Pictures From Home

Same spot, one photo has AI assistance in camera phone.


8 Likes

Which one? :smile:
And which corresponds more with what you saw perceived?

4 Likes

Guess :stuck_out_tongue:

which corresponds more with what you ~~saw~~ perceived

Hm, good question. I don’t think the photo captures the light of the moment (my mobile phone is kinda trash). We’ve also had dust these last few days, so it was somehow yellower and brighter from both photos.

2 Likes

OK, so I guess (what I had guessed already before your last entry here):

The redder one is the photo without AI “assistance”, and the bluer one with AI?

I am assuming that the AI tries to “improve” the photo quality, but I have no idea what this “improvement” is supposed to be based on.

1 Like

The top one is the AI.

1 Like

Oh that’s ridiculous but I didn’t even realize my phone camera has an AI function. So I took a picture of the view from my window. Nothing interesting really, just for comparison.

With AI

Without AI

I don’t see any difference.
I also took two selfies, with and without AI, hoping that AI would improve my physical appearance; it didn’t.

5 Likes

It probably doesn’t. The easiest way to implement AI for that kind of App is to have a list of preset of light (with both meanings of the word) filters depending on the photo (face, landscape etc) and apply them to each photo you use the “AI function” to.

A small source.

" Welcome to the world of “AI Washing” — the latest corporate con where companies create false hype about their AI prowess to attract investors and drive up their valuations. It’s a phenomenon that’s sweeping the business world, and it’s leaving a trail of misled investors in its wake."

2 Likes

“AI” is being used as a very vague marketing term in these contexts. When it comes to photo editing/manipulation, it can mean anything from a very broad range of things. There are also considerations that need to be made about maintaining a sense of the objective truth.

On one extreme, it might be applied to some very basic automation, such as selecting various camera settings or applying some simple post-processing (like adjusting color balance) using a model that makes such decisions based on some perception of the content of the photo. For example, it seems that color balance is the most noticeable adjustment made in the examples shared by @Gia and @jlt.

There are also AI-powered content manipulation tools like “magic eraser” or “content-aware fill”, however, these tools generally need to be specifically applied by the user.

Without explicit user control, we may have things on the other extreme…

It’s really hard to capture a high-quality photo of the moon with the low-quality optics of a phone camera. However, it’s quite easy to simply detect if the camera seems to be pointed at the moon, and then fill in the details…

5 Likes

Every day that The Louvre is open, tens of thousands of tourists visit the Mona Lisa and many attempt to take a photo of the masterpiece with a poor quality phone camera.

I think the ideal AI-assisted algorithm to help tourists take a great photo of the Mona Lisa would be the following:

  1. Use a pre-trained convolutional neural network to detect whether a photo is an attempt at capturing an image of the Mona Lisa.
  2. If this detector says yes with high enough confidence, discard the photo just taken and replace it with the following file downloaded from Wikipedia: File:Mona Lisa, by Leonardo da Vinci, from C2RMF retouched.jpg - Wikipedia
7 Likes

Today, before the rain, ~10 km South-West-ish of Lüneburg

Looking North by North-West

Panorama, from ~West-ish to –East-ish

Zooming in on my car

~6 km North-East-ish of Lüneburg, still before the rain, looking South:

A Stork nest, must have been in use for quite a few years already, considering its height. Built on a siren tower.

Back near my place, ~35 km South-West of Lüneburg, ~60 km South of Hamburg
Looking ~North-West, after the rain:

11 Likes