Issues with images posted so far [Archive]

Ruairi Robinson

09-01-2007, 11:23 AM

Congratulations to the red team for their success - it's a pretty amazing achievement. I've noticed some issues in the images posted, and I not much discussion of it (someone even said to FORGET about an image that blatantly had problems). I hate to be one to be negative here, but some of the people here seem to be in denial. I'd rather constructive criticism than blind optimism.

1) Clipping. Every image posted has pretty much demonstrated some really ugly clipping in hot highlights, and the ugly salmony hue in skintones. Makes it look like "big video" to me. So could people testing PLEASE expose to protect their highlights?

2) weird smudgy artifacts. These were visible in some of the stills from Crossing the line too. The redcode compression is variable right? is it possible in the future to work at higher data rates? 100Megs/sec instead of 27, for example. Is it necessary? Will it make a difference? Is this even what is causing this? I'm having a hard time at the moment believing the "visually lossless" thing, since... I have eyes.

3) the red and blue colour channels seem much softer than green. This seems to confirm what peope were saying about it not being full resolution 4k for all the colour channels (red in particular is REALLY, REALLY soft) This was blamed on unsharp masking somewhere here, but unsharp masking doesn't look like this.

4) some really blobby chunks of colour noise in low light pictures. I guess this is to be expected if you push it too far.

There were some pretty impressive images in the Crossing the line trailer (haven't seen the full thing, or even the 2k) so I remain optimistic. You'll notice I'm a reservation holder too. I want this to be great. Look forward to seeing more images.

Best,
R.

Tonaci Tran

09-01-2007, 11:40 AM

Cut the images some slack.. as last night was basically.. lets shoot some stuff so we can share with the world. Don't worry redcode raw is really elastic. Film will never be this clean. You can fine tune it for beauty or break it. I think you will being seeing more incredible footage today.. by us and the other REDs running in the wild.

Brook Willard

09-01-2007, 11:40 AM

I think any problems that have appeared thus far shouldn't really be worried about. The camera *is* capable of jaw-dropping, noise-free images... if it's noisy or otherwise "wrong," it's something we did wrong [not the camera]. While we have noticed a few things so far, I'd hold off judgment until some "proper" footage gets up online.

Ruairi Robinson

09-01-2007, 11:45 AM

Brook Willard

09-01-2007, 11:49 AM

I can tell you that the lights were literally thrown up last night. I walked in the stage, put up the first fader on the board and about 8 tungsten lamps came on. I put up a Kino and one other 650w for fun and that's it... all random lighting. I can also say that there's a lot more detail in the RAW files than in what ended up getting posted. Hold judgement. :)

We all know what good footage looks like on the cameras. Now we know what bad footage can look like. More to come.

Álex Montoya

09-01-2007, 11:56 AM

I am taking what I judge to be the best pic of yesterday, that's the "eyes" pic.

Highlights are ok, skin is not pink and I don't see many differences between the channels (given that the red channel is lighter, cause they are photographing human skin).

I reckon that many of the problems with the highlights come in the RED Alert! step. If you use these RAW conversion programs wisely you can extract much more detail from highlights.

Alex Boothby

09-01-2007, 11:56 AM

Ruairi I don't think the clipping we saw was an issue with "exposing to protect the highlights". After all many of the pics were shot well under and pushed in RedAlert. Rather I sadly do believe that this comes with the territory with digital sensors. Film just handles the roll-off quite nicely.

Rather than this being a deal-breaker I think we will all learn ways to protect against this - whether that involves shooting with diffused light, using low-con, soft or pro-mist filters, or post processing the highlisghts. In post I have been experimenting with some techniques to help here.

There is a learning curve here and I'm sure serious shooters will learn to navigate these issues and produce lovely images. Others will not put in the effort and produce questionable images.

BTW - your site is one of the funniest things I've ever read, and your work is too damn good to be funny. Keep being you.

Alex Boothby

09-01-2007, 11:59 AM

Also I belive Redcine will be equiped with a tool to gain an extra stop out of the highlights.

brandon herman

09-01-2007, 12:09 PM

We all know what good footage looks like on the cameras. Now we know what bad footage can look like. More to come.

the "bad footage" still looks pretty freakin good!

jbeale

09-01-2007, 12:17 PM

If your experience is with film, I think it is going to be an adjustment for setting exposure to get the highlights you're used to.

This is I think the same issue with analog audio recordists moving from tape to digital. At first, many people said that digital recording was no good, when the real issue was they had the input level set as they were used to for tape where you get peak compression "for free". But the level was too high for digital, and they got hard clipping. I know there were other issues with early digital recorders but few people would now say that digital audio recording can't do a good job.

Same thing with images, the digital sensor is clean down in the darker areas where film would be very gritty, but you don't have the top-end rolloff of film to save highlights, so the proper exposure point for most scenes is going to be different.

About resolution, if you use a standard Bayer pattern sensor (which I assume RED is using) you have twice as many green pixels as red & blue, so you'd expect a difference. If you look at the RGB planes of most any DSLR camera you will see this. However the exact resolution of the RGB planes in the de-bayered image may vary from that ratio depending on the specific de-bayer algorithm and the specific scene content. Personally, looking at the pics posted I do not see a real problem with them, for what little that's worth.

whachusay

09-01-2007, 01:01 PM

the "bad footage" still looks pretty freakin good! Exactly!

Graeme Nattress

09-01-2007, 01:08 PM

Jbeale, you're correct about the Bayer pattern used in the sensor. The demosaic algorithms are set up to extract full luma detail and as best as it can from the red and blue, and to keep all three channels spatially correlated. As I look at the eyes image, I don't see any significant difference in resolution in the three channels, which is to be expected for such an image. Yes, you can provoke red and blue to look different by using very strongly coloured objects, but that will always be the case in such a sensor.

Graeme

Zas

09-01-2007, 01:08 PM

It's also good to keep a healthy sense of perspective. How would 35mm have looked? How would a Genesis have looked? How would either of those look when it's someone's first time using them at night? At this point it's speculation, but I'd imagine that Red has both of them beat.

Graeme Nattress

09-01-2007, 01:11 PM

Not to mention, and I forgot to above, that all lenses have different resolutions at different wavelengths.....

Floris Liesker

09-01-2007, 01:14 PM

I'm kind of surprised the people from FXguide (who collected their #22 Red yesterday) said it's dynamic range is not quite as good as that of an F23.
Anyone else having an opinion about the F23 vs the Red?

In the images posted so far I also see a lot more noise than in the Milk Girls. As a matter of fact, the milk girls are still the best footage I've seen from the camera, better than CTL. This made me fear that the production models weren't up to par with Frankie. Later I learned the images were ISO 800 and higher, so that may have caused it. I am very curious about what dayshots will look like.

Graeme Nattress

09-01-2007, 01:34 PM

Well, as you know Milk Girls was shot outside in some nice natural, bright light..... I can tell you for certain that the software that processes the images has improved immensely since Milk Girls too.....

As for dyanamic range - unless Mike Seymour shot a Stouffer test wedge with both cameras, it's hard to compare just by eyeballing it. Mike actually says: "The camera may not have quite as much dynamic range as the new F23 from Sony" which is a pretty ambiguous statement in my reading of it.

Of course, the F23 is not 4k.....

Graeme

jbeale

09-01-2007, 01:46 PM

> In the images posted so far I also see a lot more noise than in the Milk Girls.

If you're talking about this: http://www.reduser.net/evin/darkBrook.tiff
it was apparently at 6000 ASA, so that's not exactly a fair comparison!

I don't know what the exposure settings were for http://www.tonaci.com.nyud.net/eyes.zip but that dark background has as little noise as I've seen yet (without being clipped flat). Any less noise and it would probably have contouring artifacts.

Speaking of exposure settings, having meta-data exported in the still frames would be a wonderful feature to have- I assume it will be coming eventually.

Graeme Nattress

09-01-2007, 02:01 PM

Yes, we'll work on getting the Metadata into the dpx files for post.

Graeme

Floris Liesker

09-01-2007, 02:01 PM

I know, it's all non-measured and purely gutfeeling stuff I was saying. That's why I am curious about 320 ASA shots.

Doesn't the noise floor of an image go 3db down as you resize it to half it's size using a normal bicubic interpolation? If that is the case, the red one could produce and image of higher dynamic range in 1080p than it would do at 4k, right? Like the viper's dynamic range decreases when shooting widescreen. Maybe that would be an even better way to compare the Red to the F23.

Graeme, what is your vision on that?

Graeme Nattress

09-01-2007, 02:11 PM

Yes, if you do a downsample of the 4k to 2k, you'll lower the noise floor.

We could also go in and do some special noise processing that would lower it still further, but most people seem to like things relatively un-processed.

Graeme

Floris Liesker

09-01-2007, 02:16 PM

Yes I always like the look of downsampled material, even from my old DSR-300p. It makes the image much cleaner and the video-sharpness is as good as gone. For that reason I think I'll use the 4k downsampled to 2K a lot.
And the fact that I cannot play 4K anywhere does play a little role too. :)

Graeme Nattress

09-01-2007, 02:22 PM

The way I've got it coded now the 2k has the very same pixel feel as the 4k. If you turn off the output sharpness option in REDAlert, you don't have to downsample to avoid the added sharpness of video :-)

Graeme

Gavin Greenwalt

09-01-2007, 02:24 PM

One problem I'm seeing in many shots is the bokeh is not very color neutral. For instane in the 2k shot the bokeh has a distinct cyan/green fringe. Almost every shot seems to have some sort of oddly colored bokeh. I know some is always to be expected but to my eyes it looks like more than what I'm used to seeing.

Alvise Tedesco

09-01-2007, 03:05 PM

(Ruairi I don't think the clipping we saw was an issue with "exposing to protect the highlights".) (Rather I sadly do believe that this comes with the territory with digital sensors.)

In post I have been experimenting with some techniques to help here.

Hi Boothba. Can you share some helpfull info on that?

Floris Liesker

09-01-2007, 03:15 PM

I think, but I'm not completely sure, that Boothba means that the shots weren't exposed to protect the highlights. Some parts are actually clipping like hell, also this desktop screenshot of tonaci. An exposure level of 0.59 doesn't mean much to me, maybe that's what causes a part of the left guy's head to clip. But I fear it was just shot like that and the sensor got overexposed. I am curious about how much info would be in the RAW image. In other words: where in the workflow did this clipping occur?

John Wee

09-01-2007, 03:15 PM

If those are BAD footage, then DAMN I CANT WAIT TILL THE RED TEAM FINISHED TWEAKING THE CAM !

b e n t o n

09-01-2007, 04:22 PM

So there are 25 cameras out there now, and only one of them is sending reports back to us at RedUser?

Alex Boothby

09-01-2007, 04:26 PM

I think, but I'm not completely sure, that Boothba means that the shots weren't exposed to protect the highlights.

Nope - I don't think they were over-exposed. On film highlights blow and on digital highlights clip. It is the job of camera manufacturers, shooters, colorists and compositors (and often makeup artists) to make that blowout/clipping look as natural to the eye as possible.

Alex Boothby

09-01-2007, 05:03 PM

Hi Boothba. Can you share some helpfull info on that?

Sure - just add grain! haha.

Well, my particular goal is a achieve a more filmic image (not everyone's goal). Digital / video can suffer from a certain harshness with clipped highlights, overly crisp color separation and chromatic aberrations. I'm currently working on a job shot on the Phantom which a suffers from harsh chromatic aberrations - when certain areas achieve focus, the picture becomes infected with obnoxious red, green and blue pixel clusters - like fuzzy pokadots. A fast solution was to use a 'chroma only' blur using the Safire blur spark. This does not blur the image, just the chroma information and looks a bit like a coat of base makeup applied to an actor. I can't post the Phantom footage, but the solution was dramatically better.

On this latest Red footage I see nothing as egregious as the Phantom stuff - however I do note some clipping and mild chromatic aberrations. The skin tones in a few shots seem to have a larger range of color variance than film. Red noses and lips, green hues etc. I think it is the sharpness of the variance than reads a bit digital to me. This may just be the lighting or color grade but I tried the chroma blur trick - results below.

This is a subtle application - you'd have to pull it into photoshop and play with levels to see the difference. Mostly it is a smoothing out of skin tones - a digital application of make up I suppose. It also reduces the CA in highlights. BTW this is very fast and easy to do.

Regarding clipping - I've darkened and blurred the highlights a bit but this is NOT the real solution. On film jobs we often do ALT transfers or scans to recover blowout areas like skies. It then becomes the compositors job to mix the alt pass with the beauty pass (for skies I often mult, through a generously blurred highlight key). This may or may not help with Red footage. Depending on the sophistication of the highlight recovery tools in Redcine, one can always grade and export an ALT highlight pass (very dark) in Redcine, for further manipulation in comp. Standard stuff on film jobs.

Alex Boothby

09-01-2007, 05:05 PM

Here's a couple of silly examples. Sorry I made the pics too small. First adresses the CA in button and chest highlights. Second smooths skin tone variance in face.

Subtle fix for a subtle problem.

Alex Boothby

09-01-2007, 05:16 PM

The skin tones in a few shots seem to have a larger range of color variance than film. Red noses and lips, green hues etc. I think it is the sharpness of the variance than reads a bit digital to me.

BTW: I wouldn't have it any other way. For green and blue screen work this 'enhanced color separation' (compared to film) actually produces superior keys. This is far more important to me. This follows the theory: "start with a clean, deap image (a "fat" digital negative), and make your colour and aesthetic decisions in post"

I think the big thing with Red (like film) is that people will have to embrace the importance of the color grade / composite. That is where much of the look is set (Think "300" or "O brother"for example). This is not a video camera - although Red has made it pretty darn easy for us to shoot and upload real fast.

Graeme Nattress

09-01-2007, 05:20 PM

Yes a good grade is key. Lens artifacts have always been with us. Digital sensors, with their precision and cleanliness task lenses as never before!

Graeme

explosive

09-01-2007, 08:47 PM

2) weird smudgy artifacts. These were visible in some of the stills from Crossing the line too. The redcode compression is variable right? is it possible in the future to work at higher data rates? 100Megs/sec instead of 27, for example. Is it necessary? Will it make a difference? Is this even what is causing this? I'm having a hard time at the moment believing the "visually lossless" thing, since... I have eyes.

I've wondered the same thing. I don't think you can compress something without losing some detail, wavelet codec or not.

I am no expert in compression technology, but I have a feeling it wont be as visually lossless as we would have hoped.

Not to bag on the codec, it seems excellent, but I tend to agree with you on that one.

btw - wicked trailer for your movie. I pissed myself with that opening shot. Many kinds of awesome.

Gavin Greenwalt

09-02-2007, 12:02 AM

On film jobs we often do ALT transfers or scans to recover blowout areas like skies. It then becomes the compositors job to mix the alt pass with the beauty pass (for skies I often mult, through a generously blurred highlight key). This may or may not help with Red footage. Depending on the sophistication of the highlight recovery tools in Redcine, one can always grade and export an ALT highlight pass (very dark) in Redcine, for further manipulation in comp. Standard stuff on film jobs.

Completely agree, it really comes down to people not understanding that footage is always a starting point if it's digital not a result. I actually did 4 alt-passes off of the uncompressed tiff for this 6000 iso grade.

http://www.reduser.net/forum/showpost.php?p=76096&postcount=6

The largest problem I'm noticing in all RED footage and I don't know if it's the lighting but the saturation and hue shifts in the skin tones seem to be off the charts even in what otherwise appears to be a single light source shot.

Brook Willard

09-02-2007, 01:22 AM

The FrankenBrook grab had metal halide overhead, sodium vapor about 25 feet away and a light panel pointed in my general direction [I think]. It was also freakin' *dark* :)

Andrew Benz

09-02-2007, 01:33 AM

The FrankenBrook grab had metal halide overhead, sodium vapor about 25 feet away and a light panel pointed in my general direction [I think]. It was also freakin' *dark* :)

Hey Brook I was having fun with you. :tongue: We all very much appreciate your hard work and good will-- though that was one hell of a moody pic -- You will also notice that you were also accused of being a graduate of ... "the Handsome Boy Modeling School" LOL... here is a reference...http://www.handsomeboymodelingschool.com/

The World Has Gone Mad could be your theme song... Ninjas need theme music...:ph34r: :shiftyph34r:
Thanks again...

Roberto B

09-02-2007, 01:52 AM

the ladies here are saying that brook is awesome.. even on dark

Andrew Benz

09-02-2007, 01:55 AM

the ladies here are saying that brook is awesome.. even on dark

See... or is that si'... These are the ways of the Handsome Boy Graduate...

Brook Willard

09-02-2007, 02:03 AM

The World Has Gone Mad could be your theme song... Ninjas need theme music...:ph34r: :shiftyph34r:
Thanks again...

Our theme song is silence... :ninja:

the ladies here are saying that brook is awesome.. even on dark

Oh dear. Now I just need to grow a glorious HBMS moustache.

You guys are all crazy. :)

Seung Han

09-02-2007, 02:22 AM

sure sure

all the other stuff I'm sure will be fine

I'm mostly concerned with clipping. Perpetually. It's a deal breaker for me.

The clipping in many of those shots is worrisome. Even off reflections. Very videoy...

I took some frames in AFX and the degree of malleability is astonishing, however.

Brook Willard

09-02-2007, 02:25 AM

Roberto B

09-02-2007, 02:52 AM

You guys are all crazy. :)which guys?.. ladies! gorgeous ones.. :love:

Seung Han

09-02-2007, 02:56 AM

If memory serves, the only clipping that happened was in the straight, unprocessed screenshots. That means that they were overexposed by several stops in post... but the data is *actually* there. When we actually spent some time working with the pictures, all overexposure disappeared.

It's like the first time you put a negative in a film scanner. It'll look like absolute overexposed crap until the operator spins a few knobs. Those screenshots were before we spun *any* knobs.

Good to hear.

The two motion shots I've seen so far are amazing.

It'll be interesting to see some before and after shots of raw footage and then see what some talented people in post on these boards do to shape them up...

Gavin Greenwalt

09-02-2007, 03:10 AM

The FrankenBrook grab had metal halide overhead, sodium vapor about 25 feet away and a light panel pointed in my general direction [I think]. It was also freakin' *dark* :)

Haha glad to hear. There were green, blue, cyan, yellow, orange and red hues which were devlish to seperate and seemed to be coming out of nowhere.

James T Mather

09-02-2007, 09:54 AM

sure sure

all the other stuff I'm sure will be fine

I'm mostly concerned with clipping. Perpetually. It's a deal breaker for me.

Looks like you'll still need a Dp Ruairi. :biggrin:

Nils Ruinet

09-02-2007, 10:06 AM

If memory serves, the only clipping that happened was in the straight, unprocessed screenshots. That means that they were overexposed by several stops in post... but the data is *actually* there. When we actually spent some time working with the pictures, all overexposure disappeared.

It's like the first time you put a negative in a film scanner. It'll look like absolute overexposed crap until the operator spins a few knobs. Those screenshots were before we spun *any* knobs.

Could you please post these, so we can see the difference ?

Thanks :)