Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

CineForm News

Status
Not open for further replies.
We have two concerns with this.

1. Graeme has done side by side testing and the DPX files created by this app are inferior to DPX files from RED Alert! and REDCINE. If you start with a demosaic that isn't effective and then add compression, the results will get progressively worse down the line.

2. We aren't done with our code yet. We have asked Cineform to please wait until we are finished before "helping" our customers. Our additional fear is that we will be blamed for a code change when this app stops working.

We have our hands full. Rob is almost finished with our Command Line app that works the same as RA and RC. We don't want to have our customers disappointed with the finished product and have it be due to a "workflow issue".

I hope that Cineform will show just a bit more restraint in releasing an app like this. Please wait until our code is finished, then take the time to make sure the result is worthy of the footage.

Graeme will post the side by side comparison of this and RED demosiac. If you are impatient, please do your own comparison and post it here.

Jim
 
Jim, I think I understand where you are coming from on this, but honestly I think you are not giving the customers quite enough credit. Back when MiniDV cameras were new, there were many different DV codecs (I guess there still are). MainConcept had one, Microsoft had one, Canopus had one, etc. They had different characteristics, which were thoroughly analyzed by many different people, examples were posted in forums, and people were free to choose the tool that they preferred for the job at hand.

Needless to say the first thing I did with the new Cineform tool is test it with my own footage and compare. I'd be pretty stunned if there's a single imaging professional who would do anything else.

(For those who haven't got a camera yet, and if artificial test patterns are your thing, I've got some R3Ds ready for your evaluation: http://www.bealecorner.org/red/test-080108/index.html )
 
Demosaic Comparison

(Note PDF available here http://reduser.net/Demosaic Comparison.pdf - images in it are png to avoid the JPEG compression in the images below)

This is a comparison of Bayer Pattern demosaic from CineForm’s R3D2DPX, compared to the current RED decode. Due to colorimetry differences between the CineForm DPX converter, and REDAlert!, I used levels in Photoshop to match the tonalities as best as I can. Photoshop was also used to analyse and produce the comparisons, splitting out the RGB channels for easy viewing, with CineForm on the left, and REDAlert! on the right. I’d encourage you to zoom in on the images as that will help you see things more clearly.

Image 1: Base Comparison
View attachment 2688
This is the basic decode from Cineform, compared to RedAlert. As David Newman points out, their demosaic algorithm is optimized towards detail. All other things being equal, this is reasonable, but personally, I like to take a more balanced approach, looking at other qualities as well. Looking closer, we see that we're trading sharpness for pixel level artifacts, especially on edges, where what looks like a characteristic Bayer artifact, the zipper, appears on the CineForm image. In comparison, the RED image is smoother, more continuous, and less blocky in the red and blue channels.

Image 2: Matching*Apparent*Sharpness
View attachment 2689
Adding a small amount of unsharp mask in Photoshop balanced the sharpness between the two algorithms, allowing us to further our comparison. The RED image reacts well to a small amount of sharpness.

Image 3: Adding Too Much Sharpness
View attachment 2691
Although adding excessive sharpness is not something you'd really want to do to an image, it does exaggerate image artifacts. It's a useful test to help us see what is going on. I added, unsharp mask radius 0.6, amount 500%, tolerance 0 to both images. Now, even without zooming in we can see how the RED image stands up very well to this stress test.

Image 4: Adding Saturation
View attachment 2690
Going back to our original unsharpened images we can see the effects of adding sharpness. The RED demosaic takes great care to ensure that the resulting RGB colour channels are co-sited. This is very important with Bayer Pattern sensors as their very nature means that the red, green and blue pixels are not spatially co-located in space, but are offset from each other in the Bayer Pattern. Because adding saturation exaggerates the difference between colour channels, artifacts are created when they are not co-sited. We can especially see the effect of this where the jacket logo becomes nearly unreadable on the CineForm demosaic red channel.

Image 5: Combinging Sharpness and Saturation
View attachment 2687
The RED image behaves essentially as you'd expect, whereas the CineForm demosaic now shows a lot of Bayer artifacts.

From the above you can see why I designed the RED demosaic as it is. By changing numbers in the algorithm, we can increase the apparent sharpness, but only by introducing very similar artifacts to what we see in the Cineform image. I believe that image robustness is important, and that the colour channels should appear co-sited so that even after adding an excessive amount of sharpness and saturation, the image does not break up.

Graeme
 
jbeale said:
Needless to say the first thing I did with the new Cineform tool is test it with my own footage and compare. I'd be pretty stunned if there's a single imaging professional who would do anything else.http://www.bealecorner.org/red/test-080108/index.html )
Click to expand...

If your results show different results than Graeme's, we'd love to see them. You are free to do whatever you like. We just want you to know what you are choosing.

Jim
 
Absolutely, you should do your own testing, of the whole work-flow. We have we have going through many demosaic type throughout the years, we even have a plugin API for those who wish to create there own. But then reality sets it, real world work-flows are impacted by many factors, quality, performance, easy of access, tools compatibility, output resolution (4K at 2K or 1080p.) In the past we had more advanced demosaic, and found diminishing returns for compute time versus final output quality. Many features have gone through that demosaic at 2K (with stesses it more than 4K), including a film at last years Cannes film festival. You can stress out any element of any work-flow and highlight a weakness using unrealistic post filters. That is easy. Speed and compatibility are some of your work-flow weak spots. I sure you are working on it. As we are to. We think choice is a good thing. Users can go either why, if you let them choose.
 
I have only done a very preliminary analysis so far, but I have seen what Graeme has also. Briefly; with sharpening, or even without sharpening if your scene contains high-frequency color detail, the R3d2dpx utility shows a variety of pixel-level artifacts that Redcine output does not.

However, if you shot 4k and your deliverable is 2k or lower, I'm not sure how much practical difference the debayer algorithm makes. Simple command line utilities are tremendously useful and I can do things with it (for low-res output) that Redcine cannot do right now, on my system anyway. I look forward to the equivalent Red utility when that becomes available.
 
David Newman said:
In the past we had more advanced demosaic, and found diminishing returns for compute time versus final output quality.
Click to expand...

This is a fundamental difference in our thinking...

The good news is that there is a choice.

Jim
 
jbeale said:
I have only done a very preliminary analysis so far, but I have seen what Graeme has also. Briefly; with sharpening, or even without sharpening if your scene contains high-frequency color detail, the R3d2dpx utility shows a variety of pixel-level artifacts that Redcine output does not.

However, if you shot 4k and your deliverable is 2k or lower, I'm not sure how much practical difference the debayer algorithm makes. Simple command line utilities are tremendously useful and I can do things with it (for low-res output) that Redcine cannot do right now, on my system anyway. I look forward to the equivalent Red utility when that becomes available.
Click to expand...

Totally agree to both paragraphs. Issue in real world shooting conditions high-frequency color details happen less often than than high frequency luma detail changes. Bayer has this issue that while every cell supports luma information, the same is not true for chroma. So there is a sliding scale for how luma detail you get while limiting chroma aliasing. The different is simply were on that scale you want to be. Now if you would like a softened more filter image they should ask, as that tool was only created over the last couple of weeks, we want feedback (on all factors.) Of course if you are outputting 4K at 2K, then there is no issue.

Relax, have fun with the choice. Once Red has tools that work as well, when we will happily have our customers use, or we will make something faster still. ;)
 
It seems like there really should be something on the cineform tool that says that for speed it gives up quality, because looking at those samples the quality looks pretty low.

I know Red made it very clear that the super fast (realtime?) quicktime decode is of lower quality than the redcine debayer, do they use the same methods to extract the info?

I can see someone going out and using this tool and getting inferior results and blame the camera, which doesnt do RED or CINEFORM any justice.
 
Graeme has spent all his time to extract the nth degree of quality from RED footage. Now he and Rob are working on speed. Results of their efforts will be seen soon.

It is hard for us to watch our customers give up that quality for speed. Our goal for our customers is good AND fast, not just fast. I am sure that Cineform has the same intentions for the future. When our code is finished, we hope Cineform will take another pass at it.

Jim
 
reality,

Please try for yourself, Graeme's analysis do not reflect the experience with typically imagery. Of the dozen of so beta testers, no one complained of quality. Once you try some of your image, please comment on areas you seek improvement.
 
Are you suggesting that Graeme's images are not real? I've always trusted Graeme. Maybe the beta testers didn't know what to look for?
 
Graeme used the .r3d file that I posted a couple of months ago. Anyone can do the same test with the same footage.

Jim
 
We have stated that Build 14 is a landmark build. It has many updates and changes to the code. Maybe Cineform can take another pass at this when we release it? I'm sure they will want to improve the demosaic while maintaining the speed.

It is in everyone's best interest to have as good of quality as possible in any format that will be used for RED footage. We recognize the customers need for options. We also recognize that our customers deserve the best quality.

Jim
 
Jim,
I know RED is your 'baby', but sooner or later you'll have to realize that it is growing up, and that it will make its own friends, with which you may not always agree, or find to your liking.
Eventually, it will be on its own, and won't be able to rely on its paternal protection.

:biggrin:
 
David Newman said:
Please try for yourself, Graeme's analysis do not reflect the experience with typically imagery. Of the dozen of so beta testers, no one complained of quality. Once you try some of your image, please comment on areas you seek improvement.
Click to expand...

Alright, I had our post guys try it on some footage shot earlier for a sheet metal job and passed frames of both to our vfx supe, post guys and colorist.

since we're currently posting some of this footage, I was able to get the footage up on a projector in a grading suite. Redcine and Redalert looked the same but Cineform looked very different.

the response was that cineform might be ok for editorial but not usable for final quality work and not acceptable enough to pass the network QC guys. the colorist said it felt like he was working with a dub that had dubbed.

i like the idea of having choices but can't take the risk. if someone spots a problem and realizes we did a lower quality output, my ass is grass.
 
This cineform de-mosaic is a no-go for me. You can can't anti-alias aliased footage. You can supersample anti-aliased footage. I like the Cineform compression codec though so I wish I could capture in Cineform, edit in cineofrm and de-mosaic in REDCine.

I'm getting stuff that looks a lot like the early pumpkin footage.

All the same I like where this is going.
 
Pig said:
Jim,
I know RED is your 'baby', but sooner or later you'll have to realize that it is growing up, and that it will make its own friends, with which you may not always agree, or find to your liking.
Eventually, it will be on its own, and won't be able to rely on its paternal protection.

:biggrin:
Click to expand...

I fully understand your point and agree that RED will take on its own life... it happened with Oakley. But it is hard to swallow seeing a step down in quality when we have worked so hard to get here.

Jim
 
I stick with REDCINE and REDALERT.

Cineform is very OK, but shows some weak points.

Speed is not an issue in my work.

I work in a studio and not used to "cook" a "fast food".

It's not good for my "health".

Highest quality is my goal.

"Cucina alla Mamma".
 
Status
Not open for further replies.
Back
Top