I have only experience with one of Insta360's product, the Link tracking webcam, but was left extremely impressed. The image quality was absolutely phenomenal. I imagine this action cam could only be even better working with Leica.
They've had a Leica partnership for a while in the form of the Insta360 One R (and RS) 1-inch lenses, plus with their 1-inch 360 module. There's still some fundamental physics limitations with gathering light on a small sensor, but I enjoy using the 1-inch lens on my One RS when I can over the "normal" action lens.
30w fast charge (on the Pro), good to 33ft, and a flip screen... insta360 is killing it. Having a 5nm chip on a consumer device like this, imo, shows a passion for excellence that you won't see in most competitors.
I hope these class of devices can start integrating better with phones. The screens on these things aren't big enough for comfortably complex & advanced photo-nerdery, but even my old Zcam E1 (Kickstarter 4/3" (aka MFT) 4k action camera from 2p25) has a pretty ok wifi app where I can compose shots well & relatively quickly change settings. It's also great for composing group photos. These are amazing sensor packages, and are held back by how they are marketed & targetted, but eventually those designations are going to start crumbling some, and that should be amazing.
If these devices did expose some kind of more prosumer interface I could easily see them migrating from action cams to something like camera replacements. Insta360 has already done great with cameras & optics. Their One RS platform allowing the lens + sensor to swap out, and it's easy for me to imagine expanding the lineup from mostly wide angle ish prime lenses to also incclude some optical zoom or long focal length primes.
Ideally phones would also integrate themselves well too. I can plug a USB camera (I've had much more hit & miss situation with microphones weirdly) into most Android phones and have it work (alas notably Chrome often doesn't pick up the device frustratingly!!); handy for conference rooms or desks with good fixed infrastructure. But so far we still don't good standards based ways to do this kind of thing over wifi. It'd be fun to swing for the fences & see someone bake Media-over-QUIC or something bleeding Edge into a device, but even an rtsp streamer & Android support as a sink would enable fun weird interesting advanced uses galore.
No, Insta360 has a ton of both traditional action cams and 360 cameras. The fact that they have "360" in the brand name hasn't stopped them from releasing a bunch of successful non-360 cameras.
I love my One RS 1" (the non-360 Leica lens). The only caveats are that it's really hard to search for the accessories, and doesn't have a dive case.
I went travelling recently and maybe I'm becoming grumpy and old, but it was barely possible to move for people ambling around with their phone/camera/selfie stick. It felt much much worse than I ever remembered it being.
I feel like “an excuse to skimp” has negative connotations but another way of phrasing it is “getting every bit of performance possible out of the package”.
I just listened to a few reviews of the Panasonic G9ii and it has computational photography features too, but that’s not to skimp on optics (very good lenses are available) it is again to get the maximum performance out of the system.
Why would you not spend some compute power to make your camera hardware more versatile?
I don't want that versatility if it's going to compromise the integrity of the image just because some AI was trained on what the average yokel considers a "good picture"
I want the optics and processing to produce the most detailed, widest-gamut, highest-dynamic-range image that is possible, give it to me in a raw form, and then allow me to shape the image to whatever I need it to be
AI R&D takes away from optics/processing R&D, and those are what should be getting 100% of the R&D effort
Maybe it does, maybe it doesn't. I don't know what the engineering structure of the company is like.
I just disagree that the goal or purpose of computational photography is to skimp in other areas. The G9ii has a 25 megapixel micro four thirds image sensor but it can take 100 megapixel photos using an automated process where the image sensor is physically shifted around (thanks to in-body image stabilization capability) while multiple exposures are taken and then combined using computational techniques.
But that is certainly not as an alternative to producing a 100 megapixel micro four thirds sensor, which would have very small pixels and thus poor noise performance among other things. On the G9ii, the computational features purely extend the function of the system. You could argue maybe they would have gone to 30 megapixels or 35 if they did not have this alternative, I don't know.
If you want the absolute best possibly image quality, you simply will not want a pocket action camera. But all cameras will adopt more computational techniques and I don't think it's correct to say they are doing so because they wanted to skimp in other areas. Action cameras already skimp on lens quality to get a pocket form factor, not because they are adding AI. Additionally on cell phones, the form factor has always been the reason for low quality lenses and small sensor size. AI clearly has a place in extending the capability of a camera system with other physical constraints. Perhaps some companies will use that as an excuse to use cheaper optics than what is possible, but I would argue we cannot infer they have done that simply because computational photography is used. Computers are simply cheap and these techniques will get added because they have zero or very low BOM cost impact.
And that just begs the question of why... if we are talking about the incredibly-contemptible cop-out of "the average consumer" than this 100 megapixels is literally for the marketing bulletpoint. 25 megapixels is more than enough for those clowns.
Again, we can figure out better optics and processing without having to computationally invent details that didn't exist.
We are seeing the same thing happen with DLSS, new games are coming out with frankly unacceptable performance and the cop-out is "just use DLSS"
> 25 megapixels is more than enough for those clowns.
I find it irritating to talk to people who speak this way, just FYI.
> Again, we can figure out better optics and processing without having to computationally invent details that didn't exist.
Computational photography is not just inventing details that didn't exist. The 100 megapixel mode on the G9ii physically shifts the sensor a minute amount to collect details which are there but fell in between the pixels on other exposures, and then it computationally combines multiple frames to reconstruct the details. I want this functionality because I print my photos in large format and the printer is capable of reproducing MUCH more detail than my current 21Mp camera can.
> we can figure out better optics and processing
There will always be a limit to optics, and "processing" is always going to be used. Whether a marketing department calls that processing AI or not doesn't matter. Every camera has a manufacturing budget to fit a target price, and that will always mean trade offs and choices made in lens quality. Spending on "processing" will always be part of the calculation. Even on a high end micro four thirds camera, there is just a limit to what you can do even with expensive lenses, and computational techniques are going to be used to optionally extend the capability.
Are you trying to print billboards? Absolutely nothing about large-format printing is 'mainstream'. Sensor fuckery is still a hack that can't do many types of shots (any kind of quick movement for example) while leading people to believe you can. That's the very definition of stuff you should be using specialized hardware for. Are they going to incorporate a 10+ bit sensor and at least an sRGB color space too? Otherwise it's like putting 16gb of VRAM on a geforce 1060
I've done large format printing, you need at least those things in place for the print to come out accurately.
> I find it irritating to talk to people who speak this way, just FYI.
I find the gentrified, milquetoast corpo-speak that seems to be the norm here to be just as annoying
I don't think you're wrong exactly, but I would say that computational photography is an explanation for why you might pick this camera over similarly spec'ed ones. Most people never approach the fundamental physical limits of their cameras because they are quite hard to use optimally - I think there's a lot of space for better software ("ai" or otherwise) to help out.
I’m not sure how to understand this, surely doing “computational photography” (combining exposures, AI denoising and sharpening, smart postprocessing) is in every way more complex than “traditional processing” (camera eyeballing the white balance, applying a tonal curve and calling it a day).
DC Rainmaker's review of the GoPro Hero 12 had a section on overheating. Overheating typically occurs when there no airflow around the camera. For the Hero 12, GoPro removed the GPS chip, which has reduced the amount of overheating - recording time more than doubled compared to the Hero 11.
One Insta360 Ace Pro review on YouTube compared the Ace Pro and GoPro Hero 12 and the Ace Pro kept going for at least 1h15mins while the Hero 12 shutdown after less than 30mins.
Sarcasm aside, this is mostly about their custom ML denoiser for the video:
>Using an onboard 5nm AI chip, the Ace and Ace Pro cameras use a custom-trained AI neural network to denoise the footage for a clearer image in real time, Insta360 claims.
ML denoisers are pretty useful in high-quality photography if done right. Photographers were initially skeptical at large, but quickly accepted it, as products from Topaz, DxO, and alike were giving excellent results. You are extremely unlikely to stumble upon any invented detail in a denoiser like this.
Also, any denoiser is 'inventing' details. You can't know for sure what the value of the pixel would have been absent noise. It's always a probabilistic guess. Guessing based on higher level details just produces different kinds of artifact (and possibly more noticeable ones in some cases).
When I ride my motorcycle with my insta360, the difference between one gray pixel and a slightly different gray pixel for the asphalt is really not something I care about. What I care about is that I can't notice it. If the denoising model does a better job of making up convincing pixels, that's a win. The outcome I'm looking for is a video with fewer visible artifacts caused by the limitations of the hardware.
I have to quibble on this one for technical correctness. These denoisers remove detail, not add them. The assumption is that you should have flat and even areas, not finely pixelated areas.
DxO DeepPRIME XD (their newest and greatest AI denoiser) seems to actually invent details. If I take a picture with my mirrorless of a bookshelf from across the room at ISO 32000 and let XD go crazy at it (slider to the right), it reconstructs gibberish letters on the book covers.
If I don’t let it go crazy and keep it subtle, the results are incredible. Tbh for my camera (EOS R10) and DxO, low dynamic range, not noise, is now the limiting factor for high ISO photos.
Hmm, maybe it sounded like I was saying that ‘inventing’ details is a bad thing. The point I was trying to make was the opposite. Any attempt to remove noise will involve ‘inventing’ pixel values based on probabilistic information, and it doesn’t really matter that much (to me) if that information is derived from the values of a few neighboring pixels or from a higher-level model of the scene.
