NVIDIA AI Inpainting Is Cool as Hell

FrgMstr

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,601
No matter what you think about NVIDIA and its anti-competitive GPP, its AI powered Inpainting is damn cool to see in action. It is not perfect, but what it is getting done in real time is amazing.
Taking away Borgnine's trademark crazy eyebrows is across the line however.


Check out the video.

Researchers from NVIDIA, led by Guilin Liu, introduced a state-of-the-art deep learning method that can edit images or reconstruct a corrupted image, one that has holes or is missing pixels. The method can also be used to edit images by removing content and filling in the resulting holes. Learn more about their research paper “Image Inpainting for Irregular Holes Using Partial Convolutions.”
 
That's cool. Not perfect for sure. I can think of this as a nice way to make it easier to get "the shot" you want. There was a view from a hill overlooking town, with a nice mesa in the background that I always thought would be a great picture, but there was no place you could take it without having a bunch of power lines disrupting the view. This could be an easy way to get rid of those.
 
That's all just crazy, like down right fucked up in all sorts of ways but still kinda awesome.
 
jardows said:
That's cool. Not perfect for sure. I can think of this as a nice way to make it easier to get "the shot" you want. There was a view from a hill overlooking town, with a nice mesa in the background that I always thought would be a great picture, but there was no place you could take it without having a bunch of power lines disrupting the view. This could be an easy way to get rid of those.
Click to expand...

I have photos that I took on my last vacation to Italy, they have random people in the background as taking the photos without them there was next to impossible. I would definitely try this out to remove the randoms and leave just me/family in the shots.
 
Interesting tech - and from that perspective I think it's very neat. However, like so many editting practices, I'm afraid this further blurs the lines between photographic art (as captured) and graphic arts (as generated).

I'm not a photographer in the least, so my opinion is probably ill-informed.
 
That's really cool. Great way to touch up screenshots from your games.
 
jardows said:
That's cool. Not perfect for sure. I can think of this as a nice way to make it easier to get "the shot" you want. There was a view from a hill overlooking town, with a nice mesa in the background that I always thought would be a great picture, but there was no place you could take it without having a bunch of power lines disrupting the view. This could be an easy way to get rid of those.
Click to expand...

Photoshop has had something like this since cs5 which was release in 2010.

It won't do things like adding in back eyes but it will remove power lines from the previous posters.

Here is video of a guy showing content aware fill from 2012.
 
I wonder how many (and what kinds) of training/reference images they had to use for the face inpainting?
Guess it's time to read the research paper...
 
This is with desktop graphics cards or graphics cards you need a mortgage for?
 
Looked pretty impressive to me. Other than the eye replacement not really matching the face I'd say the rest of it was excellent. I'd use this for sure. Lots of photo's with something (sock, cup, kids toy, etc) that distracts from the overall scene.
 
elavanis said:
Photoshop has had something like this since cs5 which was release in 2010.

It won't do things like adding in back eyes but it will remove power lines from the previous posters.

Here is video of a guy showing content aware fill from 2012.
Click to expand...

Did not know this. I just tried it out and its ok; works better with smaller stuff then what I was looking to do but it does work.
Thanks for the info.
 
I think Photoshop does it better with the newer versions and as already stated they have been doing this since 2012. However, NVIDIA's is good too and competition is always welcomed but I don't think it can match PS yet.
 
DangerousMan said:
I have photos that I took on my last vacation to Italy, they have random people in the background as taking the photos without them there was next to impossible. I would definitely try this out to remove the randoms and leave just me/family in the shots.
Click to expand...

Have you tried the photoshop content aware to remove them?
 
Kwaz said:
Have you tried the photoshop content aware to remove them?
Click to expand...

Yep, just now. Not the best outcome but it did remove some of what I wanted. There are still unwanted people in the pics but it does look better now.
 
That's cool. Pretty much a slightly more advanced Photoshop clone stamp / healing brush tool.
 
“To the best of our knowledge, we are the first to demonstrate the efficacy of deep learning image inpainting models on irregularly shaped holes,” the NVIDIA researchers mentioned.

Is it just me that finds that intriguing?

Mind you, the last thing I think of doing is inpainting models on their irregular shaped holes.
 
Pfft, fake.. FAKE

They already had everyones pictures when they logged into Facebook through Geforce Experience, they just put it back together, no AI needed
 
I'm waiting for when I don't need to ever take a photo, the AI will do it for me...and it better stay out of my bedroom. Heck the way things are going the AI will need one image of you then seamlessly put your imagine into any compromising situation it wants and make it look real to forensics...who needs killer bots?
 
  • Like
Reactions: _l_
like this
WhoMe said:
I'm waiting for when I don't need to ever take a photo, the AI will do it for me...and it better stay out of my bedroom. Heck the way things are going the AI will need one image of you then seamlessly put your imagine into any compromising situation it wants and make it look real to forensics...who needs killer bots?
Click to expand...

So, will AI eventually be used to make imbeciles smart so they can challenge the naturally non-imbecilic for their high tech job positions?
 
  • Like
Reactions: WhoMe
like this
_l_ said:
So, will AI eventually be used to make imbeciles smart so they can challenge the naturally non-imbecilic for their high tech job positions?
Click to expand...
What high tech job positions? And seriously while I think AI is going to change a lot of things we are starting to see it, the timeframe I'm not sure of...I remember back in 90's when a lot of people predicted computers were about to take away every computer programmers job...still waiting...but maybe it's in sight now?
 
elavanis said:
Photoshop has had something like this since cs5 which was release in 2010.

It won't do things like adding in back eyes but it will remove power lines from the previous posters.

Here is video of a guy showing content aware fill from 2012.
Click to expand...

I was just about to post this. Adobe came out with it quite awhile ago, the nvidia design seems to do it quicker which means they are leveraging the AI platform to create/fill the content. Though a lot of it was obvious it was altered it was quite interesting to see it add in eyes from a photo on its own, although it looked like the same set of eyes that were put on the supermodels face as the older actor's face( can't remember his name).
 
Local t.v. news had a story a week or two ago with Peele showing off some tech using himself with live rendering tech to create a fake Obama announcement. It looked a little odd but to someone not really used to digital quirks they might not catch all the signs. The anchor then went to a company in Cali that specializes in that stuff and experienced it with live rendering of himself as other people it was impressive and scary. Pics done thru your card seem pretty easy after seeing that.

In other NV news. OCD mentioned some other new feature involving voxel lighting techniques: https://overclock3d.net/news/softwa...soon_with_increased_performance_and_quality/1

Kind of neat but ironically they state that the compute hardware needs are pretty demanding and that's why they still haven't fully rolled it out. Even more ironic since if they hadn't effectively disabled hardware based PhysX many of us with decent older cards could probably easily benefit from this.
 
lostin3d said:
Local t.v. news had a story a week or two ago with Peele showing off some tech using himself with live rendering tech to create a fake Obama announcement. It looked a little odd but to someone not really used to digital quirks they might not catch all the signs. The anchor then went to a company in Cali that specializes in that stuff and experienced it with live rendering of himself as other people it was impressive and scary. Pics done thru your card seem pretty easy after seeing that.

In other NV news. OCD mentioned some other new feature involving voxel lighting techniques: https://overclock3d.net/news/softwa...soon_with_increased_performance_and_quality/1

Kind of neat but ironically they state that the compute hardware needs are pretty demanding and that's why they still haven't fully rolled it out. Even more ironic since if they hadn't effectively disabled hardware based PhysX many of us with decent older cards could probably easily benefit from this.
Click to expand...

OCD, lol
 
Finally something useful for AI to do. Now I can churn out memes 10 times as fast.
 
The real question people should be asking is 'what happens when you apply said tech to censored boobs?' do you get eyes instead of nipples???
 
exiled350 said:
The real question people should be asking is 'what happens when you apply said tech to censored boobs?' do you get eyes instead of nipples???
Click to expand...
Neither, it removes them and you get a pair of kneecaps in the upper chest.
 
Codename: Project Stalin.

He would of LOVED this. Best way to erase people from the party. LOL.
 
When good AI pic software goes bad:

pqacn09yoxcz.png
 
You must log in or register to reply here.
Back
Top