Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Smartphone cameras struggle to capture San Francisco's orange sky

axios.com

Post reply on HN
Print view View original

51–60 of 103 posts

Re: Smartphone cameras struggle to capture San Francisco's orange sky

#51

Post by blahblah4332 »

photoGrant wrote: post #17
30minAdayHN wrote: post #10

> This was a moment for my Canon to prove that, despite its bulk, it can't always be replaced by a smartphone. Such a naive statement. Once those phones' white balance is corrected, they should be able to capture the orange tinge. Even in a DSLR, if the white balance is set to auto, it would not capture the orange tinge. Cloudy might be a good starting point.

Your statement is also naive. White balance set in camera does not affect the raw capture. Only how it’s represented.

While I agree that RAW capture will provide an improvement in the amount of data gathered from the sensor, and available for adjustments later on, if a user sets a custom white balance while taking a RAW picture, that setting will generally be accepted as a default rendering of the scene in most RAW processing software.

This is entirely besides the point because the issue stated here involves color interpretation based on a selected white point. Left to their own devices digital cameras will compensate for scene color on purpose, because 99% of the time people don't want color casts in their photos. The author is trying to blame the camera for people's ignorance of photography, which is just silliness.

Don't agree? Set your phone's white balance to daylight for a week and see how often you agree that the captured picture is what you actually wanted

Re: Smartphone cameras struggle to capture San Francisco's orange sky

#52

Post by jeffbee »

sudosysgen wrote: post #42
jeffbee wrote: post #20

Earlier quoted context omitted.

This doesn't have anything to do with the sensor. How sensor outputs are mapped to display inputs lies 100% within the domain of software.

The fact that the sensors and lenses in phones are so small is why the software has so much less to work with and what allows so much more to go wrong.

No. A bigger lens (more light) would correct for gain errors. A bigger sensor would improve either spatial resolution or gain error or both. None of that has anything to do with color balance.

All digital color photography is computational because sensor physics, display physics, and human biology are distinct processes. Whenever you have multiple color channels you need to make assumptions about the relationships between them. This is color balance and it happens in software.

Re: Smartphone cameras struggle to capture San Francisco's orange sky

#53

Post by blahblah4332 »

dividedbyzero wrote: post #15
30minAdayHN wrote: post #10

> This was a moment for my Canon to prove that, despite its bulk, it can't always be replaced by a smartphone. Such a naive statement. Once those phones' white balance is corrected, they should be able to capture the orange tinge. Even in a DSLR, if the white balance is set to auto, it would not capture the orange tinge. Cloudy might be a good starting point.

How would I do this on iOS, if I had just my phone and no CaptureOne on a laptop or the like? Are there apps that let me use a gray card to set the white balance of raw photos?

Just set the image's white balance to daylight (3500K). The colors that you see will not be corrected. You can tweak from there.

Re: Smartphone cameras struggle to capture San Francisco's orange sky

#54

Post by blahblah4332 »

matthewowen wrote: post #11

This article is awful. It's about smartphone cameras not being able to capture, but the three photos included are all ones that at least somewhat worked. Seems like it would be better to include one that failed?

As someone with a professional photography background, this article was actually super upsetting.

The author is like "look, this gas station one is the only one that worked".. well yeah, the camera's white balance algorithm saw that the scene was about 80% daylight, so it set the white point at around 3500k, the rest of the scene is recorded accurately as a result.

The thesis of the article "I don't understand white balance, so smartphone cameras are bad"

Re: Smartphone cameras struggle to capture San Francisco's orange sky

#56

Post by mindslight »

interestica wrote: post #40
jxcl wrote: post #12

One thing that I had to get through my head when I was learning photography was that there's no such thing as "no filter". Every image you take is an interpretation of the scene, either through digital signal processing or film chemistry or retouching. When I spend time editing photos my dad would claim I was "lying" by editing them, he thought any modification to digital photos was wrong because the jpeg the camera…

It's actually pretty interesting how much processing goes on in the brain for images. We do a sort of equivalent to the phone in compensating for the colour to make it 'right'. The orange car in the article image acted as kind of an 'anchor' for the colour compensation. The blue/white-black/gold dress [1] controversy kinda illustrates that 'anchoring'. The 'Checker Shadow Illusion' [2] is another instance where our b…

IMO the orange car had little to do with the output color grading. The real color temperature anchor is the bright foreground illumination.

Re: Smartphone cameras struggle to capture San Francisco's orange sky

#57

Post by sudosysgen »

jeffbee wrote: post #52
sudosysgen wrote: post #42

Earlier quoted context omitted.

The fact that the sensors and lenses in phones are so small is why the software has so much less to work with and what allows so much more to go wrong.

No. A bigger lens (more light) would correct for gain errors. A bigger sensor would improve either spatial resolution or gain error or both. None of that has anything to do with color balance. All digital color photography is computational because sensor physics, display physics, and human biology are distinct processes. Whenever you have multiple color channels you need to make assumptions about the relationships be…

That's not how it actually works out in practice. Small lenses mean small apertures which means you get diffraction, and you get it pretty fast. With a 1-2mm aperture lens you get diffraction limiting resolution to around 8-10Mpx.

As for color balance, sure, but the parent wasn't limited to color balance.

Re: Smartphone cameras struggle to capture San Francisco's orange sky

#58

Post by sudosysgen »

blahblah4332 wrote: post #47
sudosysgen wrote: post #42

Earlier quoted context omitted.

The fact that the sensors and lenses in phones are so small is why the software has so much less to work with and what allows so much more to go wrong.

Unless I missed something in the article, this has nothing to do with sensor or lens size. This is purely an artifact of default image processing settings and ignorance of white balance. Any user with any digital camera can perfectly capture the color of the scene by simply setting the white balance manually to daylight.

The top level comment was making a more general claim than just color balance. There are a lot of weird artifacts in phone pictures introduced by image processing overfitting.

Re: Smartphone cameras struggle to capture San Francisco's orange sky

#59

Post by Defenestresque »

matthewowen wrote: post #11

This article is awful. It's about smartphone cameras not being able to capture, but the three photos included are all ones that at least somewhat worked. Seems like it would be better to include one that failed?

This is a good example of the adjustment the camera makes: https://mobile.twitter.com/ceejay35_/status/1303702328323653... (linked from article)

Re: Smartphone cameras struggle to capture San Francisco's orange sky

#60

Post by mrob »

sudosysgen wrote: post #43
mrob wrote: post #26

Earlier quoted context omitted.

I draw the line at object recognition. If the processing involves semantic meaning (e.g. this is sky, this is a face) instead of geometric meaning (e.g. this is an edge, this is a Bayer filter artifact) then it's lying. It doesn't matter if the object recognition is done by AI (e.g. faking depth of field by blurring background) or by humans applying selective processing (e.g. painting out red-eye effect, or using a g…

The line is still very blurry. How about a camera using the depth information that is already necessary for AF to automatically crop out of process differently out of focus areas, thus detecting objects? How about a camera automatically focusing on an eye?

If there's no way to prove that semantic processing was used then the rule is unenforceable so you might as well permit it. Cropping the photograph is indistinguishable from using a different camera, and changing the focus is indistinguishable from getting the focus right by accident. Cropping and focus can both be used deceptively, but it's a different type of deception from selective editing of an image, e.g. changing somebody's skin texture.
Post reply on HN