NVIDIA has introduced an amazing "state-of-the-art deep learning" method that can reconstruct an incomplete image. Last week the company released a video showcasing examples of the new modern technology tool.
The main competitor for this new tool is Photoshop's Content-Aware fill. It uses information from surrounding pixels for photo restoration. However, this tool fails badly when it encounters a more complex image. NVIDIA took a different approach and put its new tool ahead of the competitors by employing artificial intelligence technology that understands what the subject should look like. However, it remains unknown when this photo manipulation tool will be available for the consumers.
Even if this new photo editing method isn't excellent at making esthetic reconstructions, it is definitely very good at recognizing what each part of the image represents. All the imperfections aside, we can't wait to test it out.
Check out the interesting photos below and tell us what you think in the comments.
More info: news.developer.nvidia.com
This post may include affiliate links.
All people who downvited for this picture clearly did not understand how impressive work the AI actually did.
Other than the eye, most of the reconstruction is very well done. And the AI is like ANY OTHER COMPUTER PROGRAM, it's only as good as the information loaded into it. Hopefully it's able to "learn" by comparison of its finished product and the original photo. I'm not that computer savvy, but I can see that this AI programming is pretty fsking amazing!
Load More Replies...Didn't do so good on this one but this really isn't hard to do if you have photoshop. You use a clone brush and click on the orginal photo and then on the bad photo you just click the exact same place as you did in the original and it just draws it in. Not like he's recreating it pixel by pixel. It's very easy to do and on this one, he really messed up!
We can tell that Lori didn't read the description. I suggest you go back and do that, kiddo. It will help you to understand the context of these pictures.
Load More Replies...I don't think it's a house... But i don't understand why someone downvoted this...
Load More Replies...I really don't get this. It takes a good image then makes it bad. What's the point of that?
the first one is the original.. the second is what is fed to the AI.. the third is the result AI comes up with.. original is shown for comparison sake..
Load More Replies...Actually whatever tool this person is using isn't very good or the person him/herself isn't very good. I can do 100% better. Here is a sample of my work. Samplesofm...4b87a3.jpg
We can tell that Lori didn't read the description. I suggest you go back and do that, kiddo. It will help you to understand the context of these pictures.
Load More Replies...no...I think they took the eye out, and gave him a glass eye. lol
Load More Replies...Damn, it actually made her prettier XD Looks better without those unhuman eyelashes
People complaining you can see the flaws in the reconstructed pictures are missing the point entirely. It's EXTREMELY hard for a software to get a picture with blank spots like that and try to fix those missing spots with something that looks like it fits. I think that for a better effect they shouldn't have shown the originals. You have to undestand that the software isn't using the original picture as a guide, it's literally just working with the messed up picture, and coming up with a fix based on that damaged picture alone. And it's a freaking software, not a thinking human being who can look at a picture and understand what it's supposed to look like (specially when you have the orignal for comparison). The results are truly outstanding. But maybe that's something that you can only truly appreciate if you work in the area and know how extremely hard this is to accomplish...
This really is incredible. Yes, the reconstructions aren't perfect, but they're pretty damned good. Better than most people could do if not allowed to see the whole picture. Also, think of it this way; this technology can only improve. One day the results will be even better. I think the main problem right now is that some of the reconstructed parts are a bit... sharp. Jagged.
Load More Replies...Legitimately the only reason anyone would have invented this.
Load More Replies...Everyone's all like "Nah, it looks way worse than the original image!", but they're missing the main point. It's SO CLOSE to the original image. So close that the faces look creepy, edging towards the uncanny valley. But the fact that's it's this close to the original is really awesome! The program doesn't edit the original image; it takes the whited-out version of the original image and fixes the broken picture. I think it's really neat!
Lol for those that dont get it, imagine the first pic was not there. You have to reconstruct and complete the masked picture. That's what's the final pic is. The first pic is there as a reference for how good a job it did with the final pic. It would be neigh impossible to reconstruct something that erased or messed up completely.
All the people asking what's the point of this... What if you have an amazing photo of your cat being adorable, but there's some dirty underwear in the background? How about a photo of your family at a picnic, but your ex is there... white them out and run it through this, and in seconds you have a replacement photo without them. The technology isn't perfect yet, but it's REALLY GOOD and will only get better.
Lol people are TOTALLY missing the point of this whole article. You didn't even take time to read what it was about! OBVIOUSLY IT ISN"T SAYING THEY ARE IMPROVING THE ORIGINAL PICTURE!!!! The title is very odd and misleading but you can read what they are doing in the description.
People complaining you can see the flaws in the reconstructed pictures are missing the point entirely. It's EXTREMELY hard for a software to get a picture with blank spots like that and try to fix those missing spots with something that looks like it fits. I think that for a better effect they shouldn't have shown the originals. You have to undestand that the software isn't using the original picture as a guide, it's literally just working with the messed up picture, and coming up with a fix based on that damaged picture alone. And it's a freaking software, not a thinking human being who can look at a picture and understand what it's supposed to look like (specially when you have the orignal for comparison). The results are truly outstanding. But maybe that's something that you can only truly appreciate if you work in the area and know how extremely hard this is to accomplish...
This really is incredible. Yes, the reconstructions aren't perfect, but they're pretty damned good. Better than most people could do if not allowed to see the whole picture. Also, think of it this way; this technology can only improve. One day the results will be even better. I think the main problem right now is that some of the reconstructed parts are a bit... sharp. Jagged.
Load More Replies...Legitimately the only reason anyone would have invented this.
Load More Replies...Everyone's all like "Nah, it looks way worse than the original image!", but they're missing the main point. It's SO CLOSE to the original image. So close that the faces look creepy, edging towards the uncanny valley. But the fact that's it's this close to the original is really awesome! The program doesn't edit the original image; it takes the whited-out version of the original image and fixes the broken picture. I think it's really neat!
Lol for those that dont get it, imagine the first pic was not there. You have to reconstruct and complete the masked picture. That's what's the final pic is. The first pic is there as a reference for how good a job it did with the final pic. It would be neigh impossible to reconstruct something that erased or messed up completely.
All the people asking what's the point of this... What if you have an amazing photo of your cat being adorable, but there's some dirty underwear in the background? How about a photo of your family at a picnic, but your ex is there... white them out and run it through this, and in seconds you have a replacement photo without them. The technology isn't perfect yet, but it's REALLY GOOD and will only get better.
Lol people are TOTALLY missing the point of this whole article. You didn't even take time to read what it was about! OBVIOUSLY IT ISN"T SAYING THEY ARE IMPROVING THE ORIGINAL PICTURE!!!! The title is very odd and misleading but you can read what they are doing in the description.
