped frame relative to the input notably because NTSC is 59.94 fields per second and video processors may set their timing on the 60 Hz power line frequency. Should an added frame be the fourth for a 3-2 pulldown threesome or a skipped frame be one of a pair of 3-2 pulldown, judder would be more pronounced.

I'm not exactly what Lumagen is doing but "genlock" typically refers to locking the output of a device to an external reference. This is so that multiple devices are in phase with each other. Example, all the cameras at a football staduim are genlocked to a commen reference along with VTRs and graphics systems . Otherwise you could not cut cleanly or mix the signals.

Now I would assume any scaler would lock it's output clock to the input sync. But perhaps not as in Allen's description. There is a problem if you are inputting 59.94 and sending out true 60hz. You really can't lock those together, Sooner or later a frame must be dropped or duplicated. However with advanced processing it's possible to do this very cleanly.

FWEIW, any signal you get in the home is based on 59.94. Film is transferred to HD video at 23.98. So you rae better off keeping your scaler output at 59.94 or some multiple thereof. If you want 48hz then you should really output 47.956.

Lumagen's description in their release notes leads me to believe they are locking the output to the selected input; so in a sense, it would be an external clock. The notes read like they're varying the output timing slightly, but I can't see the benefit of doing this arbitrarily. I think they might have meant to say that the output tracks the input and could vary according to the input, rather than being asynchronously clocked to an internal reference.

Lumagen's description in their release notes leads me to believe they are locking the output timing to the selected input's timing; so yes, it would be locking to an external clock. The notes read like they're varying the output timing slightly, but I can't see the benefit of doing this arbitrarily. I think what they might have meant to say is the output tracks the input and so could vary according to the input, rather than just being asynchronously clocked to an internal reference.

Glimmie,

When they genlock to 60Hz, they're doing 59.94 or it's not genlocked. There's no way to adjust the 1% difference without losing a frame every 100 frames.

The lock is solid and can be an integer multiple of the original sync (2:2 or 3:3 pullup).

Well yes, you have to drop a frame or field sooner or later. But you could analyze motion and do it where it wouldn't be noticed. But that's most likely beyond these <$5K scalers.

1920X1080, allan, Glimmie, Oferlaor, et. al.:

Many thanks for responding to my query regarding GENLOCK...bottom line, though, is as follows:

1. What is the optimal setting for GENLOCK -- on or off.

2. What are the criteria/inputs/parameters for determining wheter GENLOCK should be on/off?

Should it be of any help I am using an InFocus 7210 DLP projectir which is being fed sy a SA 8300 HDPVR and a TMA DVD32R DVD player porjecting on a 92" Stewart FireHawk.

TIA,

Joel

Please see the 2nd post here: Genlock - How/When to use
All things considered, my guess is you're not doing any sort of frame rate coversion, so you're a candidate for GENLOCK=On.

Thanks a lot...I will give it a whirl.

Joel

Glimmie,

I have never seen "Context sensitive" frame drops...

I had a similar idea to wobbling (tried sending it to everyone I knew in the industry that could benefit from it, to no avail.

The idea is to do wobbling (the annoying feature that's important to reduce burn-in on plasmas) in a context sensitive manner. Many people are deeply annoyed at a static scene that suddenly gets shifted by one pixel.

If it gets done when scenes change, it can be done a lot more often and a lot more dramatically, causing no more than logo shifts.

Some of the original iScan deinterlacers had a feature to do exactly this. The deinterlacer could detect when a scene change occurred (since there are deinterlacing issues to be concerned with around such an event) and would shift the image by a pixel or two synchronized with the scene change. This was set up to happen only when a 4:3 image was pillarboxed inside a 16:9 frame, as that was the case in which we were trying to avoid burn-in.

- Dale Adams

Dale,

Are you still doing that today?

Any idea when the 1080I film de interlacing will be added to the Lumagen. Is there a chance this could be added in time for an Xmas present :)

Alan,

They're working on it now. That's all I know - no ETA.

They seem to work miracles getting their current machine to do so many things so well, I wonder what about this one last but very important feature is proving so difficult to implement?

Mark,

there's no doubt it's a very hard thing to do.

Thanks for the insight. :)

I would like to ask if Patrick would confide:
How many fields do the Lumagen processors work with for motion adaptive HD de-interlacing?

I am guessing that three field -- current, n-1, and n+1 -- processing will come very close to correctly de-interlacing film source without actually sensing 3-2 pulldown. This does require sensing sameness between odd and even fields. Whereas two field-- current and n-2 -- requires only even-even or odd-odd sameness detection, is not close to 3-2 pulldown sensing. Meanwhile two field -- current and n-1 or current and n-2 -- can give reasonably good motion adaptiveness.

Or maybe as a poll question I should just ask everyone with a Lumagen processor with the latest software, how good do HD film sources look so far? (I don't have any HD sources at home to see for myself on my HDP)

I think HD film looks great. If I concentrate and scrutinize the picture, I can pick up a subtle strobe at times; I'm not sure how to explain it. But if I "relax and enjoy the show" I'm good. As far as resolution, I'm tapped at 768p, so perhaps it will be interesting to compare what, an upscale from 600-700 apparent lines to a downscale from 1080p?

Thought it might be time to bring this thread back up with this query. :)

Joseph- a quick update on 1080i inverse telecine---we are running it in the lab now. Its working very well but we are tweaking a few things. It was harder to get done than we had thought but we should have an initial release of it middle of next week.

A great product is just about to get a whole lot better. Well done Lumagen.

Patrick, you guys are the best. Anybody that has followed this product should see that. Thank you yet again for the continuing improvements! :)

No kidding!

I lose track of how long this has been in the pipeline, with no roadmap in sight, so to get this particular update is great news.

Can't say how much so many are looking forwards to this!

Nick

Patrick,

This is great news. It sounds like Lumagen performed a miracle to pull this off. I heard that the chip(s) you have to work with were somewhat over-spec'd by the suppliers. Plus, running within tight timing constraints on an FPGA is tricky.

Are there still plans to finish the 10-bit pipeline work?

Patrick?