Fuji XE2s - "dark frame" needed

Could some kind soul please upload a “dark frame” exposure (RAW) from a Fuji XE2s? If you are not familiar the term this would be a long exposure (say 30s) with the lens cap on, temperature in this case does not matter. Files can be uploaded to my google drive at https://drive.google.com/drive/folders/102_FPaLEKgp9cUXjxaHSnAZkVEUKJeGB?usp=sharing. Why? I’m playing with using dark frames to correct long exposure images and the dark frame from my XE2s does not look anything like what I expected it to, so I’d like to compare it to another sample. If anyone is interested I’d be glad to elaborate on what my sensor’s “dark-frame” looks like. Thanks.

I guess X-T1 or other 16mp xtrans files from the same generation of sensor would work as well for what you want.
Can share your dark frame so we can have look at it ?

@jd_bugs you haven’t explained what it is you’re doing, so with that in mind, perhaps the cause of your problems is that your premise is wrong.

All of these things matter - the dark frame should be captured using the same ISO, using the same exposure time, and at the same time when photographing the actual scene so that the temperature of the sensor when shooting the dark frames matches the temperature of the actual photos (one or more dark frames at the start of shooting, one or more after, as the sensor will warm up while shooting).

2 Likes

In this google drive folder, you’ll find some dar frames I took a couple years ago with my X-T1 (same sensor as the X-E2s if I’m not mistaken). They are labeled DSCFxxxx-Dark.RAF.

Thanks for your reply. This pretty much confirms that something odd is going on with my files. I am comparing dark files from my XE2s (16MP) with your XT1 (16MP). I’ll post screen grabs where I increased exposure by 12 stops (DT) in each case, no other changes to the files.

In the XT1 file there’s a fairly uniform distribution of r g and b pixels with a small over abundance of the g channel (as seen in the histogram), also a small amount of amp-glow along the sides - exactly what I would expect from this sensor. In my XE2s there’s only g channel pixels with no amp-glow even though this was at a significantly higher ISO (1600) and exposure (110s). I’ve looked at several different dark frames that all look the same. Any ideas? I’ll post the RAF files if necessary. BTW, thanks again for your Siril tutorial.

Ambient temperature does not matter in this case since I am only looking for gross difference in how the files look when the histograms are stretched by 12 stops (see details below).

It looks like the noisier area corresponds to the PDAF zone. I think I’ve seen this type of noise somewhere elses.

Yes, I think the central area does have something to do with auto-focus but where are the red and blue pixels and why is it so much less noisy than the other sensor. Where’s the amp-glow? Both files were treated the same. My main concern is to determine if my sensor is functioning correctly which is why I wanted to compare it to another XE2s. The camera functions normally otherwise. Anyway, thanks again.

Here’s an example of higher heat noise in the PDAF region, from an XPro-2 camera: https://www.dpreview.com/forums/thread/4081628#forum-post-58691423

But it’s not as strange as your example with mostly green pixels. Can you share a raw file?

I’ve posted the RAW files at https://drive.google.com/drive/folders/102_FPaLEKgp9cUXjxaHSnAZkVEUKJeGB?usp=sharing They are all the same (at least should be). The discussion about sensor noise is interesting but as you said the examples all looked pretty “normal”. I’ll dive into the discussion later today.

Thanks for the raw files. If I push the least exposed one (85 sec) by +12 eV, I see quite number of blue and red pixels:

It is expected to see more green pixels, as they are over-represented in the CFA array (for a 6x6 unit area, there are 20 green pixels, and 8 of each of blue and red pixels, if I’m not mistaken). So statistically one can expect to see at least twice as many hot green pixels as of the blue and red ones.
I don’t know if you can find many dark frames with both high ISO and long exposure time as you did, and I’m not sure that combining both high ISO and long exposure time is desirable.

1 Like

That’s correct. The ratio is (number of red pixels + number of blue pixels) * 1.25 = number of green pixels

Just a couple of final thoughts - I spent much of the day in a deep dive re: sensor noise and have come to the conclusion that manufacturers can and sometimes do mask sensor noise in raw files. In other words raw files are not always “raw”. I think this may be one of those cases but I can’t be sure until I find another sample from the same camera model. If true this makes using dark frame subtraction useless in the case of the XE2. I’d also point out that a 110s exposure is not unusual when tracking is used and the ISO of 1600 is pretty typical for this x-trans sensor as an upper limit. In fact the combination of very long exposures with high ISO is exactly the situation when dark frame subtraction should do the most good because sensor noise and amp-glow are higher than normal. Thanks for your interest and help with this.

I just thought that tracking + stacking multiple frames allowed keeping ISO low.

Do you see any evidence when using the camera that it might be performing an automatic dark frame subtraction? (I’m guessing you probably would have noticed this, but if for example a single 30s shot has two shutter clicks heard and takes 60 seconds, but only records one image, that’s a telltale sign.)

It’s usually pretty obvious, but perhaps if you started a tracking sequence and went you might not have noticed you had half the images you expected?

Just a wild guess.

Not sure why Fuji’s auto-DFS doesn’t remove those green outliers though.

@Entropy512 I was also wondering about this so-called “long exposure noise reduction”, but my intuition was that the dark frame subtraction would only affect the JPEG output, not the raw itself.
I think I should try experimenting with my Fuji X-T2, and see if the long exposure NR bakes the raw.

LENR is baked into the raw in all files I worked with (else we wouldn’t ask for LENR samples for camconst.json).

Yup, this was my experience with Pentax. They had multiple settings for high-ISO/long-exposure noise reduction, but even with everything “off”, they would force auto-darkframe-subtraction past a certain exposure time on older cameras. It’s one of the reasons I left Pentax for Sony. (Supposedly Pentax has fixed this.)

I think a lot of people want to assume RAW is RAW and that manufacturers don’t “cook” their RAWs, but the reality is that nearly all of them do. Sony at least allows you to turn off automatic dark frame subtraction, but is notorious for engaging spatial noise reduction algorithms at higher ISOs and at longer exposures. (The latter being infamously referred to as the “star eater” algorithm)

Usually automatic DFS is pretty obvious (your camera takes 2x as long per image recorded), but I know I myself have spent lots of time digging into the less obvious and missing the obvious cause on numerous occasions!

I use a barn-door tracker so my exposures are limited to 90-120s otherwise I get star trails. There’s no polar scope and alignment is by laser sighting. So I need to keep the ISO as high as possible. With this very inexpensive setup I can get 2-3 stops of additional exposure vs. un-tracked.

The LENR is turned off and there’s no second exposure.