One of the few times a resolution increase does dramatically increase the clarity—probably due to the camera module also having a much nicer lens to match the higher-resolution sensor.
It also helps that the sensor is slightly larger (though still a bit smaller than the one on the HQ camera.
I think that opens up a discussion about sensor size not being the biggest factor in image quality and whether it's worth it or not in practical terms to have a large sensor. Personally I have found that my full-frame DSLR gets a lot less use than my phone with a sensor size that is a fraction of the DSLRs! Yes, artsy bokeh shots/low light photos (although this is subjective due to recent developments with smartphones) are easier to take with a larger sensor, but I almost feel like they're going obselete now at least for every-day use.
I still use my old Nikon D700 for some things, like as a timelapse camera or for simple product shots.
The 12 megapixel sensor in the 35mm format is still much better at gathering light (without any fancy AI manipulation) than any modern small sensor camera, and the few Nikon lenses I still have around—one manufactured in the 70s—are extremely sharp and still great for many situations (though lens coatings, even on tiny sensors and smartphones, are vastly improved nowadays).
I agree - I have a D600 myself and still use mine occasionally, Nikon's backwards compatibility is great! I am equally impressed by the AI manipulation by modern smartphones and their ability to take handheld shots that say a 10 year old APSC camera (with a huge sensor comparitively) might produce a blurry mess with. That being said I don't really know what modern APSC cameras can do nowadays!
Your phone camera's images look 'great' because they're been very heavily processed (now with all sorts of "AI" nonsense that is leading to us not being able to trust photos from cell phones) and are taken with a wide angle, small aperture lens that yields very wide depth of field.
Compare images side by side, especially at indoor lighting levels, and you'll see just how much more detail even a 10 year old micro four thirds or digital SLR captures compared to the cell phone camera. Capture some images around 6400 ISO, too, and you'll really see what a mess cell phone cameras are.
There's no getting around physics - the larger the sensor well, the more light it can collect and convert to electrical impulses. Top end cameras from Canon, Nikon, and Sony have been bumping up against the limits of physics for a while now - efficiency, noise, etc.
My m4/3 camera which is now 10 years old still takes better pictures than any current cell phone camera on the market.
I've done some work on barcode scanning, so I've spent quite a bit of time taking images of reference patterns and comparing them in minute detail.
In my experience, the current iPhone SE produces very slightly more detail than a m4/3 Lumix DMC-GX80. Which is crazy, because the iPhone's lens and sensor are tiny compared to the m4/3 camera.
We still have some ways to go in low light. My room is lit by 2 40W incandescent light bulbs. This is more light than a typical urban sidewalk or restaurant at night. This is low enough light that the difference in video quality of my iphone 13 pro (2021) and nikon D750 (2014) full frame camera is night and day.
However I agree that there are times now where I can confidently use my phone where 10 years ago I would have had a dedicated camera. Phones are only going to keep getting better!
I also wanted to highlight a discussion about this camera from two weeks ago, from user lemariva: https://news.ycombinator.com/item?id=29727231 (has some other interesting details).
Are there any affordable development platforms for those cameras right now? Paying 100 € for a second hand Raspberry Pi and another 25 € for the Camera seems like a bad idea when I could get a USB GPIO adapter and a smartphone with a better camera and touchscreen for less.
Raspberry Pis (at least Pi 4 model B) can be found new at list price in a few places, but you have to be patient.
Of the 8 or so Pis I currently test on, the average time from when I placed the order to when I got the Pi was about 3 months (the CM4s I bought took about 8 months).
But I did get them all at list price, plus a small shipping fee.
Do you have any specific links of suppliers you recommend checking? All of my common ones, including my local MicroCenter, are almost perpetually out...
Thanks - I've been keeping track of Pishop and Micro Center (always forget the space...), forgot about PiHut as I usually don't need to check stores for other countries (though they've got Pi Zero 2 boards in stock, so may be placing an order...).
In an ideal world getting a random android phone camera component and getting the appropriate linux driver should be easy but I guess not enough people are interested in doing that.
Always wondered why you upload yt videos in the morning (in US). From my limited experience it appears the “algorithm” seems to favor videos released in the (early) evenings.
There are probably some stats out there on release time/viewership, but I wouldn’t be surprised if there is an optimal release time (“its 5’oclock somewhere…”)
Every channel's a little different—I might get more initial views if released later in the day, but it seems like YouTube's algorithm is biased more towards how the clickthrough rate and watch percentage go, then it gives another pass at a new audience on day two and sees if it changes—if it does, you can get a significant boost on the 2nd day (or I've seen up to a few days later, for a good video).
The basic answer is more, "it's complicated"—but for me it's more convenient to release in the morning so I can check in on comments early on (sometimes I get some new ideas or re-test based on early comments for things I missed!).
I think this will unlock a lot of University robotics/computer vision projects. The existing options for doing embedded computer vision were pretty terrible.
I believe most modern smartphones have more powerful cameras than ArduCam. They're more expensive, but broken / older smartphones can be had for cheap.
I believe the grandparent is suggesting using a smartphone in place of using a raspberry pi, not trying to extract the camera and connect it to the pi.
There's been about 5 years now, where the common refrain is to just use a smartphone camera. And in some cases people are literally writing an app and strapping a cellphone to a drone, but it's not ideal. This arducam camera is the closest thing to an embedded cellphone camera that is actually usable in robotics.
I;m not sure but I'd expect autofocus doesn't work well with computer vision. Most of the lenses and sensors I've worked with in raspi land have various image aberations (spherical, color). They are very different from the computer vision cameras I have.
This is great. I don't care at all about the 16 megapixels (to misquote Bill Gates, 6.4 megapixels should be enough for anyone), but AF and reasonable close focusing have been available in cheap smart phones but missing from the Pi til now. There hasn't even been a usable way to read a QR code with a Pi until now. The $25 price is great too.
The Pi foundation should have shipped something like this long ago. Maybe they can work out something with Arducam to make it a standard Pi product.
While it is non-intuitive that the camera doesn't focus on a center point, the good news here is that it should be possible to control the AF motor in a more fine grained manner. I'm guessing it would take some firmware hacking, but I don't see any technical reason why it wouldn't work.
True. I hope someone with more knowledge will work on some algorithms for this camera that are reasonably fast and can do contrast detect in a given region, or based on face recognition.
32 comments
[ 0.17 ms ] story [ 78.3 ms ] threadIt also helps that the sensor is slightly larger (though still a bit smaller than the one on the HQ camera.
The 12 megapixel sensor in the 35mm format is still much better at gathering light (without any fancy AI manipulation) than any modern small sensor camera, and the few Nikon lenses I still have around—one manufactured in the 70s—are extremely sharp and still great for many situations (though lens coatings, even on tiny sensors and smartphones, are vastly improved nowadays).
Compare images side by side, especially at indoor lighting levels, and you'll see just how much more detail even a 10 year old micro four thirds or digital SLR captures compared to the cell phone camera. Capture some images around 6400 ISO, too, and you'll really see what a mess cell phone cameras are.
There's no getting around physics - the larger the sensor well, the more light it can collect and convert to electrical impulses. Top end cameras from Canon, Nikon, and Sony have been bumping up against the limits of physics for a while now - efficiency, noise, etc.
My m4/3 camera which is now 10 years old still takes better pictures than any current cell phone camera on the market.
In my experience, the current iPhone SE produces very slightly more detail than a m4/3 Lumix DMC-GX80. Which is crazy, because the iPhone's lens and sensor are tiny compared to the m4/3 camera.
However I agree that there are times now where I can confidently use my phone where 10 years ago I would have had a dedicated camera. Phones are only going to keep getting better!
Of the 8 or so Pis I currently test on, the average time from when I placed the order to when I got the Pi was about 3 months (the CM4s I bought took about 8 months).
But I did get them all at list price, plus a small shipping fee.
Micro Center always had some model in stock until last September, now they only seem to have Pi 400s available.
In an ideal world getting a random android phone camera component and getting the appropriate linux driver should be easy but I guess not enough people are interested in doing that.
There are probably some stats out there on release time/viewership, but I wouldn’t be surprised if there is an optimal release time (“its 5’oclock somewhere…”)
The basic answer is more, "it's complicated"—but for me it's more convenient to release in the morning so I can check in on comments early on (sometimes I get some new ideas or re-test based on early comments for things I missed!).
The Pi foundation should have shipped something like this long ago. Maybe they can work out something with Arducam to make it a standard Pi product.
Currently there is a shortage of RPi's, they take long to ship, costs more and somehow also have a lower quality compared to the older RPi's I have.