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 :)

Scale and Debayer settings - some examples

This is the test frame...

image.jpg
 
and here are the different debayer settings...

debayer.png


The footage was rendered to a TIFF at 1920x1080. This is an enlargement of a small section of the frame. The uploaded web image you're looking at is a PNG 24 which a lossless compression, the 'jpeg' looking artifacts in the 1/4 debayer are in fact debayer artifacts.

1/2 debayer is in fact '1/2 high'
 
This is a comparison of noise reduction from a different shot. This is a car number plate at night, from the corner of a frame that is basically completely black. I've adjusted the levels in photoshop in order to see the differences.

It's hard to judge from a still image, the difference looks subtle. In order to see what's going on, I isolated the blue channel. It looks like the noise reduction algorithm blurs the blue channel mostly, some on the red channel and nothing on the green channel.

noise.jpg


* note - I adjusted the levels of this test in photoshop so that the noise was MORE visible. That way, we can see what the noise reduction is actually doing. It's not like you would try to bring up a shot this much in a real world situation.

Without bringing up the levels, the noise was too hard to see in a static image.
 
Thanks for the comparison. Very interesting. As far as resize filters... which one do think it is best fro general use? They all seems so similar. :-)

Sorry for my ignorance, but that's in Redcine right?

I've only used Redalert. Can you also check Max Noise reduction on Red Alert as well (that's the Chroma-noise tab right? The test of the noise is the one that it is more clear to me.
 
Gunleik Groven said:
What was the kelvin/iso settings on the noise-images?
Click to expand...

I actually raised the exposure in photoshop using the levels layer. I raised it a LOT... because the idea was to see what the noise reduction was doing. This portion of the frame with the number plate was almost pure black...
Obviously you would never actually try to bring a shot up as much as this. I'm basically increasing the noise in photoshop so that we can see what's going on.

Kelvin, was most likely on default.
 
Impagliazzo said:
Is it guass or gauss? Confused.

But Lanczos and G(au)uass look very sharp, not sure about if those introduce artifcats or not.
Click to expand...

I can't comment on my spelling... it's usually due to speed of typing, or possibly slowness of brain, depending on the time of day :)
 
muffinny said:
Thanks for the comparison. Very interesting. As far as resize filters... which one do think it is best fro general use? They all seems so similar. :-)

Sorry for my ignorance, but that's in Redcine right?

I've only used Redalert. Can you also check Max Noise reduction on Red Alert as well (that's the Chroma-noise tab right? The test of the noise is the one that it is more clear to me.
Click to expand...

These renders were all done in Monkey Extract... (shamless plug :biggrin: ) but the settings are the same as Red alert. So yes, that's the same noise reduction.

I personally like the sharper filters for resize. The wold 'roll' on the clapper is sharp in real life, so I prefer the render to match that, but it's all personal taste.
 
For info, we're currently finishing a Red shot doc for the BBC and we used Lanczos to resize. The result are stunning and we've had no down conversion problems.

Will post some samples when I get a moment.

BBC here in the UK are showing Wallander, a new drama shot on Red in Sweden and it looks gorgeous!

Scott
 
Scott Brown said:
For info, we're currently finishing a Red shot doc for the BBC and we used Lanczos to resize. The result are stunning and we've had no down conversion problems.

Will post some samples when I get a moment.

BBC here in the UK are showing Wallander, a new drama shot on Red in Sweden and it looks gorgeous!

Scott
Click to expand...

Yeah, we chose Lanczos too...
 
There is also subtle changes in black levels across the different resize filters - looking at Gauss and Lanczos which are the sharper of the two. The differences are noticeable in this resize portion of the frame, what are your thoughts on the full frame in motion? These types of frames renderings would be great for documentation on any of the RED tools being offered.

Thanks for doing all of this work! I was going to some of that over my Christmas vacation...

Michael
 
MichaelP said:
There is also subtle changes in black levels across the different resize filters
Click to expand...

The way many of such filters are typically implemented results in the violation of the certain principles of interpolation. If somebody is using sinc-derived filters such as Lanczos, or otherwise non-time/space limited filters (non-compact) you can be certain that the intensity of uniform areas may not be preserved all of the times. It does not have to be a resizing operation. One can rotate an image with these filters and still see an intensity shift. Unfortunately, even resampling to a finer and finer grid may not make the problem go away.

Is there a solution to avoid that? Of course, there is. But that will mean graduating to approaches that are different from a straight-forward implementation of apodization-based filtering.
 
If I do a 4K dpx, does the algorithem make a diference?
I would think not if it is a scaling tool, but if it is a Debayer tool I gues it would.
I mean if it is a tool that "creates a RGB pixel" then i will affect the 2K dpx.
But if its not, and only works after the debayer, I gues it doesn't make any diference.

Have anyone seen any diference in aliasing with the diferent algorithem?

/carl
 
You must log in or register to reply here.
Back
Top