Even if you have good computational photography software in the phone, a good starting point for light collection matters just as much: rather than megapixels as pointed out already, focus on sensor size, and as large as possible aperture. Google’s Pixel main camera got the balance nicely it seems (1/2.55" and f/1.7) and iPhone is very similar as well; though the sensor could be larger for both (e.g. 1/1.7" is not unusual these days: Samsung S21, Sony Xperia 1/5 III). I’d also stay away from overly wide focal length of the main camera: 26-27mm equivalent is ok, 24mm looks too distorted in the corners (looking at you Sony). OIS is also very desirable.
1 Like
I agree. I’d even say go for a ~50mm equivalent if you can. On my iPhone I barely use the ultrawide but the tele lens I use all the time.
Computational photography is hyped for a very good reason IMO. My first gen Google Pixel with a GCam app in Night Mode captured files that came eerily close to the low light shots from my α7 III. However, these Night Mode captures were super cooked JPEGs which were pretty much unworkable from a post processing standpoint.
This brings me to Apple ProRAW — which is where things are getting really exciting to me. The critics say: it’s not a real raw file! Which is of course true as it’s “nothing but” demosaiced data in a rebranded Linear DNG file (which has been around since the age of dawn).
The big difference from the Linear DNGs we used to have is that these are the result of multiple exposures and some computational trickery. Cooked? Yes! But, much better than a “real” (mosaiced) DNG file from your phone. You get some of the pros of computational photography and you also get a very flat file with plenty of editing headroom.
So, if I was truly interested in phone photography (which I’m not), I’d get the iPhone 12, without hesitation. I say that as a LineageOS user that hates iOS (despite having used an iPhone at work for the last 10 years).
that’s interesting. I am strongly against iphones because of… lots of things, but the most practical aspect, the less “philosophical”, is price. As I said above, I can’t understand spending close to or more than a thousand euros or dollars for a phone – when a google pixel or similar does the same things for half the price (and I still see a Pixel as a bit too expensive for me).
However, I did try to work with raw files from my Pixel and they are next to useless and much worse than the jpegs; as you said, you can’t do much more with these jpegs.
So your comment on these pseudo-raws from the iphone is interesting – it would be nice to see and example btw.
I certainly hope that google will do the same for the Pixels (or android phones in general).
1 Like
Note that GCam also allows you to save the ‘raw’ (DNG), even in HDR+ Enhanced and Night Sight modes.
AFAIK, the Google camera app on the Pixel will export a 14-bit Bayer DNG after all the multi-frame HDR+ wizardry (see RPU samples for 4 XL and 4a).
1 Like
I have a 3a and I initially thought as @aadm stated about the raw files but I found one thing was the heavy vignetting made them look bad and hard to edit. RT will use the gain map to correct this using a Flat field correction from the meta data. It makes a big improvement or I cheat in DT as there is no such support and use the lens settings for the Huawei P10. Not technically correct but the raw files look much better IMO . When I started to really look at the jpg the treatment in many photos was way too harsh and colors were sometimes actually washed out from what I could get in the raw. So now I still will use the jpg but surprisingly I did an icc profile using the colormatch script with a jpg and raw pair and I get nice results with raw files and find I use them as often as I choose the jpg. I will go back and try to do an edit to this post with some examples if I get some time…
Thank you guys, thank you to everyone who answered here. It’s very useful and educational.
I think I will get S10, since it’s better than others I mentioned and I will get the Pixel’s camera app to get the most out of it.
There is only one thing, I checked the M51 specs again and it seems like it has much bigger sensor than S10, because it’s 1/.73" vs 1/2.55". But then I checked Google Pixel 4a and it has 1/2.55" as well. So is M51 much better than both S10 and Pixel 4a, sensor-wise?
And the 52mm camera that isn’t present in S10e (which is significantly smaller than S10, which means more desirable for me) is only 1/3.6". Would that make the S10 much better than S10e (since it has this additional camera) or not (because the additional camera is quite poor anyway)?
Here are two DNGs (HDR+ Enhanced and Night Sight mode) from GCam on a OnePlus Nord. darktable complains about not finding the matrix. They don’t look sharp at the pixel level. They seem to be mosaiced; I don’t know about the bit depth.
HDR+ Enhanced mode:
IMG_20210630_140220.dng (10.3 MB)
Night Sight mode (yes, I know, it’s broad daylight):
IMG_20210630_140232.dng (9.7 MB)
Currently ART is the only OSS raw processor that supports the GainMap (lens shading map) included in dngs from the Pixel and other phones. RawTherapee, darktable, etc, do not support it yet. When it’s included in the dng file and is not a no-op, it’s very important to apply it because it’s not just vignetting but it affects each color channel differently. So if it’s not applied the colors can be correct in the exact center of the image and there’s a heavy color cast closer to the edge.
The example in this thread shows how incorrect the colors can be when the GainMap is not applied. In that case it was mistaken for an incorrect color profile.
1 Like
I must have been in Art but I swear I did try it in RT…funny…too many programs on my PC 
I think open camera saves tiff’s in a DNG container. You can’t extract a jpg preview from open
camera DNG and the tags showed references to tiff somewhere I think. They are on average 10MB larger than the DNG from the native pixel app so that might make sense. I may be wrong but for sure they are not the same DNG.
Just too a quick shot out my open office door… one pixel and one for open camera…which shoots a very flat raw for me…
IMG_20210630_130155.dng (23.3 MB)
PXL_20210630_165800503.NIGHT.dng (12.7 MB)
TIFF is also just a container. DNG is a TIFF.
The difference in size you’re seeing is most likely due to compression - original Pixel DNGs use JPEG lossless.
Yes, without having inspected the files it’s very likely DNG lossless compression vs. no compression. Open Camera does not compress, while GCam derivatives do. A linear DNG has three channels and a mosaiced bayer raw file has one, so one can guess by looking at the size what flavor it is. 
Ya the open camera DNG tags say full size preview uncompressed but I have not found any thing that could extract a preview. I think most tools expect a jpg for the preview not a tiff so I think the open camera DNG is just the image as you say DNG is a container for tiff file…I am sure there is a nuance but no utility I found or exif could pull out a preview so that is why I surmised what I did…
I believe that kmilos was not talking about the embedded preview data “being a TIFF”, but the mosaiced data (raw DNG) or demosaiced data (Linear DNG) itself conforming to TIFF/EP. If I recall things right Open Camera does not embed any preview data at all, just uncompressed mosaiced data, but don’t quote me on that.
1 Like
Ya that is what I was trying to articulate badly I guess… I think it just has the data no preview…
Ah, I see. Yes, that is likely the case. You can save plenty of storage by running them through the Adobe DNG Converter with lossless compression enabled.
I haven’t used OC too much I have ON1, Open Camera, Hedgecam, and the Google app. Mostly just shoot in the google app and those files are pretty small around 13MB.
Instead of speculating, just run exiftool -a -u -s -g1 or exiv2 -pa on your files.
For the OpenCamera one:
Indeed, just single image (IFD0) with the full resolution Bayer raw, uncompressed, and from a single frame (10-bit)
For the original (Google camera app) one:
- A small resolution JPEG preview in IFD0
- The full resolution Bayer raw in the SubIFD, but: 14-bit after multi-frame processing, and JPEG lossless compressed
@mikae1 Adobe DNG Converter will indeed provide the identical space saving, as the same JPEG lossless compression will be used (the only one allowed by the baseline DNG spec).
1 Like