Artificial Intelligence on PhotoLab: what do you expect for?

Hello,

After the first implementation of an AI tool in DPL with DeepPrime, here is the summary of our AI wishlist:

#1. AI mask
Select quick masks of parts of the image, assisted by AI. For example: sharpness mask (automatic detection of areas of maximum sharpness with an opacity gradient depending on the transition between sharpness and blur); sky; faces and skin; vegetation, etc.

#2. AI highlight recovery
An AI tool for recovering/simulating “lost information” in overexposed areas in a realistic way. It will be very useful for cloudy or white skies in landscape pictures, or colored spotlights that appear white because they are overexposed. Maybe an AI algorithm could restore their natural color, despite their overexposure.
In short, a way to improve the current performance of DPL in the recovery of highlights (a weakness for the moment, in my opinion) and I hope, to overtake the competitors.

#3. AI restoration after perspective transformation
It would be very useful to widen the crop after ViewPoint transformations. By recreating “out of original picture” areas with AI algorithms, it would be possible to complete the empty parts of the frame by a “realistic” pattern, copied from closest areas of the picture (sky, vegetation, grass, rocks or concrete ground, etc.).

#4. Noise removing improvement
Done: it is deepPrime! :+1:

#5. AI-assisted DAM
It would be very useful for those who use a DAM.

#6 AI sensor’s dust removal
I would like the AI to successfully perform this step without any manual intervention.

#7. AI help to select to best picture of a series (new)
AI could assign a score to each picture to help choose the best one among a series of similar pictures. Examples of criteria: sharpness score (the best sharpness), portrait score (those where the eyes are not closed, orientation of the gaze, smile, etc.), bokeh score (where the background is clean); strength of the composition (rule of thirds, symmetry, etc.). A help to eliminate quicky the pictures that will not never be good, to avoid demosaicing them just to find out they do not worth it.

#8. AI Object detection and replacement (new)
You select very roughly an area of the image, AI precisely selects its boundaries and replaces it with the background. Magic!

Please don’t hesitate to come up with new ideas: a new DPL is coming!

Regards,

3 Likes

Some solid suggestions.

#1, #2 and #6 seems useful and make sense for PL.

#5, #7 is better done externally because it will create bloat

#3 and #8 are tricky since there is a lot of manipulation involved which is counter to the premise of PL, we have Photoshop/Affinity Photo for those tasks

But the latter is a thin line since ultimately there is already some of that going on.

I would personally suggest Auto White Balance or Real White Balance as a good task for AI.

4 Likes

Tout à fait :slightly_smiling_face:

1 Like

#2 I Think this is more for a pixel editing app like Photoshop of Affinity. But…

Smart Lighting is already a AI tool. It isn’t advertised as AI, but is in a very smart way handling the shadows and highlight.

I really would like to see the RAW editing tools to be devoloped first. Think like Auto color correction, AI skin improvement (faces are already recognised), AI (output) sharpening.

3 Likes

While using iMatch a DAM I recently trialed and then purchased ( reasonably priced) is Excire Foto. The AI tagging is quite impressive. In fact, at times I’m blown away by some of the keywords that are AI added…possibly overkill! There is a German and an English version at this point. The US version is $59 USD as of today. For those who may want to give it a 30 day trial the US website for Excire Foto is
Excire Foto_US

Barry

Hi Barry,
I’ve also tested the trial, but there are some limitations in my opinion.
The last version1.1.1 tagged some photos with food, but there are things I do not eat normally.
So i see little elephants in the zoo, a wooden puppet and other strange things :grinning:
You are not able to change the excire keyword hierarchy and, and and.
The idea is good and I think in one year or so, they will be ready for the workflow of a semi prof photographer, but at the moment I’m afraid to spend time for 30k Fotos in an non prof way.
The test I made was wit 1500 pictures, the same I use for every test for photo software thats 5% of my photos.
What is very good is searching for persons based on a photo with this person…that works great
I have tested since the first version and you can get a lokk at Excire Forum - General Questions for first impression.
And some infos here DAM-Take a look at Excire Foto - News and announcements - DxO Forums

stay tuned

Be aware that IMatch includes AutoTagger IMatch Help System. You would need to set up an account with one of the supported cloud-based services provided by external companies like Google or Microsoft.

I absolutely agree with your assessment Guenterm. Some of the added keywords are a bit wacky for the photo. I purchased it basically for two reasons;

  1. I like supporting companies that have a vision as to where it wants to go, even if it’s not there yet.
  2. Along with some effort on my part and the capabilities o Excire Foto, I can easily and very quickly find, for example, a blue sky with no clouds or many clouds or whatever I’ve keyed in. Also, finding textures and overlays is very quick and easy because I had Excire Foto analyze the external drive where all my resources are stored.

Guenterm, I 100% agree it’s a work in progress. I’m curious to see where it goes. But, I guess that philosophy is in part why we do photo-editing, that is, here’s my vision…where will it go.

Barry

1 Like

Dear Barry,

that’s correct…I was part of the first affinity photo testing, and glad to see companies that develop good software not working in a cloud system. And yes…the search function for faces, persons, colors, geometry and so on are very impressive, but not so necessary for me.
The keywording was the only thing i looked for as I get the first informations about excire.
What I don’t understand is that Excire search is better in hierarchical keywording( I’ve tested in my old LR) than Excire Foto. But I want to leave LR.
My dream is working with only DXO for sorting, keywording, developing and using AFF Photo and Designer for all the other stuff.
Dreams will come true in the future :innocent:

Allways good light to you

Guenter

1 Like

Personally, I’m OK with AI on all aspect but “artistic” ones.

  • Suggesting keywords based on DeepLearning,
  • DeepPRIME,
  • etc.
    These are examples where AI applies well imho. They are practical and non-artistic stuffs. For DeepPRIME, there’s no “artistic” choice involved. Basically we want the least noise while retaining the most details and that’s it.

I wouldn’t want to see DxO turning into some kind of AI-powered software that “improves” everything for you. I still think it’s cool to say that the end result is “my” end-result.

2 Likes

You may be interested in this:

Looks like I’m late to the party but….
I think I don’t need AI sky replacement because I just got it yesterday with a new Photoshop.

First thing I would like to have is Auto Mask with more AI capabilities. Auto Mask as it is works fine most of the times but sometimes it just can’t handle certain photos like hair or fur for example.

Second thing I would like to have is a separate AI slider for highlights. This slider would not interfere with mid tones. It would be something like ‘’DeepHIGHLIGHTS’’ or ‘’Highlights Plus’’.
It would have two functions – to recovery highlights and also to recreate highlights (artificially rebuild/rescue/restore part of the image with unrecoverable highlights).

1 Like

Joanna wrote:
Please don’t take this the wrong way but, if you used ETTR (expose to the right) metering, then you wouldn’t need to fix this kind of problem.

I think this is a matter of opinion from which side do you look.
I could also say – if you use exposure to the left you won’t have to deal with shadows in Photolab; you just replace the sky.
I could also be a smart a** and say – if someone is such an exposure master he could expose to left and right and shoot perfectly exposed jpegs straight from the camera so there is no need to use Photolab to ‘’repair’’ the images.

Mark wrote in other thread:
I’ve never run into a properly exposed file that I could not recover shadows and highlights on with PL4 using a combination of smart lighting, selective tone sliders, advanced contrast sliders(Filmpack 5), local adjustments and very rarely exposure compensation. I understand that this process is too lengthy if you have 100s of photos to be delivered under a deadline. In which case I suggest just using C1 or LR and not altering PL4’s exceptionally good highlight/shadow recovery toolset and leaving it as it is for those of us who like it and have the time to use it properly.

If I understand correctly: file is PROPERLY exposed but someone would still had to use a combination of 5 (five!!!) tools in Photolab to deal with shadows and highlights.
I wonder how many tools and combinations would you need if the image is a little less ‘’properly’’ exposed?

And if I continue - ‘‘time to use it properly’’? :

‘’ … leaving it as it is for those of us who like it and have the time to use it properly ‘’.

Hmmm; do I need to think first for a few days if I have enough time to ‘’use Photolab properly’’ before I buy it? How much time do I need to have? Half an hour per image? Perhaps more?

We have espresso machine at home. I don’t like it and I don’t use it. I use caffettiera.
But just because I don’t like it and I don’t use it I DON’T LIMIT the others. WHO AM I to tell them what they need and what they don’t need. If they like it they can have it and I have nothing against if they use it.

2 Likes

Perhaps I should clarify:

By “properly exposed” I mean that the highlights on the RAW files are not overexposed to the point of being completely white(no ink applied to the paper on a print) with no detail left to recover. No app is capable of recovering these lost highlights without manufacturing false details to replace the lost ones. Anything “less properly exposed” would fall into this category. The detail is either there to be recovered or not there at all.

Five tools? I should have said " using a combination of up to these 5 tools". Only rarely do I need more than smart lighting, selective tone and advanced contrast. The selective tone Highlight adjustment alone will recover the highlights but, as you point out, also affects the mid-tones and shadows as well as their selective contrast ranges so the other tools are used in order to compensate in a seamless manner.

I don’t know what your thought process is or should be before you buy a photo editing app but most files take 30 seconds to 5 minutes to recover highlights and shadows and adjust the mid-tones to seamlessly match. This process may take a little bit more time if local adjustments are required, but nowhere near 30 minutes. Still 5 minutes per image is too long for someone with 500 images to deliver tomorrow. I suggested the use of LR or C1 in these cases.

I’m not sure that I understand your espresso machine analogy here, but I’m guessing that the espresso machine is the more complicated process and represents PL4 here. If so, I’m not trying to force you to use it, that’s why I suggested the use of LR or C1 if the process was too lengthy. I love espresso and don’t mind the process of making it, so I just don’t want the espresso machine to be disposed of.

I want to evolve.
I want to nail the exposure the first time.
I want to find the composition that no one but me thought of.
I want to push myself beyond what I did last week. Last year. Or ever.
I want a tool that refines and encourage my ideas of creativity.

I don’t want something that replaces the sky I invested my soul in after travelling thousands of miles.
I don’t want something deleting objects I that moment choose to capture.
I don’t want something interfering with my thoughts.
I don’t want an application to outwit my work.

I want to go to bed knowing that today I have out done myself ones again. And that tomorrow I’ll push my creativity beyond what I did today and ever have done before.
I want my tools to support me and my vision - not to replace them.

And yes. It’s way to late and today I did not push myself enough.
Tomorrow - I will.
No one else will do that for me.

4 Likes

Laudable but lofty goals. I wish you the best in your efforts. :smiley:

2 Likes

« You used to think that it was so easy
You used to say that it was so easy
But you’re tryin’ you’re tryin’ now
Another year and then you’d be happy
Just one more year and then you’d be happy »

1 Like

Still 5 minutes per image is too long for someone with 500 images to deliver tomorrow. I suggested the use of LR or C1 in these cases.

I suppose that’s why professional photographers are using LR and C1P.
I hope this is not official DxO philosophy: ‘’If you don’t like Photolab use something else’’.
I think you know what sooner or later happens to companies with philosophy like: ‘’If you don’t like it go somewhere else’’?

I’m not sure that I understand your espresso machine analogy

It was an example – different users have different needs. In our home there are lots of different ‘’tools’’. I use some and I don’t use other but I don’t tell no one what he needs and what he does not need.
The same goes for new tools in Photolab.
You stance ‘’go use LR or C1P’’ will not bring new users and more profit to DxO.

1 Like

Thanks for all your last contributions.

What do you expect for AI features?

  • Best results (e.g. #2. AI Highlight recovery)
  • Improved ease of use (e.g. #6. Sensor dust removal, #1. AI mask)
  • Extending capabilities (e.g. #3. Out of the frame filling with AI after a perspective transformation, #8. AI Objects detect, and replace)
  • New features (e.g. #5 AI-assisted DAM, #7. AI photographic criteria scoring / help to selection)

and overall, a high level of reliability and performance for the best efficiency. No one here is talking about giving AI the responsibility of making photographic or stylistic choices.

Efficiency is really the key: if AI helps me to do what I want better than me, quickly and easily, I want it! And if I no longer need to switch to another software to complete my treatment, that’s fine.

1 Like

#9. AI auto color correction or auto white balance
It is a proposal that @Corros and @florisvaneck made, but I confess it is not clear for me how AI should improve the color balance process.
• By selecting a mood and apply the choice? For example: candle light, cozy living room (warm incandescent light), creepy basement neon, surgery block’s halogen lamp, sodium street lamp, bright winter day, rainy Sunday afternoon, misty morning, summer sunset, full moon night, arctic midnight day…
• By offering new tools to control or drift colors (temperature, tint, hue, saturation… or color wheel) for low, medium, and high lights, in order to make more easy and ergonomic carrying out new preset styles, in a creative spirit
• By selecting a set of pictures and giving to all of them the same balance color (and brightness too maybe)

What do you need?