What do you mean? Statistics and ML is used in astrophotography regularly. For example, the famous black hole “photo” [1].

https://www.nasa.gov/mission_pages/chandra/news/black-hole-i...

Yes, but the computational pipeline is designed and implemented by the end user to achieve a scientific result. Here, it’s a black box where the resulting image cannot be treated as raw data.

Yes exactly. The training set used for the famous black hole image was a database of images created using black hole simulations based on our best available models. This is pretty far away from these consumer ML models trained on publicly available images which may as well be consider garbage from a scientific perspective.

Nothing about phone cameras involves "ML models trained on publicly available images".

You haven't been following the story, then, because Samsung phone cameras have been accused of doing exactly that: using a training set of publicly available images to "enhance" people's images by filling in details with a ML model.

Those people don't understand how anything works though so I don't care what they said. And how would they know if it used "publicly available images"? It's the moon, of course it's publicly available. It's in the sky.

You're focusing on the "publicly available" part but it's irrelevant to my point. Samsung's camera software is literally doctoring people's photos with fake data from its ML models. From the perspective of photographic integrity, it does not matter whether they scraped their training set off public photography sites or if they hired a team of photographers to build a dataset from scratch. To the user the effect is the same: it is not really an authentic photograph that is being created, it's a hybrid derived from (to the user) unknown sources.

All photos work like this - cameras don't know what color things are, the color you see is made up by the camera.

You've never taken an authentic photo on a digital camera.

But… they don't use ML models made from publicly available images.

There's a difference between ICC colour profiling (which is embedded in the RAW file and can be changed by the photographer without any degradation in image quality) and what Samsung is doing. The former can only affect the image colours globally (for example, allowing adjustment of white balance) whereas Samsung's changes are local, affecting image data in a profound way that can add hallucinated details which were not present in the original scene. One rather disturbing example is that the tool added teeth to baby photos!

White balance adjustments to camera raws don't generally use ICC profiles, they do whatever the combination of camera format and raw processor want.

Which can involve local adjustments, since you may want to process people differently from skies. So object recognition/segmentation models can definitely be involved, or what's essentially an upscaling model for better demosaicing. (That wouldn't be trained on public images though, unless they were the right format of camera raw.)

That is completely unrelated to astrophotography as a hobby, taking pictures of space objects from the surface of the earth. Like, totally and completely not-comparable in any way.

As somebody who has a Pixel 7 I completely agree. The AI makes my photos unusable. The worst part is I can see a usable image before it finishes processing. But then it turns it into pixilated garbage.

I am often taking pictures of small text that is borderline unreadable to get part numbers off stuff. The AI processing is absolutely a bad idea.

You can install an alternate camera app if you just want raw rixels with no computetional stuff.

Having said that, I find the computational stuff turns a fairly poor sensor and lens into kinda acceptable photos, as long as I don't want to zoom in (I can see far further with my eyes than the camera can when it comes to taking photos of tiny far away things)

on the iPhone you can enable raw image output. No need for an external app. Just open settings.

ProRAW in the system app is still heavily processed, it just saves more of the intermediate data to let you tweak more of the processing stages.

You still need a third party camera app to capture true RAW photos.

E.g. with ProRAW you can't remove most of the noise reduction, and you can't rescue images with large white balance issues (you get fringing in one of the processing steps since the white balance is baked into several of the intermediate images)

A 48 Mpixel 12 bit sensor with 128 frame stacking is 9 gigabytes of collected data per image.

The flash memory isn't fast enough to store all of that in the ~2 seconds it is collected, nor is the RAM large enough to store it all.

The iPhone is literally incapable of taking true RAW photos (at least not if you want to collect all the same info that the built in camera does)

Not sure I follow your logic, how did you end up with 9GB for a single shot, and why exactly 128 frames in 2 seconds?

iPhone's hardware is more than adequate to capture RAW videos at 60fps with binning, not just photos (too bad Apple doesn't allow it, so it can't). People are using Android smartphones to shoot videos in RAW for several years already. OnePlus 8 Pro and later can easily capture a RAW stream in 4K 60fps.

When you take a photo on iPhone, the image is composed of up to 128 frames taken before and after you hit the shutter button. Those frames are aligned and stacked using dedicated image processing silicon which can do that stuff reasonably power efficiently while working at those data rates.

The stacking also allows things in the frame to move - it uses optical flow to track movement.

That's how iPhone manages to take a tiny very noisy sensor and give non-noisy output.

If you want to see how the camera performs without this stacking, try taking a photo from a moving car of something behind railings. The moving railings screw up the optical flow, so most of the 127 other frames will be discarded (if regions of an image arent a close enough match, then the region will be discarded), and you'll get to see all the sensor noise.

Sure I know, GCam pioneered this years ago. But all those frames are merged into one which is written as a correct scene-referred DNG with the usual dimensions and the CFA pattern, as if it was taken by the sensor directly. Which is what GCam and every other app do. They don't write gigabytes of RAWs for a single shot. (although some apps do have that option, within the hardware capabilities)

And all this has nothing to do with manual astrophotography, where you ditch computational apps, use a mount and shoot actual unstacked raw sensor frames with several seconds exposure each, interleaving them with dark frames, and do the computation yourself with the dedicated astrophoto software which is far better than any app if you calibrate the sequence and do the math correctly. Every phone is capable of this.

Yeah the RAW photos from third party apps are just a single frame, that's the point. It's the same thing as a RAW photo you'd get out of a standard digital camera.

You can disable lot of AI in settings, you can also enable saving RAWs.

On that subject one of the recent samsung models had a rather interesting feature, where you could print a blurry picture of the moon, put it in a dark room and the camera would "enhance" the image adding detail it never had. I think smartphone cameras are reaching the limits of their potential, the only way forward now is ever protruding camera bumps.

On that front, there is also one called the vaonis hyperia that seems totally automatic, but perhaps we miss out on the challenge of amateur astronomy..

I've found it very useful for showing curious people (especially kids) that there's a whole universe up there mostly unseen; you can whip it out and a minute or two later they can see the Milky Way in at least a little of its splendor. For those who show interest, I follow up with our 6" backyard telescope to see the Moon's mountains, Jupiter's moons, Saturn's rings, and Mars's ice caps.

There's something different about experiencing it in person, even if the photos from a professional or a space telescope are empirically better.

If you read further, at first they mostly talk about stacking, even though it has nothing to do with "AI", but then they suddenly mix it with synthetic ML techniques, which defeats the point. Confusing.

Nevertheless, you can do some proper astrophoto with smartphones without any AI, just like you would with any other camera. It won't be that good for obvious reasons, but it's possible and it will be better than using automated astrophoto modes in any app. Calibrate your lens and sensor, make a RAW stack with fixed exposure which is not too long, and some dark frames; use something like Sequator (or more sophisticated software) to process the stack.