Hacker News Reader

Comment by snapetom | Hacker News Reader

Yeah, I was vehemently anti-mirrorless until maybe just like 6 weeks ago. I tried to use every reason to stick with the latest Canon DSLR, but there are just too many advantages with the mirrorless R5.

And a lot of advantages are the computational tech in the cameras that you speak of. The eye tracking and focus is absolutely incredible. Those that claim cellphone cameras will replace dedicated cameras have no idea the advances that are taking place in the cameras. Camera companies are not sitting by idle and waiting for cellphones to make themselves obsolete. They are making incredible strides. With the advantages of the sensor sizes and lens physics, I'm not going to say cellphones will never catch up, but I find it hard to believe they will.

View on HN

The new Sony flagship[0] is awesome. It’s a $7,000 body that they can’t keep in stock. Sony also has great glass.

[0] https://www.bhphotovideo.com/c/product/1619483-REG/sony_ilce...

At that MP and that price, it's dipping its toe into medium format photography.

Ultimately, in the hardware sense, I think making medium format technology accessible to the masses, is where Canon/Sony/Nikon is heading. It's going to be about striking a balance between camera size, sensor size, using software to fill in the compromises, and of course, cost. However the point is that while cellphones are catching cameras, cameras are also moving ahead.

Medium (and large) format has more to do with the size of the sensor.

Big sensors require big housing, a minimum optical distance, and big, expensive glass.

Ansel Adams must have been pretty diesel. He used to haul these gigantic cameras around in the wilderness. They often had a plate size of 8” X 10”.

And then came the next iPhone.

I am by no means an Apple zealot but its astonishing that with every iteration they set new standards.

My wife has the latest Pro Max. I have a mini. I also have one of the latest Canon mirrorlesses and pounds and pounds of lenses.

Give me my setup any day, but I'm looking at it for what I love to do - wildlife, sports, cityscapes. I don't care about portraits, I don't care about taking picture of food, I don't care about capturing life memories/events. Sure my use case is not what a regular consumer uses a camera for, but a cellphone can't come anywhere close to what I do.

In my experience even the cheapest (mirrorless) digital snapshot camera runs circles around any mobile phone's. I used to have a cheap Nikon camera and its pictures still stand-out compared to those of even high-end mobile phones (like my Samsung Galaxy S9), especially in low-light or imperfect lighting situations.

If you want to take great looking pictures, buy a camera. Don't settle for a mobile phone.

I agree. Light issues are like distance issues. It's about capturing data, which cellphones will never match given their sensor and lens limitations.

Sure Apple/Samsung boats how good their phones are in low light situations these days. But people forget how awful they were in the past. Cellphones were useless for low light even just a few years ago. They've made great strides, but the lower and lower light you go, cameras are the clear winner.

I read somewhere, the photos from phones are looking great when are viewed on phones. The moment you view them on a big screen, you immediately understand the picture is not that great as you thought. One could argue the great screens of phones nowadays and the vivid colors affect the perception of the image

What really is the mark of a pro photographer, is composition, and control of light.

My wife is a far better photographer than I am. We could be standing next to each other, shooting the same subject. When people look at my photos, they go “That’s great!”, but when they look at hers, they forget all about mine.

Absolutely. How you view photos absolutely come into play. Digital? Print? Viewing them on a monitor, even a huge one, has different requirements than, say taking a photo and making a wallpaper (as in a real wallpaper, on a real wall) out of it. I have terabytes of photos that look fine on screen, but if I printed it on an 8x10 to hang up, they look like crap.

This feels a little like a decade-old maxim that needs a refresh. Most comparisons that I find of modern flagship handsets against even mid-range point-and-shoots show the dedicated cameras failing to dominate. Particularly with computational processing it gets really tough to honestly just wave a hand and scoff at the mobile as a primary choice for most casual photo situations (and some professional, as well!)

I too love my manual-operation DSLRs and mirrorless swappable cameras, and just bought a new lens for my go-to. But I've not found myself in years compelled to spend what I'd need to on a pocket camera that'd beat what my phone can do.

My experience is similar, and I bet our experience is very common (among people who've tried both). My prosumer DSLR and medium-expensive lens is so much better than any phone for wildlife photography. For indoor low-light portraits, though, my Google Pixel 3 is actually better than my DSLR, and I say that as someone who very much expected the opposite.

Different purposes of computational capabilities in modern cameras vs. phones place them reliably in different categories.

The former aim to computationally adjust settings and make it easier to capture raw scene data the way you intend. The “magic” happens on lower levels (e.g., tracking subject), and you generally feel in control of how it’s applied.

The latter focus on computationally interpreting raw data and generating a display-referred image. The “magic” happens throughout the process, and is mostly a black box (recall that case where iPhone camera replaced human head with a leaf on a group photo). If you use phone camera to capture raw image, phone software may still be able to assist to a degree, but challenges arising from sensor size and lens limitations would become more obvious.

(I would, however, watch out for the line becoming blurred if/when phones—hopefully not cameras!—start to capture computational raws. Potential improvements offered by applying extensive magic at unmosaiced raw signal level would for me be killed by that human-head-turned-to-leaf possibility, which would ruin phone photography for me.)

1. GCam, MotionCam etc were always able to output stacked raws. In case of raws, "computational" is pretty much a marketing buzzword that means very little, it's just good old exposure fusion and stacking in bayer domain, with in-camera alignment robust enough to shoot a long stack handheld. They pretty much have the same exposure/iso priority and controls you expect to simplify capturing the most of the scene data unchanged. There's even experimental RAW video recording in MotionCam, however ridiculous that might sound.

2. RAW is also pretty much meaningless with modern mirrorless since it's not literally raw. They also do a lot of stuff under the hood for years, including "computational" stuff like noise suppression or pixel shift/superresolution in bayer domain.

So the line isn't just blurred, it doesn't exist at all. All this "computational" stuff is here precisely to capture more of the real data, not to draw something that never existed.

I don’t know what are things you reference in (1), some apps?

In (2), either you are speaking about consumer point-and-shoots, or you are speaking of features that can be turned off, or more typically have to be turned on. (Have you ever tried to use pixel shift, by the way? What kind of magic do you think does it do, except for capturing multiple pixel-shifted exposures?)

Mirrorless cameras I use get me a dumb array of raw light values, and I am free to apply (or not) noise reduction as I interpret the data. And if they were applying noise suppression, I would not put it anywhere in the same domain as computationally detecting and smoothening human skin or the like.

The line is there, and at this time it’s pretty sharp. If you don’t believe, try some of the latest “dumb” pro mirrorless cameras with the latest iPhone and compare their respective raw captures.

GCam is Google Camera, the app that started all this. Funny how everybody praises iPhone when Apple actually have been very slow to notice this trend, and their camera apps really sucked for many years. I hate GCam specifically because it requires lots of fiddling with mods and ad-hoc setups if you want full potential, but it doesn't change the fact that a "computational" camera outputs real scene data.

>computationally detecting and smoothening human skin or the like

Seems like there's some misunderstanding. I'm not talking about stuff like face detection or beautifying, computational bokeh, one-size-fits-all tonemapping etc. I'm talking about raw capture, which simply aligns and stacks many frames (discarding bogus ones) without demosaicing. There's really nothing more to it. Not much different than, say, Pentax' pixel shift. Similar process, same limitations and issues, no opaque magic that tries to think for you. I've used similar process for years with HDRMerge to shoot landscapes with my DSLRs.

>If you don’t believe, try some of the latest “dumb” pro mirrorless cameras with the latest iPhone and compare their respective raw captures.

I currently use the a7S III, but I fail to see how it's relevant. The difference is not due to them being dumb but because of having better sensors and optics, variable aperture, mechanical shutter etc. There's nothing smart or special in smartphone raw capture.

We seem to mostly agree. Here’s the line—on one side of it are phones, able to produce decent images despite hardware limitations by applying black-box magic when interpreting raw captures (but not faring that well when used to shoot raw); on the other side are cameras, employing advanced tech (and sometimes, indeed, magic) to assist you with capturing raw signal of much higher fidelity and yielding full control as to its further interpretation.

That’s what I meant by cameras being in a qualitatively different category, and why I wrote my comment arguing against the next iPhone somehow rendering mirrorless cameras useless.

(The only point where it might become more of a debate in my eyes is if phones start applying similar extensive processing that they do, but at raw levels. There could be genuine gains there—but also unwelcome unpredictable adulteration of DNG captures.)

My main objection is this:

>able to produce decent images despite hardware limitations by applying black-box magic when interpreting raw captures (but not faring that well when used to shoot raw)

Not really, the quality boost in recent smartphones comes exactly from bayer domain stacking, which is scene-referred capture. You can absolutely get precise photometric values from this process, as precise as your hardware and lighting allows you to, and as long as you calibrate your device. Those apps contain profiles for specific sensors precisely for this reason, albeit you can do it yourself better, as many people do just like they do it with proper cameras. There's no black box magic here, the output is just the same ordinary DNG with the CFA photometric interpretation, but with significantly lowered noise floor and better dynamic range. Moreover, the pipeline is strictly separated into two parts: bayer domain operations, which give the quality boost, and post-demosaicing processing (auto white balance, tonemapping, skin smoothing, other features aimed at point-and-shoot users - what you call magic).

I'm not claiming smartphones are somehow better, which would be silly. I'm objecting to a claim that apps do some opaque magic behind your back when writing RAWs. They don't, you can get entirely normal RAWs with all the quality benefits of stacking (and its limitations).

And to be honest, I feel like our discussion is creeping into a no true scotsman/audiophile territory a bit. There's nothing sacred or qualitatively different about single-shot capture, no modern normal-sized sensor gives you truly unprocessed internal bucket values, and even what is counted as single shot is often unclear (many sensors are capable of simultaneous multiple exposure capture requiring the same exposure fusion to get the result; is it single shot or not?). What smartphones, and some cameras, do is essentially temporal denoising, it's a routine imaging technique employed by videographers and in technical imaging (which is as precise and scene-referred as you can get), and I'm not sure what's magical here. Same with temporal super-resolution that many cameras do.

Sure, some of the black box processing that phone camera does relies on known algorithms. The difference is that it also makes use of ML, which IMO either plays a significant role or will do so in near term.

> auto white balance, tonemapping, skin smoothing, other features

This, for example, I see as an opaque step with nondeterministic outcomes, which if you take away from phones and shoot raw reveals how far they are away from mirrorless cameras. (Cameras that shoot raw do not do this all that well, nor are they expected to.)

> There's nothing sacred or qualitatively different about single-shot capture

I stack exposures too. Whatever process works, works. The difference is that my flow is A -> N -> X, my camera’s built-in processing is A -> M -> Y, while ML-powered black-box compute in a phone does A -> ? -> R today or A -> ? -> Z tomorrow.

Due to its automatic nature, I see this processing as taking more of the creative part of photographer’s role, unlike compute in cameras where it is a capture-time assistant.

It will have to play a bigger and bigger role in how phone camera output improves in near future, as there is not that much space for enlarging the sensor or the lens; and as a user you either let it do the magic or you end up with a very mediocre camera.

It is not inherently better or worse, but it relates to photography in the sense that I appreciate it in kind of as Nvidia Canvas et al. relate to hand-created art: different tools, different goals.

It still seems we're talking about completely different things, or maybe you're trying to derail the conversation...

Both ML-based processing and tonemapping take place after demosaicing to produce the final JPEG. RAW capture is just your typical laplacian pyramid stacking/exposure fusion with motion compensation and sensor-specific calibration. That's the main contributing factor that somewhat compensates for the sensor size, not some kind of ML processing. If you think that ML is somehow involved here, I have zero idea where you are getting it from. The entire point of pre-demosaicing pipeline is to robustly converge the captured stack into a neutral scene-referred baseline with provably better SNR and precise photometric parameters. And that's what you get in a resulting DNG, not some neural tonemapped RGB image (which is not even created at this point).

Perhaps you should look at the pipeline yourself, here's the source code for a typical "computational imaging" camera app that produces stacked RAWs, for example. [1] And here's what Google Camera does [2] [3] (you can ignore everything that comes after demosaicing). Google's particular method is also partially implemented for arbitrary RAWs as third-party software, so you can stack any burst capture from your camera with the same method smartphones do. [4] (there's much less benefit for normal sized sensors, for obvious reasons)

[1] https://github.com/mirsadm/motioncam/

[2] https://static.googleusercontent.com/media/hdrplusdata.org/e... (PDF)

[3] https://static.googleusercontent.com/media/hdrplusdata.org/e... (PDF)

[4] https://github.com/martin-marek/hdr-plus-swift

> It still seems we're talking about completely different things, or maybe you're trying to derail the conversation...

I am making literally the same point I made when replying to jhoechtl: phones and cameras are different beasts, and the next iPhone is unlikely to obsolete semi-pro mirrorless cameras. Judging by your tone, you are dead set at proving me wrong, but I can’t see a meaningful argument.

Why do you keep reverting to ML involvement at pre-demosaicing level? “Magic” computed raws is not something I claimed to happen now, nor am I certain it ever will—I just speculated we are logically headed there and if we do get there it may affect my view at phones vs. mirrorless.

So no, not interested in a hypothetical typical “dumb” camera app or raw capture from a phone. I mean the computationally more advanced capture (like the one done by bog standard Camera) that produces good images, which is what makes people think iPhone can rival a mirrorless. If you take away (by shooting raw or using a dumber app) semantic masks and all the magic it uses to obtain display-referred data for you, you just get a mediocre camera compared to a mirrorless, and it will remain the case for the next few iPhone releases.

> (recall that case where iPhone camera replaced human head with a leaf on a group photo)

Though that leaf was just a leaf, hanging in the wrong place - no magic involved, not the iPhones fault.

https://petapixel.com/2022/01/03/the-mystery-of-an-iphone-re...

Didn’t know this was debunked, my faith in iphoneography was restored a little!

Still, I occasionally notice artifacts of computational processing—most recently, wrong color of a building (it somehow decided it’s yellow, instead of reddish beige), other times some skin improvements, eye sharpening and so on. (I don’t shoot in raw on my phone though.)

Reply on news.ycombinator.com