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 :)

steps to avoid Noise and Softness

Ok, some of you guys need to back off the user error thing. I have a R1 and a Scarlet and there is definitely something going on here. Side by side tests with the same lens and settings are completely different. Now some of my shots are great with Scarlet but some are just unusable. It seems to be completely random too. I can literally pan the camera over on an exterior daylight shot two feet and get completely different results in noise. It's pretty weird. I can do a daylight shot for b roll and it's a snowstorm but an interior is almost noiseless. I've sent some samples to tech support and I'm waiting on a reply.

Like Carey, I've never had one speck of noise in my R1...it's rock solid. I have a great lighting set up and I know what I'm doing. All of my Directors are seeing the same thing. I think this is happening to certain cameras.

The difference between RED and every other camera company is that they will make it right when they figure out what's going on.
 
Whether or not there is something more than user error here will take some time and effort to determine. My eyes are open to the possibility, but I think our attentions are in the right place for the larger portion of cases. If the determination is other, I have no reason to believe that Red would shirk responsibility. If you have a legitimate reason to say otherwise let's hear it.

Ryan Nichols said:
Ok, some of you guys need to back off the user error thing. I have a R1 and a Scarlet and there is definitely something going on here. Side by side tests with the same lens and settings are completely different. Now some of my shots are great with Scarlet but some are just unusable. It seems to be completely random too. I can literally pan the camera over on an exterior daylight shot two feet and get completely different results in noise. It's pretty weird. I can do a daylight shot for b roll and it's a snowstorm but an interior is almost noiseless. I've sent some samples to tech support and I'm waiting on a reply.

Like Carey, I've never had one speck of noise in my R1...it's rock solid. I have a great lighting set up and I know what I'm doing. All of my Directors are seeing the same thing. I think this is happening to certain cameras.

The difference between RED and every other camera company is that they will make it right when they figure out what's going on.
Click to expand...
 
Ryan Nichols said:
Ok, some of you guys need to back off the user error thing. I have a R1 and a Scarlet and there is definitely something going on here. Side by side tests with the same lens and settings are completely different. Now some of my shots are great with Scarlet but some are just unusable. It seems to be completely random too. I can literally pan the camera over on an exterior daylight shot two feet and get completely different results in noise. It's pretty weird. I can do a daylight shot for b roll and it's a snowstorm but an interior is almost noiseless. I've sent some samples to tech support and I'm waiting on a reply.

Like Carey, I've never had one speck of noise in my R1...it's rock solid. I have a great lighting set up and I know what I'm doing. All of my Directors are seeing the same thing. I think this is happening to certain cameras.

The difference between RED and every other camera company is that they will make it right when they figure out what's going on.
Click to expand...
Ryan - you sound like a sophisticated user, and your issues don't sound like they are fixable by proper Black Shading, lighting, etc. Agreed.

It would be useful if there were a way to limit the variables.

For users with problems, to shoot the same calendar cover available at Barnes & Noble, at the same lighting (daylight), with sensor temperature at the same temp as black shade temp, 3 stops above wide open, and all identical settings.

I'm probably dreaming that it's possible - to eliminate the variables, but ... setting up some kind of standard test would seem useful to understand what the problem is - especially for the new users who, unlike you, are making mistakes.

If it's a RED problem (hopefully this will be the minority of users), everyone benefits if we get to the bottom of it sooner than later, and can give them data on the users who are affected by issues caused by the camera.

What do you think? Have any ideas? I bet yours are better than mine!

BTW - Are sensor temperature and last black shade temperature included in the meta data?
 
Last edited:
No that’s exactly what I was trying to say, they will make it right. I’m confident in that. So all is good there.

Perhaps you are right about in some cases it is user error. I just felt I needed to speak up about my experience.

I would just find it hard to believe that if I were doing something wrong that I would have such different results with the two cameras and even within the Scarlet itself. If I have any black at all in the image typically it shoots that goal post up to about 80% and wham - snowstorm. The same image on the R1 and the goal post is almost nil with zero noise. I know that optimally you want to overexpose all the sensor to the right but sometimes that is just not possible and I think there is no reason I should have to when I don’t with the R1.

I’ve done all the tricks with the black shading and have mixed results still. Overall I’d say the percentage of usable shots went up but it’s still way too often. I’ve tried to do it every time I get it booted up to the correct temperature...but there will be many times that I don’t have the time to do that.
 
Last edited:
I honestly don’t know if that’s included in the metadata or not. I think the black shading settings are however.

I think you may be onto something with your idea. Whatever is the solution it needs to be standardized with a clear set of instructions. It would have to be something very simple that lighting kits won’t make a difference.

Also, I always put CTB on my lights. The image on both the R1 and Scarlet seem to grade so much easier with the blue channel activated in that way. That also reminds me that the noise I’m seeing is white sensor pattern noise NOT blue noise.
 
Ryan - when you are speaking to RED about this, see if you can find out what kind of testing each camera gets?

Is there testing done as far as noise levels of the sensor over a range of light intensities (your problem increased when panned into exterior daylight)?

Semiconductor wafers don't generally get 100% yields for all the chips on the wafer. It's certainly possible that there's some variance that they are not detecting or testing for.

If we can put our minds together, root out the users who are having user errors, and come up with some tests that would indicate other non user-error problems, we all benefit.

Having to leave the camera at home for a re-shoot doesn't benefit you, or RED, or me (as I want them to make a lot of money to create even better cameras and upgrades).
 
I’m with you on that. I’ll try to find out for you.

No, my problem is throughout every scene I’ve shot no matter the location, with of course some solid shots mixed in there. I’m just using daylight exterior to illustrate the point that in that circumstance, that would be the least likely place I would expect to get a noisy shot.

These are all normal settings too. 1/48th, 800 iso, 24 frames.
 
Ryan Nichols said:
I honestly don’t know if that’s included in the metadata or not. I think the black shading settings are however.

I think you may be onto something with your idea. Whatever is the solution it needs to be standardized with a clear set of instructions. It would have to be something very simple that lighting kits won’t make a difference.

Also, I always put CTB on my lights. The image on both the R1 and Scarlet seem to grade so much easier with the blue channel activated in that way. That also reminds me that the noise I’m seeing is white sensor pattern noise NOT blue noise.
Click to expand...

Thanks! I added your idea on CTB to my original post dealing with tungsten.

It would be great if that temperature info was available in the meta data, and if there were a color gradient warning (in the camera LCD) if the current black shade temperature varies too much from the sensor temperature.
 
I love the gradient warning idea. I think they are working very hard on getting the black shading time down. It’s just hard to wait 40 minutes to do it properly.

Also, to add to your list about sharpness make sure you’re checking your image at full rez. The default in RCX is 1/8th and that will always be soft. That is a very common mistake. Also if you are judging that off the .tiff stills it will usually be softer as well. I’m not sure if it’s the .tiff compression or motion blur, but that is not a good way to perceive sharpness. A better way might be importing the R3D still into Lightroom and process it from there. It’s worth noting that almost everyone has trouble with focus moving from a small sensor to S35 sized sensors. It just takes a little testing and they will nail it.
 
Ryan Nichols said:
I love the gradient warning idea. I think they are working very hard on getting the black shading time down. It’s just hard to wait 40 minutes to do it properly.

Also, to add to your list about sharpness make sure you’re checking your image at full rez. The default in RCX is 1/8th and that will always be soft. That is a very common mistake. Also if you are judging that off the .tiff stills it will usually be softer as well. I’m not sure if it’s the .tiff compression or motion blur, but that is not a good way to perceive sharpness. A better way might be importing the R3D still into Lightroom and process it from there. It’s worth noting that almost everyone has trouble with focus moving from a small sensor to S35 sized sensors. It just takes a little testing and they will nail it.
Click to expand...

40 minutes is a long time (especially if a black shade fails and you have to start over), but if you can apply saved black shades from different temperatures, maybe that's workable?

I've heard that some don't de-bayer at 100% when in RCX - and get concerned about the quality. I'll try to write something up about de-bayering at 100% and viewing at full resolution. But, too tired now, time to sleep as big day watching NAB news. Thanks! (BTW - what's your technique for nailing focus?)

Good luck with your issue, please post about the resolution of the problem.
 
Well, for your debayer settings in the transcode Half Pristine works. Full Pristine gives you a huge amount of more data for very little gain. That’s been covered by many people, not just myself. In RCX doing a first light however, you have to bounce back and forth under the resolution tab at the top to get your desired sharpening. Almost all the time I just sharpen to three at the top in the pre gamma stage and then now I use the default setting for unsharp mask and it suits me just fine. Sometimes I might even dial the unsharp mask back a little if it’s too sharp. RPPs resolve VERY sharp. I love them.

On set I use the 1:1 (image magnify) on both cameras and verify with the focus check feature since that is very close to traditional peaking. I always get focus using that. I always pull again when changing focal length on a zoom because I never assume it’s parfocal, even though mine seems to be. I usually always go for the sweet spot on my lens if I can as well. I pull tape to verify sometimes. You have to know your lens for that though...I would never pull tape with a still lens or a lens you don’t know for sure if the marks are on. Do not assume they are. All that coupled with a second check by the Scripty or someone watching a t least a 22’” monitor and you’ll be fine. Even if you don’t have a monitor you should be fine doing the other two. If your subjects are moving and going out of focus when someone is pulling, the first thing to think about is your aperture. It’s tricky to nail at first but it comes very easy once you have it.
 
Great job on the "sticky"

Great updates and links to information, thank you for a one stop info check thread.
 
Les C. said:
3. verify enough Light. Improperly lit scenes = more noise.
Click to expand...

While this is essentially true, it is also misleading. Noise isn't actually the result of improperly lit scenes, it is the result of the abuse of lift, gamma and gain. When you attempt to inject light into places where there was none, you get noise. The parts of a frame that were black when you shot them should be allowed to remain black when you print them. It is not incorrect to do that. It is not incorrect to light a scene and leave parts of it in darkness. To fully illuminate a scene that should be played in darkness is incorrect.

When highlights exceed the range of the imaging system we say they have "blown out", meaning that there is no detail information in them. Sometimes it is right to allow brights to blow out. We could also say that noise is the result of blacks clipping, or rather it is what happens to blacks when they clip and details are lost, but somebody tries to recover them. If absolute blackness is your intent, don't fear it, but know your intent and be prepared to defend it. There is a difference between darkness and a black hole.
 
Scott Crawley said:
While this is essentially true, it is also misleading. Noise isn't actually the result of improperly lit scenes, it is the result of the abuse of lift, gamma and gain. When you attempt to inject light into places where there was none, you get noise. The parts of a frame that were black when you shot them should be allowed to remain black when you print them. It is not incorrect to do that. It is not incorrect to light a scene and leave parts of it in darkness. To fully illuminate a scene that should be played in darkness is incorrect.

When highlights exceed the range of the imaging system we say they have "blown out", meaning that there is no detail information in them. Sometimes it is right to allow brights to blow out. We could also say that noise is the result of blacks clipping, or rather it is what happens to blacks when they clip and details are lost, but somebody tries to recover them. If absolute blackness is your intent, don't fear it, but know your intent and be prepared to defend it. There is a difference between darkness and a black hole.
Click to expand...

You bring up some good points.

I'm curious about shooting dark scenes (as I really haven't)? Whether it's better to light it enough to get valid data on the sensor by exposing to the right (so as to have less noise in the darkest areas - a side effect of linear sensors having fewer bits in the darker ranges), and then crush the blacks - and then grade to achieve the desired 'look' - or to hold back 'injecting light'?

When I wrote "verify enough Light. Improperly lit scenes = more noise", I agree that certainly needs to be fleshed out a bit.

Do you have a good link that covers your points, or care to summarize in a sentence or so - which I could link to your recent post (which you could re-phrase)?

But, somewhere in there, we need to convey that they will get less noise if they actually capture data that's in the range the sensor wants. I ?think? if they have the data to begin with, they can grade to get the desired effect. What you can't do is create information from data where it doesn't exist.

And to some degree, it is true that improper lighting (as to what the sensor needs), will result in more noise and less valid data stored?

Ah ... I'm dealing with my taxes, not fun, brain is mal-functioning - so may not work on this much until sometime tomorrow!

Thanks!
 
I site no reference, these are just my recent thoughts as I get more familiar with Scarlet and the craft. I am not formally trained, or even particularly well read. One day soon I will go to the ASC Manual for some definitive answers.

As we have mentioned, exposing to the right and bringing levels down to where you want them is best. It yields the most flexibility with the widest range of grays, but the larger point was don't be paralyzed by the possibility that some of your blacks are clipping. If you do it, know exactly what you are doing and why. Have a good reason.

Think of it this way; There is no noise in clipped blacks. There is only blackness, nothing, no data. If you grade the footage properly your image will bear that out. If you don't you will introduce (or reveal) noise. You can not create data where it doesn't exist, just noise. (Which I guess counts as data, but I digress.)

More important than the presence of black clipping itself is where you place the middle, as this will determine what falls off and how rapidly it falls off.

This evening I decided to plumb the depths a bit in search of my Scarlet's noise threshold. This was not a "test" or anything formal or scientific, just a guy and his new girl getting to first base. Here is what I did.

I hung up a black chipboard, a tungsten light with CTB and warmed Scarlet To normal operating temp six feet from the chipboard. It was the same temp at which black shading was done today; 29/45c. ISO 320, 1/48 24-70L @70mm f3.5 (Sharpness was not being measured. In retrospect f4 would have been more useful just the same.) I turned on false color and used it to measure three points: no false color at upper threshold cyan, cyan at upper threshold of purple and full on purple (noise). I varied lamp focus to achieve the desired intensity levels.

My meters are not ideal for this exercise, but I used the Sekonic 758R for incident readings and the C-500R for color temp and footcandle readings, though it is not an exposure meter. The 758 is not a cine meter, so it doesnt read fc or lux, and I had to read in 1/30th shutter measurements.

Results were as follows:
no false color at upper threshold cyan- 18.1fc/1/30 f8.1 @320 5240k
cyan at upper threshold of purple- 13.2fc/1/30 f7.7 @320 5160k
full on purple (noise)- 9fc-1/30 f7.2 @320 5070k

So my Scarlet's noise threshold (Black clipping point) appears to be between 9-13fc @f3.5 and f7.2-7.7 at 320 1/30. Your mileage may vary.

I will repeat this exercise tomorrow night with straight tungsten. It should make for an interesting comparison.

Les C. said:
You bring up some good points.

I'm curious about shooting dark scenes (as I really haven't)? Whether it's better to light it enough to get valid data on the sensor by exposing to the right (so as to have less noise in the darkest areas - a side effect of linear sensors having fewer bits in the darker ranges), and then crush the blacks - and then grade to achieve the desired 'look' - or to hold back 'injecting light'?

When I wrote "verify enough Light. Improperly lit scenes = more noise", I agree that certainly needs to be fleshed out a bit.

Do you have a good link that covers your points, or care to summarize in a sentence or so - which I could link to your recent post (which you could re-phrase)?

But, somewhere in there, we need to convey that they will get less noise if they actually capture data that's in the range the sensor wants. I ?think? if they have the data to begin with, they can grade to get the desired effect. What you can't do is create information from data where it doesn't exist.

And to some degree, it is true that improper lighting (as to what the sensor needs), will result in more noise and less valid data stored?

Ah ... I'm dealing with my taxes, not fun, brain is mal-functioning - so may not work on this much until sometime tomorrow!

Thanks!
Click to expand...
 
Last edited:
Sorry for that but as I mentioned, this was not science. I was just trying to find out how far she would let me go. ;-)

If you really want it to look scientific I will have to repeat for all aperture settings in both daylight and tungsten and plot the data out on a graph. Graphs look so definitive. :-)

Doug Beatty said:
The science of art...or is this the art of science? Either way, Scott, you have managed to make my head hurt.
Click to expand...
 
Last edited:
I went ahead and metered light levels for each f-stop using false color as described above. Instead of measuring cyan, blue and purple I measured no false color, blue and purple to represent a continuous range from acceptable exposure to full noise.

It was a useful exercise and I encourage anyone who wants to know their Scarlet fully to do it. I now have a number of reference points that will help me make informed decisions going forward and I learned the value of false color. More importantly I am developing a feel for my Scarlet.

ScarletLowLightNoiseTest-2.jpg


ScarletLowLightNoiseTest-1.jpg


ScarletLowLightNoiseTest-3.jpg


The graphs bear out the fact that you hit the noise floor sooner under tungsten light than daylight. Tungsten also performs worse than daylight as you close the iris down. There is a sharper rise on the right of the tungsten graph than daylight.

Since doing this I updated my firmware to 3.0.6 which is described as more accurate in the representation of the metering tools, and therefore If I did this again, my results may be quite different but I wasn't going for hard numbers in the first place. Instead I have a greater respect for false color, and that should serve me well.
 
Last edited:
Scott...
wow.
just - wow.

Its uber-geek but impressive.

Stu Aull
Alaska
 
You must log in or register to reply here.
Back
Top