Hacker News Reader

I can't relate to this at all. I have a Canon EOS 450D (Rebel XSi, 2008 model year) with a few lenses that I used to love, but haven't really used for years now. Its low light performance sucks, it's almost impossible to photograph people at night. My phone does a much better job of that. Yes, the DSLR's optics run circles around any phone, but the sensor really does show its age. And since you specifically mention the autofocus, I just want to add that I find it infuriating. It's awful compared to my phone! It reliably focuses on the wrong thing and is really rather slow. At night, it barely works.

To me, the old DSLR is simply not worth the weight of lugging it around any more. It's funny how we have such different perceptions of our old DSLRs.

I kind of straddle both the positives and negatives - recently pulled out my 7D from a while back and the images are fantastic even only JPG.

I don't expect to do low-light without long exposure (ie, no impromptu shots) and I do have an external flash for bounce lighting indoors.

It's still better than my iPhone (which admittedly is a couple of years old), but weight is a huge factor I only use this when intending to get really good shots. Another is the extra time it takes to transfer files, sort through bursts and put into iCloud or smugmug.

I strongly suspect you may be putting everything on auto and hoping the camera will focus on the right thing. It wont, because the camera has a very crappy auto-mode. It will, when you learn how to use manual mode and pick your own focus point. I fully accept that this is not something most people are interested in, but that camera can and does take wonderful images. Here are a few samples (randomly picked by flickr - I hope they're good :P ).

https://www.flickr.com/cameras/canon/eos_digital_rebel_xsi/

To be fair to other DSLRs, the Rebel XSi's autofocus was considered one of its weakest parts when the camera was new. There are lots of other autofocus systems which are much faster and more accurate.

The bad low-light performance wasn't universal at that time. I made low-light performance a priority, so I got the Konica Minolta Maxxum 5D.

Contributing factors may be that the pixels on the sensor are physically larger, and that the sensor-shift image stabilization helps with longer exposures.

You're comparing a modern phone to a very old camera in DSLR terms. Much like the nokia you probably had in 2008, things have very much moved on.

Well, the parent was talking about their even older DSLR and their equally modern phone, so I'm not sure what your point is. Nobody is claiming that a 2020 phone is anywhere close to a 2020 (or 2015, for that matter) DSLR or mirrorless camera.

Are there DSLRs with image processing features like Google's "Night Shot"? Where you can take a very long hand-held exposure and it will never be blurry?

Not quite but close. My Sony A7ii will take a 2/3s of a second exposure by stacking four shorter exposures and using stabilization.

2/3 of a second with an f/2 or f/2.8 lens means you can take pictures lit by not much more than starlight.

Actually, even without this, with an F/1.4 lens I can take a handheld picture using only light pollution.

To a very old and entry level camera.

I was not referring to phone lens, although those things are being applied there too.

It is now possible to build a telephoto lens for your DSLR, that costs $2 to make, is short and flat, yet has better optics than your current kit. When these become common, it will likely cause another seismic shift in the pure-camera market. Perhaps some of them will actually be bonded with the sensors, unlike what we have today.

Regardless, new lens technology -- while likely to make a greater impact on mobile phones -- is largely orthogonal to them.

> It is now possible to build a telephoto lens for your DSLR, that costs $2 to make, is short and flat, yet has better optics than your current kit.

Do you have any references you can share? I'm super interested.

edit: For anyone else interested, the key phrase that worked for me was "metasurface lenses".

... can you actually buy this yet?

Metamaterial lenses seem to be the fusion power of photography. Just one breakthrough that might not happen before they're commercially viable.

I have a Sony RX-100 (Mark-IV specifically). Also with a 1" sensor but in a compact body. Even without flexible lenses, I agree with all your points. I make sure to bring it on trips where I expect to be taking lots of pictures. The cellphone is great in the sense that I always have it on me, and the best camera is always the one you have. But the comparative quality is not even close.

But, I mean, it better not be considering both were about the same price!

>Doing great things with 10X less light is impressive, but at some point, you need a big sensor and good quality glass.

This problem is basically solved already by taking multiple exposures and blending them in software. Once we're at the point where stacking 10 exposures works reliably even in challenging conditions (and we're pretty nearly there with the latest phones) then many of the advantages of a larger sensor are moot.

As far as glass "quality" is concerned, smart phone manufacturers benefit hugely from economy of scale. I wouldn't necessarily assume that smart phone lenses are manufactured to lower quality standards just because you're paying less for them.

There are still a great many. First, much greater resolution. Then, being able to take pictures of things going fast, then also being able to take non-blurry pictures in very low light, then there is the dynamic range, and then there is also constant aperture zoom as well as telephoto focal ranges.

Lenses on phones are limited by diffraction. They can never provide the quality of a larger lens because their aperture is so small that their airy disk is often over the size of a pixel.

Phones are already better at taking non-blurry pictures in very low light. They generally have pretty fast shutter speeds available for moving subjects. Dynamic range is not particularly important now that exposure blending works so well.

Diffraction all comes out in the wash. Smaller sensors allow wider apertures (and phones have some room to grow in terms of sensor size).

The resolution advantage is actually quite moderate (people are often mislead by comparing the number of pixels rather than its square root).

If you use multiple exposures for exposure blending you lose sensitivity, and in the case of fast shutter speeds image quality plummets.

Phones are simply not better at taking non-blury pictures in very lowlight. Cameras evolved since the 90s, we have 5 stops of IS, sharp f/1.4 lenses and exposure stacking.

The resolution advantage is not moderate at all. A 12MP phone is around as sharp as a theoretically perfect 8MP sensor with a perfect lens, while current cameras are as sharp as 40+MP perfect sensor with a theoretically perfect lens.

Diffraction absolutely doesn't come out as a wash. What matters is the actual diameter of the aperture, not the ratio. Phones are a tenth of ILCs in this metric, or even less for non-optimal focal lengths.

>Diffraction absolutely doesn't come out as a wash. What matters is the actual diameter of the aperture, not the ratio

It's actually the f stop that matters for diffraction, not the absolute diameter. Roughly, this is because the light spreads more with a longer focal length because it has a longer distance over which to diverge. If you crunch the numbers this exactly compensates for the larger absolute diameter of the aperture. Here is a post with a reference: https://www.dpreview.com/forums/thread/21428

It's amazing how much enthusiast discussion about cameras is based on basic misconceptions about optics.

I'm not talking about linear resolution. I'm talking about angular resolution.

Here, the airy disk diameter is the image size. At the same magnification, the image height is proportional to the object size.

Since when the f number is smaller, at the same magnification, substituting in the thin lens formula with the magnification, you find that the object size is inversely proportional to the focal length.

From this you come to the conclusion that, at the same focus distance, the smallest object you can image is a function of the absolute diameter of the aperture.

Intuitively, this is because even though the airy disk has the same size, the focal length is much smaller, therefore the field of view is much wider for the same sensor size, therefore the angular confusion is much bigger, therefore for the same FoV at the same f number detail goes down.

Or put another way, yes the light spreads more, but the sensor on a camera is bigger, so in the end you have much more detail anyways.

This is why, for example, spy satellites despite operating at f/11 or so can resolve much more detail than a cellphone :)

This is a standard exam question if you ever take a Wave Optics class, by the way. But as you say, enthusiast discussion, such as on DPReview, is often based on basic misconceptions about optics :)

Apologies for the snark. This would probably be a fun discussion in person. In my experience, most apparent disagreements like this just come down to confusion over which factors are being held constant out of field of view, sensor size, focal length, total amount of light gathered, etc. etc.

If the phone was using a slice of a DSLR sensor, then yes, the absolute aperture would be what was relevant, as you obviously cannot improve the resolution of an image merely by cropping it (if you imagine a ultra ultra wide DSLR lens which becomes a normal phone lens after the crop).

But two points:

First, you're then comparing resolution at a focal length that's irrelevant to DSLRs, so in practical terms it's not a like for like comparison.

Second, phones are using sensors with higher pixel density. The question is just whether the phone can use a wide enough fixed aperture to avoid being limited by diffraction at the desired resolution. A fixed aperture of e.g. f1.6 is practical at phone focal lengths but not (as a fixed aperture) at full frame focal lengths. If you calculate the theoretical linear resolution limit imposed by phone camera f-numbers, it's approaching 1000 lp/mm for the better models (e.g. an iPhone with an f1.6 aperture and a ~10mm sensor). I'm sure the real limits are lens aberrations and noise at the moment.

It's not an either-or. Diffraction makes things a lot harder, and some phone lenses are indeed essentially diffraction limted

Secondly, I'm actually comparing resolution at fixed FoV, not a fixed focal length.

Again, the sensor size doesn't matter, what matters is the focal length. The focal length of an iPhone main lens isn't 10mm, the sensor size is 8.5mm but the focal length is actually around 5mm. You find the theoretical maximum linear resolution is more around 300-400 lp/mm. If you run the numbers, at f/1.6 the airy disk diameter is already bigger than the pixel pitch (2.11 microns airy disk and 1.7 micron pixel pitch). The resolution is already being seriously degraded by diffraction, although it's not completely diffraction yet. Lens aberrations make it even worse, of course. All in all, realistically the resolution is pretty much at its limit.

If you work at this regime where the linear resolution is close to equal to the airy disk, you start having a type of abberation where changing the wavelength of the system causes a change in the resolution, which can't be fixed. So pretty much in practice you can't really exceed the resolution we see now even in perfect light with a perfect lens, because you start having unavoidable abberrations that are very significant to the image.

>Again, the sensor size doesn't matter, what matters is the focal length.

Well, yes, but the two are obviously tightly linked if we're comparing a DSLR to a phone camera.

I'm going by the calculation in the 'Cameras' section of https://en.wikipedia.org/wiki/Airy_disk It explains why only the f-number is relevant. If you plug values for an iPhone into the 'x =' equation, you'll see that an iPhone camera is not really close to being limited by diffraction. The minimum separation distance at 500nm wavelength at f1.6 comes out to 976nm. Along an 8mm sensor dimension that's 8188 line pairs. Now sure, that's by the Rayleigh criterion, so you're not resolving 8188 perfectly sharp line pairs. But that's a still a good margin over the sensor resolution, and it's an open question how good sharpening algorithms can get.

I'm afraid that the radius, not the diameter. The diameter of the airy disk at f1.6 for 500nn is of 2 microns, and for a line pair you'd need double that, not a single, so a line pair would fit on 4 microns. Over an 8mm sensor that's 4000 lines, 2000 line pairs. Note that even here contrast is reduced by diffraction as we are operating from the first null. If you use the Rayleigh criterion and relax your standard for resolution you're at 4000 barely distinguishable line pairs. IIRC that would 4000LP at MTF15, so at MTF50 you might be at around 2500-2600?

And this is of course the very best you can do at 26mm equivalent focal lengths, if you want a tighter or wider shot your image quality degrades because once again the optics don't fit and you'll have to reduce your f number.

2000 line pairs in perfect conditions is quite low. Better is expected from cheap DSLR zoom lenses.

>I'm afraid that the radius, not the diameter

Yes: the quantity calculated is the smallest separation two objects can have before they blur together (by the Raleigh criterion).

According to the standard notion of lp/mm, there is no need to halve the reciprocal of this quantity to get lp/mm. See e.g. the table on p.14 here: https://zenodo.org/record/3518178/files/Verhoeven%202019%20-... An airy disc of ~2um radius translates to ~500 lp/mm. The following page also agrees with this usage, as does every other source I can find at the moment. http://www.normankoren.com/Tutorials/MTF1A.html

I think you are basically out by a factor of 4 by using the diameter instead of the radius and requiring line quads rather than line pairs. I believe the lp/mm value obtained is for ~9% MTF, so it's a fairly generous estimate of the available resolution, but not excessively so.

Regardless of the correct definition of lp/mm, the iPhone has plenty of room to increase the sensor resolution at f1.6. And phone cameras could certainly move to somewhat wider apertures and somewhat larger sensors.

Phone cameras are already at the limit of the space they can occupy. You would have to trade off the number of cameras.

As for the rest, lp/mm only makes sense at a given MTF rating. MTF50 is the standard. MTF9 is very, very, very generous. Incredibly so.

But anyways, sure, let's take 900lp/mm on the short edge for 4590lp/mm MTF9 - maybe I misremembered. A 5DSr with a real world zoom lens, the 24-70mm F2.8L, gets 7700lp/mm in the real world, and that's the MTF lens-sensor system, not just the lens. And that's not even close to the maximum possible resolution, nor to the maximum you can actually get in real life as you can. If you take the sharpest full-frame lens I know of whose MTF9 is limited to 400lp/mm because no testing apparatus was available that could go higher, the actual maximum is around 9600lp/mm. In any case, even the highest theoretical resolution of a phone is far lower than the practical resolutions of ILCs in use today, and is actually less than the sharpness of a 90$ lens on the very cheapest full-frame body you can find.

Indeed, take the Canon 50mm EF II, with an MTF50 yielding ~2000-2400lp/mm on the short edge of FF sensor. According to this source (https://www2.uned.es/personal/rosuna/resources/photography/D...), at around f/1.6 we should get around 420lp/mm, on the short edge that's 2142lp on the short edge, which is less than the cheapest EF lens in production right now. I don't know about you, but the best possible performance in theoretically perfect conditions being under the performance of the cheapest EF lens in the real world says something.

In real life, of course, the lens-sensor system will barely get close to the theoretically perfect result, and you will peak at the best possible result being around half the linear resolution of the cheapest production Canon lens.

Keep in mind, if you were to go for a larger sensor, you'd have to increase the focal length to compensate too. So 20% increase in sensor size length would need a 20% increase in focal length and at the same f number you'd need a 20% wider aperture too. And again, you'd be fixed at around 25mm FF eq. focal lengths, if you try more practical focal lengths like 50mm FF eq. you would see a drop in quality. They're already pretty much at their optical limits and have been for a while, actually f/1.6 isn't even the fastest phone lens - but those had smaller sensors still.

> This problem is basically solved already by taking multiple exposures and blending them in software.

Here's the problem.

Doing "things in software" is akin to saying ultraprocessed food is gourmet "food".

Image stacking is extremely common in low-light photography, especially astrophotography (where it is essentially the only way to get the 30min+ long exposures needed to image some deep-space objects). This need is somewhat mitigated by a sky tracker, which moves your camera together with the sky, but these are expensive and will slowly drift, making it still require multiple exposures (though you can get to several minutes before you get trailing, while without one you can only really get to a few seconds).

Basically, it's a pretty common technique for increasing SNR, it's a very different kind of "doing things in software" than, say, an iPhone's fake bokeh. It's no surprise that phones are starting to use it.

I would love smartphone postprocessing over DSLR raw images.

I feel that phones usually make better decisions for contrast, white balance, exposure than DSLRs do. With a DSLR, reaching the same finished product usually requires a ton of manual work editing photos.

Of course, the DSLR has far better detail on the pixel level, and if you need to zoom in at all, it wins in a landslide. I wish I could combine the strengths of both platforms.

DSLRs (and other high-end cameras) typically have a ton of JPEG options too. Though I usually just shoot raw and let Lightroom apply presets.

That said, the traditional camera manufacturers are pretty much laggards with respect to software so their better hardware isn't really used to full advantage a lot of the time.

I use a little bit newer Nikon DSLR (~2013), which adds another thing (the first generation of DSLRs wasn't really good at this): Color. The color of iPhones is just bleirgh in comparison, which is a little bit weird considering the cargoculting on HN of Apple's "computational photography".

The color of iPhones is just bleirgh in comparison

They're trying to solve two different problems. Nikon want to reproduce colors that are balanced and accurate. Apple want to produced colors that 'pop' and catch peoples eye when scrolling through Instagram.

This is the first comment here on computational photography, which I think is really the reason for traditional camera manufacturers going away. Most high end cellphones this generation can do stellar photography.

My dream is a camera with a nice sensor/lens and an iphone brain. It likely already exists, up there orbiting the earth...