Demonstration of X-Trans Under-the-hood RAW Spatial (Noise) Filtering

tradesmith45 wrote:

razorfish wrote:

This happens because people are too lazy to use dark frame these days. For years it was never a problem for anyone, but then someone started complaining it takes too much time, and I guess manufacturers are now trying to find ways to do it ”on the fly”, with these issues as the result. Note that dark frame in itself never eats stars, and can always tell the difference between a point of light and a hot pixel. Just goes to show that taking shortcuts never pays. Take the time to do the job right, you’ll get the best result in the end.

Actually dark frames are a poor solution because they often add noise. This happens for several reasons. One significant problem is matching sensor temps between lights & darks. During a long shoot, the sensor warms & can take 2-3 hrs to reach equilibrium. Creating darks to match that is impractical. Applying darks that are from a different temp adds noise. Some cameras report sensor temps in the EXIF but many don't. For the cameras that report sensor temp, it is theoretically possible to match temps but not easy.

A better solution for hot/stuck/dead pixels than star-eating is available in some camera specific RAW converters. A bad pixel map is used by the converter to ID which pixels need to have interpreted values applied. I believe some astro specific software will also do this.

Fuji & Oly have a bad pixel in-camera mapping function for some models but IMHO, they don't work very well.

Dark subtraction on DSLRs and mirrorless should use what is called adaptive darks that require a bias as well as a dark. These are then calculated on the actual image and adjust to that specific image.So temperature variations are not as significant. I don't think that adds any significant noise and definitely cleans up an image.

I wonder if Sony showed them how to do this filtering as Fuji uses Sony sensors. Perhaps its a "service" that comes with buying their sensors!

Greg.

tradesmith45 wrote:

crosson wrote:

may explain the waxy/wormy artifacts

FWIW you admit yourself that this does not effect images under 5 seconds. The wazy/wormy artifact is therefore still an all together separate and overly documented issue. It is a dead horse long since settled. Best to keep it separate from this particular concern so as not to add confusion, see above example.

Secondly where did you get the 5 second mark? Do you have anything referencing this or documentation on this? Or is this just personal and anecdotal observation? forgive me if you have already posted this and I don't see it.

Regarding 5 sec., go over to my post on Fuji filtering. The easiest way to see filtering is in the histogram of dark frames. These are shown in that thread. Those for dark frames less than 5 sec., show no stair steps so are unfiltered.

I did see that. Thank you.

Astrophotographer 10 wrote:

tradesmith45 wrote:

razorfish wrote:

This happens because people are too lazy to use dark frame these days. For years it was never a problem for anyone, but then someone started complaining it takes too much time, and I guess manufacturers are now trying to find ways to do it ”on the fly”, with these issues as the result. Note that dark frame in itself never eats stars, and can always tell the difference between a point of light and a hot pixel. Just goes to show that taking shortcuts never pays. Take the time to do the job right, you’ll get the best result in the end.

Actually dark frames are a poor solution because they often add noise. This happens for several reasons. One significant problem is matching sensor temps between lights & darks. During a long shoot, the sensor warms & can take 2-3 hrs to reach equilibrium. Creating darks to match that is impractical. Applying darks that are from a different temp adds noise. Some cameras report sensor temps in the EXIF but many don't. For the cameras that report sensor temp, it is theoretically possible to match temps but not easy.

A better solution for hot/stuck/dead pixels than star-eating is available in some camera specific RAW converters. A bad pixel map is used by the converter to ID which pixels need to have interpreted values applied. I believe some astro specific software will also do this.

Fuji & Oly have a bad pixel in-camera mapping function for some models but IMHO, they don't work very well.

Dark subtraction on DSLRs and mirrorless should use what is called adaptive darks that require a bias as well as a dark. These are then calculated on the actual image and adjust to that specific image.So temperature variations are not as significant. I don't think that adds any significant noise and definitely cleans up an image.

Hi Greg, isn't the use of adaptive darks limited to a few astro specific programs?  I've seen debates about how well this works in PI when sensor temps differ.  Also acquiring the numerous darks/bias in the same session as the lights lengthens the session.  Can you point us to a good video or written demonstration of this approach so we can decide if we want to make the additional effort?

I wonder if Sony showed them how to do this filtering as Fuji uses Sony sensors. Perhaps its a "service" that comes with buying their sensors!

FWIW Oly cams also use Sony sensors & they clearly do not filter RAW based on examination of long dark histograms & the substantial amount of long exposure noise present.

Greg.

tradesmith45 wrote:

saltydogstudios wrote:

tradesmith45 wrote:

We've found that Fuji applies filtering to X-trans RAW data in camera that slightly smooths images & produces some star color artifacts but no filtering is used in the Bayer sensor X-T100. This is easily seen by comparing X-T2/3 images w/ X-T100 images. The later is much sharper. Ironic isn't it?

I've done a similar analysis and found the same. Though probably not as in-depth as yours.

Fuji seems to use an "area detect" smoothing algorithm that tries to find like areas & reduce contrast within those areas. This is just a guess.

I go into much depth about it here

https://medium.com/ice-cream-geometry/x-trans-vs-foveon-a-mostly-monochrome-photowalk-1a931f8fb277

The image on the left was taken with a Foveon sensor camera (and then upscaled), and the image on the right an X-Trans sensor camera with in ACROS color mode, it's the out-of-camera JPG.

Notice how "smooth" the bricks look on the right. I describe the difference as being like a charcoal drawing (left, Foveon) and an oil painting (right, X-Trans).

This is much reduced when processing in RawTherapee.

What you describe seems to be a more extreme version of this - perhaps that algorithm kicks into high gear at long exposures? Like... If this is Fuji's version of noise reduction, then it would stand to reason that when doing long exposures, their noise reduction does more of the same.

I don't do astrophotogrpahy, but I think I'm going to have to go spend some time over there, you guys are doing some fascinating stuff.

With regards to what someone else said - I haven't done any kind of analysis but my gut hunch is that the 16 megapixel generation X-Trans sensor didn't do this - or at least not as much.

Thanks for these comments & tests Salty. I'm on the road w/ only an iPad so can't dig into your images. Can't say if Fuji uses the same filtering for jpg & raw.

In my earlier tests did report that RAW filtering can be found in XT10 images too. Unfortunate. When I compared LRCC, RP & CP1, found no detail differences but did see they all use different color profiles. The XT2 was less likely to overdo green in the background w/. RP.

Yes, I'm not saying that Fuji uses the same noise reduction algorithms for JPG and RAW.

But if their noise reduction algorithm works the way I described - then if/when they turn it on for RAW I'd expect it to act in much the same way.

May I just add that whilst there is some filtering it does not mean you can’t do stunning nightscapes with Fuji X cameras.

i have gotten some great results with Fuji X cameras.

Greg

tradesmith45 wrote:

TOMMMMMM wrote:

Don't mean to hijack this thread, but I have both an X-A3 and an X-T2. I was wondering if anyone knows if the X-A3, with it's Bayer sensor, would be an improvement over the X-T2 for astro. Unfortunately, I won't have the opportunity to test this out any time soon, but it would be helpful to know for the next time I do get to go out.

Tom, it's not hard to check the A3. Take a 2-4 min. ISO 800-1600 dark frame (body cap on in dark room). Open the file w/ RAW Digger & look at the histogram. If there is filtering, the histogram will be chopped off or have steps. A dark frame from the T2 will show you what filtering looks like. Be sure to turn LENR off.

I'd offer to look at your file but am on the road for couple weeks.

Hey Tradesmith, never used Raw Digger before, but this is what I got from my X-A3 using ISO 1600 for 3 min exposure. To my eyes, it looks like there is no RAW filtering, but I'll let you chime in on that. Seems like a normal distribution of noise for a totally dark frame.

TOMMMMMM wrote:

tradesmith45 wrote:

TOMMMMMM wrote:

Don't mean to hijack this thread, but I have both an X-A3 and an X-T2. I was wondering if anyone knows if the X-A3, with it's Bayer sensor, would be an improvement over the X-T2 for astro. Unfortunately, I won't have the opportunity to test this out any time soon, but it would be helpful to know for the next time I do get to go out.

Tom, it's not hard to check the A3. Take a 2-4 min. ISO 800-1600 dark frame (body cap on in dark room). Open the file w/ RAW Digger & look at the histogram. If there is filtering, the histogram will be chopped off or have steps. A dark frame from the T2 will show you what filtering looks like. Be sure to turn LENR off.

I'd offer to look at your file but am on the road for couple weeks.

Hey Tradesmith, never used Raw Digger before, but this is what I got from my X-A3 using ISO 1600 for 3 min exposure. To my eyes, it looks like there is no RAW filtering, but I'll let you chime in on that. Seems like a normal distribution of noise for a totally dark frame.

Glad you did this.  The pixel values seem too high for a truly dark frame though.  If you created the dark in a lit room, may have had slight light leak.  Try this again with the cam in a dark room or dark enclosure.

If there is filtering it will only affect the lowest pixel values.  That's why you have to use a truly dark.

tradesmith45 wrote:

TOMMMMMM wrote:

tradesmith45 wrote:

TOMMMMMM wrote:

Don't mean to hijack this thread, but I have both an X-A3 and an X-T2. I was wondering if anyone knows if the X-A3, with it's Bayer sensor, would be an improvement over the X-T2 for astro. Unfortunately, I won't have the opportunity to test this out any time soon, but it would be helpful to know for the next time I do get to go out.

Tom, it's not hard to check the A3. Take a 2-4 min. ISO 800-1600 dark frame (body cap on in dark room). Open the file w/ RAW Digger & look at the histogram. If there is filtering, the histogram will be chopped off or have steps. A dark frame from the T2 will show you what filtering looks like. Be sure to turn LENR off.

I'd offer to look at your file but am on the road for couple weeks.

Hey Tradesmith, never used Raw Digger before, but this is what I got from my X-A3 using ISO 1600 for 3 min exposure. To my eyes, it looks like there is no RAW filtering, but I'll let you chime in on that. Seems like a normal distribution of noise for a totally dark frame.

Glad you did this. The pixel values seem too high for a truly dark frame though. If you created the dark in a lit room, may have had slight light leak. Try this again with the cam in a dark room or dark enclosure.

If there is filtering it will only affect the lowest pixel values. That's why you have to use a truly dark.

I will try again but it will be hard to exceed shutter speed of 30 seconds without a shutter released, since 30 is the max before bulb mode needs to be set.

Even the Fuji camera remote app doesn't let you use bulbmode with this phone. Will 30 seconds suffice (since it's been mentioned that the filtering kicks in after 5 seconds), or do we need a 2-4 minute shutter speed as you recommended?