Odyssey7Q + FS700 - 4K Modes

I'm asserting that:

1. It appears we can get better than the arbitrary 80% quoted but not backed by science (math and/or empirical tests). I believe it's perhaps 94% or more from empirical tests and top modern deBayer methods.
2. If we use a high quality deBayer, we can get excellent RGB (444); when converting to YCbCr and converting to 422 and 420 there will be a (statistically) significant drop in color fidelity that cannot be recovered in post (since the original Bayer information is lost).

The Malvar, He, Cutler (Microsoft) algorithm is not the top method (dates back to 2004); ACR and others are now better.
Top, original, bottom, converted to Bayer then deBayered:
deBayer.jpg
Does the bottom image look 80% of the top? Looks over 90%, right? It's not perfect, but is very good. Modern methods are even better.
 
Last edited:
If you ask me, the bottom image looks to be about 70%.
Disclaimer: The opinions expressed in this thread are strictly my own and not based on scientific fact, rather based on a WAG.
 
jcs said:
Does the bottom image look 80% of the top? Looks over 90%, right? It's not perfect, but is very good. Modern methods are even better.
Click to expand...
Man, I'm gonna look like a jerk for asking this, but -- is this a serious question?

That bottom image looks like ass. It looks as aliased and color-fringed as something that was shot on an original Canon 7D. In 720p mode. There's orange and blue color fringing absolutely all over it.

I would be furious if my camera did that to my footage.

That isn't what 4:4:4 should look like. That looks more like 4:1:1, on its best day.

Sorry jcs, we're not going to see eye to eye on this. That is not close, that doesn't look good -- I mean, if that was a JPG saved at 10% size, then I'd say, yeah, that looks pretty good for 90% compression, well done guys. But empirically, judged on its own, does it look "good"? No. It looks, seriously, like it was shot on a 7D in 720p mode. It looks pretty bad. It's recognizable, certainly -- and in terms of luminance detail it looks pretty good, but in terms of color accuracy (which is what 4:4:4 is all about) it looks really bad on every edge.
 
Note, it took me so long to type out that response and to look up some supporting info, that olgagroza swooped in and posted before I could, and demonstrated what I was seeing in a much more effective way.

I repeat -- getting full resolution from a bayer sensor, and also getting 4:4:4 (or anything close to it) is not possible.
 
Remember- the 80% was for luma. The gripes with the Marval et al algorithm are with chroma. In any case, not too shabby for an 'ancient' 9 year old method (which uses every single Bayer pixel for luma and combinations of colors for chroma (counter intuitive)). When I scaled the original down to 80% with nearest neighbor then back up with bilinear or bicubic, it was clearly lower resolution.

The AMaZE and DSB algorithms in Rawtherapee look very good (source code is available). Once I find a way to convert TIFF/PNG to a RAW-Bayer format (cr2, DNG, etc.), I (or anyone) can post examples of what modern deBayer methods can do. I don't see the Marval et al kind of artifacts in ACR or Rawtherapee: looks excellent.
Again, I believe we can store useful information in 444 from Bayer vs. just tossing it with 422 or 420 (not stating 100% reconstruction, but useful reconstruction for 444).

Olga- thanks for the A/B animation- very cool.

I'm moving this weekend, so won't have time to explore this for a while. However the plan is to rasterize the ISO 12233 chart to RGB and convert directly to a Bayer format (taking away any optical/camera variables). We can then compare methods of Bayer reconstruction and find where any limits are with current algorithms. High quality real images with no source artifacts can also be used.
 
Last edited:
Hello folks,

The "popular compressed codec" Convergent Design is going to use is probably Prores, in the 4:2:2 10 bit flavor.

Okay, now my stupid question : don't you think there would be a benefit to be able to record Prores4444 instead ?

Mainly for the 12 bit color depth, as I understand there is no consensus concerning 4:4:4 sampling and that we may not need an alpha channel (I'll appreciate if someone could light my lantern on this last point, thanks)

Does the OD7Q even have enough power to handle this codec ?

Cheers,
Steph.
 
did that 4k footage CD was talking about on facebook ever get posted? They said there were shooting some the other day.
 
You must log in or register to reply here.
Back
Top