> Something that surprised me is that very little of the computation photography magic that has been developed for mobile phones has been applied to larger DSLRs. Perhaps it's because it's not as desperately needed, or because prior to the current AI madness nobody had sufficient GPU power lying around for such a purpose.
Sure, and my camera can do bird eye detection and whatnot too, but that's a very lightweight model running in-body. Probably just a fine-tuned variant of something like YOLO.
I've seen only a couple of papers from Google talking about stacking multiple frames from a DSLR, but that was only research for improving mobile phone cameras.
Ironically, some mobile phones now have more megapixels than my flagship full-frame camera, yet they manage to stack and digitally process multiple frames using battery power!
This whole thing reminds me of the Silicon Graphics era, where the sales person would tell you with a straight face that it's worth spending $60K on a workstation and GPU combo that can't even texture map when I just got a Radeon for $250 that runs circles around it.
One industry's "impossible" is a long-since overcome minor hurdle for another.
A DSLR and mobile phone camera optimize for different things and can't really be compared.
Mobile phone camera's are severely handicapped by the optics & sensor size. Therefore to create a acceptable picture (to share on social media) they need to do a lot of processing.
DSLR and professional camera's feature much greater hardware. Here the optics and sensor size/type are important it optimize the actual light being captured. Additionally in a professional setting the image is usually captured in a raw format and adjusted/balanced afterwards to allow for certain artistic styles.
Ultimately the quality of a picture is not bound to it's resolution size but to the amount and quality of light captured.
> A DSLR and mobile phone camera optimize for different things and can't really be compared.
You sound exactly like the sales guy trying to explain why that Indigo workstation is “different” even though it was performing the exact same vector and matrix algebra as my gaming GPU. The. Exact. Same. Thing.
Everything else you’ve said is irrelevant to computational photography. If anything, it helps matters because there’s better raw data to work with.
The real reason is that one group had to solve these problems, the other could keep making excuses for why it was “impossible” while the problem clearly wasn’t.
And anyway, what I’m after isn’t even in-body processing! I’m happy to take the RAW images and grind them through an AI that barely fits into a 5090 and warms my room appreciably for each photo processed.
There are many things wrong with this. I have an iPhone 17 Pro Max and use it to capture HEIF 48 and ProRAW images for Lightroom. There’s no doubt of the extraordinary capabilities of modern phone cameras. And there are camera applications that give you a sense of the sensor data captured, which only further illustrates the dazzling wizardly between sensor capture vs the image seen by laypeople.
That said, there is literally no comparison between the iPhone camera and the RAW photos captured on a modern full-frame mirrorless camera like my Nikon Z6III or Z9. I can’t mount a 180-600mm telephoto lens to an iPhone, or a 24-120mm, or use a teleconverter. Nor can I instantly swing an iPhone and capture a bird or aircraft flying by at high speed and instantly lock and track focus in 3D, capture 30 RAW images per second at 45MP (or 120 JPEGs per second), all while controlling aperture, shutter speed and ISO.
Physics is a thing. The large sensor size and lenses (that can make a Mac Studio seem cheap by comparison) serve a purpose. Try capturing even a remotely similar image on an iPhone in low light, and especially RAW, and you’ll be sitting there waiting seconds or more for a single image. Professional lenses can easily contain 25 individual lens elements that move in conjunction as groups for autofocus, zoom, motion stabilization, etc. They’re state-of-the-art modern marvels that make an iPhone’s subject detection pale by compare. Examples:
I can lock on immediately to a small bird’s eye 300 feet away with a square tracking the tiny eye precisely, and continue tracking. The same applies to pets, people, vehicles, and more with AI detection.
You can handhold a low-light shot at 1/15s to capture a waterfall with motion blur and continue shooting, with the camera optimizing the stabilization around the focus point—that’s the sensor and lens working in conjunction for real-time stabilization for standard shots, or “sports mode” for rapidly panning horizontally or vertically.
There’s a reason pro-grade cameras exist and people use them. See Simon D’entrement, Steve Perry, and many others on YouTube for examples.
For most people, it doesn’t matter. They can happily shoot still images and even amazingly high-quality video these days. But dismissing the differences is wildly misleading. These cameras require memory cards that cost half as much or more than the latest iPhone, and for good reason [1].
With everything, there are
trade offs. An iPhone fits in my pocket. A Nikon Z8 and 800mm lens and associated gear is a beast. Different tools, different job.
most likely one reason is that to do that, you'd have to pair the price of a fancy smartphone to a nice camera, so adding ~$1000 for a feature professionals often prefer to do offline since they can get good focus and color using optics and professional lights
Sony Alpha 6000 had face detection in 2014.