P: Introducing the Project Indigo camera app

[Rikk Flohr: Photography]

This post applies to the Project Indigo iOS camera app. 

Adobe Labs is excited to share an early look at Project Indigo, an iPhone camera app we've started to develop, to get feedback from the photography community. The app offers full manual controls, a more natural ("SLR-like") look, and high image quality in both JPEG and raw formats. It also introduces some new photographic experiences not available in other camera apps. For more information on the underlying technology, please refer to this Project Indigo blog post.

Before you start with Project Indigo 

• We recommend using Project Indigo on iPhone 15 Pro/Pro Max or newer devices.
  (Also supported are 12 Pro/Pro Max, 13 Pro/Pro Max, and all 14-series devices.)

• You should have at least 1GB of storage space left for the app, the downloadable AI Models inside the app, and for captured photos. 

• Download the app from the App Store. 
• Check out the Project Indigo FAQ, which is updated regularly with new explanations. 

Recipes for success when using Project Indigo 

To get the maximum out of your images captured with the app, follow these guidelines: 

• When reviewing the results, focus on Project Indigo's more natural look (in both SDR and HDR). If you haven’t done this before, try viewing the images on your laptop or desktop device, preferably on an HDR screen. 

• Capture with both JPEG and raw DNGs with file saving enabled. Project Indigo produces computational photography DNG files, which have the same natural look as JPEG images, but much more latitude for editing after capture. 

• Take control of the camera with the built-in Pro Controls, including controls that are exclusive to a computational camera: Frames to Merge and Merge Method. These may be intimidating for beginners, but with Project Indigo, you can try them for free, and nothing will break—you can always reset the settings to ‘Auto’ and let the camera take back control. 

• Go to the Indigo Labs page and play with the latest innovations our team can offer. These are only available on mobile via Indigo! 

• Be patient! Project Indigo is doing a lot of heavy lifting under the hood, and it will reward you with great photos. In return, it may ask you for a bit of time to set up captures when needed, and to wait a few seconds for the image processing to finish. 

Sending feedback 

Please try the app and share feedback in this community forum thread. If you report a problem you encountered, it would help to include details like which device you are running Project Indigo on, what kind of scene you were trying to capture, what you were trying to achieve with the camera, and as much information as possible about what you like or do not like about the resulting photo quality. Our team will continually monitor this thread to track issues and improve future experiences.  

To improve the performance and results of Project Indigo, it is important that examples of images that do not meet your expectations are forwarded to the team via your report.  A large variety of file formats are allowed as attachments in these forum posts. The best option is to attach your image's raw file directly to your feedback post. Note that there is a 50 MB limit on an attachment's file size. If your raw file is too large to attach, the best option is to share the file via a file-sharing service (Dropbox or similar) and then share the link in your feedback post. Thank you for continuing to provide feedback on the Project Indigo camera! 

Boris Ajdin: Product Manager, NextCam 
 
Posted by: 

[nhan_8084]

Attached here is a night scene consisting of streetlights that is tinted led vs daylight color led or the halogen yellow and orange street light typically seen.  All shot is done without night mode, nothing else touched.

Indigo kept the color rendition this time very good vs stock, the light color was that blue vs the purple tone from stock.  Jpeg from Indigo is very soft, no details at all vs stock which has very good amount of fine details and not extensive oversharpening applied.  The raw of Indigo is great.  I have attached crops of the area of interest showing default profile look of Indigo vs nuking sharpening and noise reduction, and applying color noise reduction to 10 vs 25.  This look should be applied as the default jpeg output as it allows a balance of sharpness to noise ratio, and snr at this iso is expected.  The foliages rendered very nicely imo now.

[nhan_8084]

Here's another low light shot indoor using the 4x tele.  The light source is my tv to the left so it's quite dim.  The speaker grills are not visible to the human eyes from the distance of about 15' to the phone.  Stock did amazingly well picking up the individual mesh sizes of the speaker grill while Indigo cannot render anything at all in the picture.

All shots auto as usual, no night mode used.

[nhan_8084]

@BorisTheBlade I took a look back at this slide you sent a few months back and sent to couple of mobile camera enthusiasts on telegram and they seem to be bothered by the word "processed" in the begining portion of the steps.  They kept thinking that apple's bayer raw is still baked in similar to proraw.  Can you explain what that word of "processed" means?

[nhan_8084]

Another low light scene, with mixtures of lighting that includes xmas decors and street lamp, both different color temperatures.  All shot auto, no night mode in Indigo and in stock.

Indigo 1x shot fars way better than the 4x version, which is extremely unusable vs stock 4x.  I also included the rendered Indigo dng to jpeg that I think is manageable vs the default ouput with settings I used vs the default metadata set by Indigo.

[nhan_8084]

Had to attach more as it has limited me due to the sizes. @BorisTheBlade Tag u to not miss any of this.

[nhan_8084]

@BorisTheBlade I picked the darkest possible spot on a cold, wet, grey skies outdoor conditions possible to test normal and night mode on Indigo vs stock.  Auto night mode stock is 3s, and Indigo is 6s but it somehow does not render better than stock.  The jpeg is still mushed especially the pine needles, crushed blacks (yes it was dark to not see anything with the eyes, but night mode is supposed to be see in the dark, which is what stock did), and normal mode Indigo is unuseable vs stock normal mode.

 In the dng Indigo night mode, shadows are recoverable with minimal noise even with all nr off, so plus for that!  See crops for point of interest.

[nhan_8084]

@BorisTheBlade this set involves 1x, and 4x in same conditions but no night modes used on both stock and Indigo.  I am showing street light color, one is incandescent bulb type very orange about 2700K lighting temp, and one is pure led bulb (yes, I actually looked at the bulbs on the street lamp!) at 5000K ish daylight temp.  I made sure to have the lamp in frame in both sets.

The 1x is good!  The telescope failed again as usual vs stock even at the better brighter led lamp in frame!

[nhan_8084]

@BorisTheBlade This set is same, 1x and 4x vs stock normal mode, but it's special as it contains both incandescent color lamp AND led lamp in the same frame for 1x.  The 4x, I resort to taking the led in frame as it's obviously brighter but the tele of Indigo is still very unusable vs stock.

[leed24280355]

Have the developers considred using tiled JXL for the DNG compression? This reduces the DNG file size substantially.

[BorisTheBlade]

Have the developers considred using tiled JXL for the DNG compression? This reduces the DNG file size substantially.

By @leed24280355

That is an interesting question - I'll check with the dev team on this. Thank you.

[nhan_8084]

Indigo's DNG is already lossless compression.  If it wasn't lossless it would be bigger, based on what I've handled in DNG from Gcam lossless vs no compression toggle.

[nhan_8084]

Happy new year @BorisTheBlade to you and the team!  I hope you guys have a great time off the past few weeks, and a safe break as well.  Just wanted to ask if the team had a chance to download and inspect all the photos I have sent here during my time off, and if the team has any findings to solve the issues I mentioned?  Were the team able to replicate the scene scenarios I sent to see personally these issues that I pointed out as well?  For the next release, what can we expect next?  

Also, when will the beta testers program be rolled out?  Will you guys have the possiblity to have 1:1 direct contact with the testers, to send test pictures and findings, and fixes so that way it will be easier and faster to implement for the final version to the app store?

[nhan_8084]

@BorisTheBlade Could you please respond to this so I can get an idea moving forward?

[nhan_8084]

Hello, here is another sample set vs stock.  The point of focus and tapped to focus using stock and Indigo is the yellow couch cushion.  The distance from phone to pillow is 25ft with the straight view, the diagonal view is 27ft based on the Measure app of iOS.  All shot are auto with the exception of tapping.

The pillow has patterns that Indigo could not resolve, stock although oversharpened did resolve the pattern decently but not the best.  The raw of Indigo cannot resolve the pattern either even if all noise reductions are killed!

 I will include samples from my Pixel 10 Pro XL, using stock google camera and tuned google camera mod using my own parameters from the defined hex address deep inside google camera lib file.  You can see how well the pattern are resolved from my tuning values.  Can the team achieve this since the sensor of the 17 pro max is pretty much identical to that of the pixel 10 pro XL (1/1.28" vs 1/1.3")?

note:  due to the saving from one device to another, the file name got overwritten, please look at the exif to see which shots are the pixel shots.  Crops are included from both devices as well, with the obvious clarity of pillow pattern being the Pixel device!

[nhan_8084]

Ran out of attachment limit amounts so here are the crops @BorisTheBlade 

[Telllinex]

I guess you'll help more by providing exactly what you did change in a gcam mod lib, since PI=Gcam, I believe this can be fixed, as well as excessive luma denoising. The result from your IMG2227 should definitely be achievable on PI, but this has to be changed by devs and tested. From what I remember, modded Gcam was good in one scenario, but worse in others - for example, when you have a strong dynamic range difference - you start to see those motion artifacts that AFAIK have not been fixed since GCam 7 or 8 came out with Merge 1,2,3 modes - 3rd is what has been used since then, it improved shadows because had "shasta factor" parameter, but had some areas where denoise failed for some reason.  Because of this, we had an invisible config button near a shutter button and switched configs for day/night scenarios.

Additionally, since 48MP is blocked without getting it BEFORE the ProRaw pipeline, I think the app, as of now, is pretty ok to use, even with the 12MP limitation.  

[BorisTheBlade]

I guess you'll help more by providing exactly what you did change in a gcam mod lib, since PI=Gcam ...

By @Telllinex

One correction here - Project Indigo and GCam, although in some sense similar, are *not* the same. While the computational photography spirit behind them is very close, how they are implemented is not. Any assumtions stemming from "In GCam it works like this so it should be the case in PI as well" are likely not going to be accurate.

[Telllinex]

I understand that, because I already see that it outputs different (better images), just trying to help here and wish it becomes better!
BTW, is there any chance you can add denoise sliders somewhere deep in settings (perhaps under experimental settings or something like that)?

[BorisTheBlade]

Post-processing sliders are on the ToDo list, yes. 

[nhan_8084]

@Telllinex I don't want to overstep my boundaries with the Indigo devs, after all they are the true devs.  I am just a novice who are EXTREMELY DEDICATED to mobile photography as mentioned here.  What I am simply showing to Boris and the devs here is that even with Gcam who was Levoy's child had issues as is in stock form vs what I have done myself.  Also Indigo is not the same as Gcam I'm sure due to just being developed differently on the iOS platform.  If it was on android, I can see it being "ported" but definitely not the same for iOS.

I have fixed all the issues you mentioned of merge method, and shasta as you see my samples are high dynamic range and no issues with shadow recovery.  There's no need to have multiple configs anymore.  

[Telllinex]

I am glad that you had them fixed(though I still see my old friends testing a lot of parameters on the latest versions). I wish I were in the loop about all of these gcam things, but I left that community long ago after moving on with iPhone. The last Android phone I had was Xiaomi Mi 8, and Gcam was perfect on it - IMX363 did its thing. 

[nhan_8084]

The Mi 8 series was what the gcam modder BSG has/had so he can do his things efficiently.  

[BorisTheBlade]

Hello everyone - Happy Holidays from the Project Indigo team! Thank you for the support and feedback you have been providing since we launched back in June, and we are excited to keep adding improvements to Indgio going forward. It has been a bumpy ride so far, but that is not unexpected for an experimental app. We will continue to iterate on various aspects of the app, as well as add new feature, and we are looking forward to receiving more feedback from our patient users!

[powerful_Elixir5E29]

@BorisTheBlade I think you should definetely focus on improving the super res zoom algorithm because in areas with subject motion it produces strong artifacts, quality degrades significantly even more than to a single 12mp fallback in these areas and the super res zoom does bad with foliage or grass like textures it smoothens them out and overall even in dng there is strong denoising pre-applied which you cant remove even in lightroom. There are also alignment issues. I think the team can definetely do a better job of also detail preservation even in moving subjects like gcam super res didnt have many of those issues. Also i want to ask you this question: Will the improved super res zoom on a 12mp bayer sensor produce the same 2x zoom dngs as apples 2x crop from a 48mp quad bayer sensor in a moving subject scene specifically since they both fall back there to single frame demoisaicing and as we know quad bayer is hard to demosaic and has aliasing artifacts or apples 2x crop will always beat your software?

[nhan_8084]

I'm pretty sure apple's ISZ will always be superior to any SR because you have true access to the entire sensor pixels.  However, speaking from Gcam standpoint when it originated, their SR is NOT that far behind of optical zoom back then for the non XL/Pro series.  Indigo has a lot of barriers especially with iOS locked down while Gcam had full access to the entire phone/sensors.  Also I'm not sure what you mean can't remove the strong denoising in lightroom of Indigo's DNG.  Their DNG on optical shot are fine in most cases vs their jpegs!  The only issue are foliages which they mentioned is high frequency rendering that is being looked at.  I have showed samples to Boris and team above of the case in point couch pillow that oddly could not be rendered at all in both dng and jpegs.  @BorisTheBlade it's a good sample set so please make sure the team look at them for analysis!  Once I see that Indigo can render that pillow cushion from that distant in the next few updates, I know that imaging from Indigo will be extremely amazing!