HD to SD; lot's of aliasing...

[Leo Ticheli]

Thanks, John,
The problems manifest themselves with motion, not the still frames.

I'll be trying more software solutions when I get back to the shop; I'm shooting Monday, so will have little time tomorrow. I'll dig into it as soon as I can.

Very interested in trying the DPX method.

Best regards,

Leo

 

[JohnF]

Hi Leo,

Yes I understand the problem with horrible aliasing on motion footage. The pre-blur before resize works for motion too!

Though as RedGuy mentioned AfterEffects is much better at resizing than programs like FCP.

Hope you find a solution.

JohnF

 

[Adam Glick]

Hi Leo,

I was working on this issue in the lab yesterday and generated NTSC DPX frames - with the same aliasing artifacts...

I've done the transcodes with the debayering set at both full and 1/2 res with roughly the same results but I was thinking...

What if you you shoot at 2K? I wonder if the filters in Red Cine (and Scratch Cine) would have an easier time of the downconversion?

I have some 2K stuff I will try this with when I get back Monday morning.

Also, 720P stuff looks pretty darn nice coming out of my Aja Xena (same as Kona on the PC) at NTSC... Maybe there's an r3d-to-HD-to-SD workflow that won't be too much of a stretch.

Adam
BOXXlabs

 

[Ivan Cortazar]

Hello there,

I don't know if this will help, as I don't have my Red yet and haven't play with the footage. What Graeme says makes sense. Some filtering needs to be made in order to reduce the extra detail that wouldn't fit.

I have been doing that in Final Cut or After effects every time I drop a High Res photo into an SD sequence. Before I import the photo into AE or FCS I use a 1, 1.5 or 2 pixel vertical "motion blur" in Photoshop. I think I read somewhere that it is better to use a Vertical motion blur filter instead of the standard Gaussian blur, as the aliasing problem is more relevant on vertical resolution. The vertical motion blur won't blur the horizontal so results in a crisper SD image.

This may make no sense, but for me, it has worked pretty well on importing Stills on FCP. But now that I think.... this filter was to help the flicker of the photos because of the interlaced footage, but if it is progressive ... Oh, well... Now I'm confused... :blink:

IC

 

[Michael Schrengohst]

I like AE because I have used it for years. I know how to get the results
I need. REDCINE is only a conversion tool. I don't like to use it for
"final results". If you are cutting SD I still like to have an HD source I can
play with. If you scaled to final in REDCINE you can screw yourself
if you need to crop and re-frame.

 

[donatello b]

i'm not seeing this aliasing on clips that i've taken from 4K 2:1 to 16x9 SD ...
i tried to use Leo's clip but it's 4k 16x9 and it crashes Redcine when i attempt render so i can't tell you about that clip ...
my normal 4k 2:1 to SD .... take the 2k proxy into FCP .. render out to 2k (2:1) cineform ( rendered clip will be 8 bit) .... then i take it into Sony Vegas
project set up as 1080p 23.976 .. the clip will be letter boxed in the preview window .. use the fx/pan on the clip and change the 2:1 to 16x9 so the image fills the 16x9 window .... add CC etc .. then render out to widescreen SD 23.976 + pull down ( SD cineform clip & make a mpeg2 for SD DVD )...i only have one interlace viewing device, 14" sony 14u4u monitor and the image looks very good .... i project the DVD onto a 110" screen and it looks very good ( projected progressive).... perhaps there is something in the 10bit to 8bit that knock out some of the high detail when i go from 4k to 2k ??
the cineform clips will work on both PC & Mac .. i have not taken the 2k cineform into FCP and rendered out to SD = i'm a PC'er ... perhaps that would be a test to see if it's the FCP rendering process ?

 

[Dj Joofa]

Hi Leo,

Yes I understand the problem with horrible aliasing on motion footage. The pre-blur before resize works for motion too!

The proper blurring filter for motion is a temporal filter that operates on time axis as opposed to pre-blurring filters that operate on spatial axes. It is possible that Red is not using temporal filtering.

 

[Graeme Nattress]

We're not talking about temporal aliasing (wheels spinning backwards and the like) but how spacial aliasing is made more obvious by images moving over time. Spatial pre-blurring to stop aliasing before a downsample will reduce aliasing and as there is now no aliasing, the temporal effects will not be able to make more obvious what is not there.

Graeme

 

[Dj Joofa]

We're not talking about temporal aliasing (wheels spinning backwards and the like) but how spacial aliasing is made more obvious by images moving over time. Spatial pre-blurring to stop aliasing before a downsample will reduce aliasing and as there is now no aliasing, the temporal effects will not be able to make more obvious what is not there.

Graeme

It is not about wheels spinning backwards. Temporal filtering is used for noise smoothing. Even if there is no motion hence no wheels spinning backwards, there is always noise on the same pixel location that varies from image to image. Now you can suppress noise by spatial filtering also as you rightly point out. However, a better way in some sense is to tackle this noise in temporal domain, as it is after all temporal in nature. Because two adjacent pixels have can and does have different levels of temporal noise, hence noise statistics in time domain helps here instead of blurring them spatially.

Spatial is also used in conjunction with temporal filtering. However, spatial filtering is for tackling other types of noises/artifacts.

We see cases all of the time when no temporal filtering results in artifacting, poor compression, and overall noisy appeal to the image.

 

[Graeme Nattress]

I thought we were talking about aliasing, not noise.

Graeme

 

[John Tissavary]

I like AE because I have used it for years. I know how to get the results I need. REDCINE is only a conversion tool. I don't like to use it for "final results". If you are cutting SD I still like to have an HD source I can play with. If you scaled to final in REDCINE you can screw yourself if you need to crop and re-frame.

RedCine is definitely more than a conversion tool - it's got a fine set of first light color tools, and it's got some of the best scaling algorithms currently on the market. I wouldn't consider it as my first choice for a finishing tool, but I used it in earlier versions before getting Scratch and was able to finish quite a large percentage of shots with it, and get others within 10-20% of finished (mathematically & visually speaking). I always recommend getting as far as possible can before debayering when working with r3d. Ideally, debayering happens as the last step in the processing chain. Short of that, get as far as possible before adding additional steps, particularly color & gamut related operations.

Familiarity with a set of tool is truly a valuable asset and sometimes needs to override another technically ideal path, but it can also be of great use to become 'familiar' with a tool that is dedicated to providing the absolute best quality from a given medium, in this case RedCine (RedAlert/Line... or Scratch for finishing).

cheers,

John T.

 

[Dj Joofa]

I thought we were talking about aliasing, not noise.

Graeme

I have not seen the footage from Leo. However, I read this comment from him:

Thanks, John,
The problems manifest themselves with motion, not the still frames.

which lead me to theorize that the issue is more complicated than just scaling a single image down as it is reported that a scaled movie is troublesome. A single frame scaled down might look fine, but a sequence of scaled down images displays some artifacting and we want to know why that happens, and to possibly remove it.

Scaling down a single image is a filtering operation using spatial filters. So if one uses a spatial filter, then of course, it has a neighborhood defined by its coefficients (otherwise it will reduce to decimation), and it "absorbs" the signal "fluctuation" from neighboring pixels because of its co-efficients. Certain type of filters will do better than others (e.g., should I use lanczos or a pure sinc function), but the issue persists.

The fluctuations in neighboring pixels are stochastic in nature, typically not deterministic. As I said that you rightly pointed out that one can use a spatial filter to suppress them to some extent by "spreading them over the neighborhood." However, typically, many times people will use a "constant" pre-blur" filter, this process is not adaptive, and does not give equal variation in different parts of an image. Additionally, the next frame has the noise values in different arrangement, and so on.

Both aliasing (as folded frequencies over a "mirroring" frequency) and noise typically manifest themselves as high frequencies, and actually it is the higher frequencies we are removing, in different dimensions. A low-pass filter, (sinc, lanczos, polyphase, etc.) tries to smooth an image by getting rid of spatial higher end frequencies. However, the temporal higher frequencies persist, and it is a moving sequence, and not a single image, therefore, they start degrading the user experience.

It is always good to start with a cleaner image before filtering / scaling if one can, and temporal filteriing can do that. It can be made adaptive also with not a huge effort.

For best results, a joint spatio-temporal approach will be very effective.

 

[John Tissavary]

I have not seen the footage from Leo. However, I read this comment from him:

In this case I don't believe temporal filtering will help. Motion enhances the visibility of the artifacting, but it's aliasing rather than noise.

I have posted about this several times, including methods of obtaining acceptable scaling results, but I still see this issue being talked about as if there were no acceptable solution. I'm not sure what the reason for that is.

I have had footage that scintillated terribly during downconversion to SD, including the sample of Leo's footage. I used the Lanczos, Mitchell, and Gaussian filters in both RedCine & Scratch to produce SD downscales that show no edge crawl.

Obviously it's fine if people want to come up with other methods to work this out, whatever works. But there is a solution, everybody with RedCine has it, and it works today.

regards,

John T.

 

[Dj Joofa]

In this case I don't believe temporal filtering will help.

Lunacie, I don't want to make this issue as a choice between what type of filtering should be done. The issue of temporal filtering is much more essential than just applied to scaling, etc. Just as with any high end camera it is difficult not to imagine an OLPF to reduce unwanted elements, it is almost unacceptable for a sequence of moving images acquired from a camera not to be routed through a temporal filtering unit, even if you don't want to do any scaling. It is just not the normal practise. Any deviation from this behavior should be view with the same token as the absence of OLPF causes in certain situations.

Therefore, if you are seeing motion artifacting in a sequence when you are scaling down (with the assumption that the camera did indeed provide the temporal filtering much earlier upstream), then you are left with no choice but to use spatial filtering as you rightly point out.

 

[Michael Schrengohst]

So Leo, after reading all of this -
what are you going to do?

 

[Leo Ticheli]

Update, no resolution yet...

Update, no resolution yet...

Sincere thanks to all for helping; this is where we are this morning...

Thanks to BJ at Custom and Teranex, I've got a Teranex Mini coming in for a demo next week. Custom has offered us a very attractive price, as they always do.

I've tried most, but not all, the suggestions offered in this thread and am continuing to test as time permits. We've been a bit busy, both in the suites and shooting, so I'm a little behind in doing everything I wish to try.

We did have to get a spot out, so we processed the 4K, 16x9 R3D files in RedCine to 720P and looped out of FCS, through a Panasonic HD deck's down-converter, to another Mac FCS. This produces acceptable results. I hope the Teranex will do an even better job.

I received an email response from my trouble ticket report from Kevin Stanley at Red with a suggested workflow, which involves going from FCS to Compressor, using DVD Group, Best Quality to the job/target. We are testing that solution now. Thank you, Kevin.

I would like to say that this is a problem that may not be immediately noticed, if the material is free of interference patterns. Shots of people with little movement may not blatantly show the problem; shots with lots of straight lines and movement will look terrible, even with cursory examination.

I think we are so overwhelmed by the beauty of the HD images, we may overlook the SD issue.

At the risk of redundancy, let me say again that I consider this issue grave. Not everyone will have a hardware solution available; there should be a completely straightforward HD to SD software solution, a one-button method.

In my opinion, RedCine should have a prebuilt SD output mode than produces stellar results. I'd settle for simply acceptable results.

Best regards and many thanks to all,

Leo

 

[Graeme Nattress]

I have some very pristine digibeta footage of a car, a bit like your shot. Not only is there aliasing visible on the grille, as the shot moves, but it's overall softer too.

Graeme

 

[George A.]

Dear Leo,

I completely agree. I hope RED is working on it.

 

[Leo Ticheli]

I have some very pristine digibeta footage of a car, a bit like your shot. Not only is there aliasing visible on the grille, as the shot moves, but it's overall softer too.

Graeme

For the past six or so years I've been shooting sheet metal with a Panasonic VariCam and the footage was regularly used in SD by simply spitting it out of the Panasonic HD decks as SD. It looks really good.

Of course it's softer than the HD, but it doesn't have the completely unacceptable aliasing of software converted Red footage, at least the methods I've tried.

Don't think I'm ungrateful for your help, Graeme! I think you guys are doing amazing things and I appreciate your taking the time from other tasks to look into this.

Best regards,

Leo

 

[Rob Gardner]

Thanks to Leo and everybody else working on this. I think this is a serious issue for anybody delivering television in SD. I'd hate to think it requires more hardware.
Rob Gardner