nesdev.org

I saw this at Walmart the other day and it was too ridiculous to pass on taking a picture. If this is what I think it's trying to say...

It does look like bullshit...

Oh, absolutely. I mean, there's no God damned way it could take that undiscernible jumble of (non square?) pixels and turn it into a photo-realistic image with only a simple algorithm. It's so blatantly false that I was actually questioning if I somehow interpreted what they were trying to say incorrectly.

Espozo wrote:that undiscernible jumble of (non square?) pixels

The "before" image is obviously being displayed by a Commodore 64.

Assuming the image on the right is 4K, that's probably close to about the size of C64 pixels.

The same company also has bridges to sell.

No! It's not false advertising at all!
Before upscaling:4K image as shown on classic 320x180 resolution.
After upscaling: 4K image upscaled to 8K and shown on 4K screen.

There has been a lot of research effort put into image super-resolution techniques lately, so this is not as implausible as it might first seem.

But yeah that image does look like an exaggerated artist rendition of what's happening.

so this is not as implausible as it might first seem.

It's entirely impossible here. Unless you're going to say that it used an object detection algorithm and guestimated what the original image looked like, but disregarding how absurd that is, I couldn't even tell what the heck the "before" image was.

koitsu wrote:The same company also has bridges to sell.

But apparently not a reputation.

Espozo wrote:

so this is not as implausible as it might first seem.

It's entirely impossible here. Unless you're going to say that it used an object detection algorithm and guestimated what the original image looked like, but disregarding how absurd that is, I couldn't even tell what the heck the "before" image was.

"Guesstimation" is essentially what those super-resolution algorithms do.

Here's a quick example I found:
https://www.slrlounge.com/zoom-and-enha ... a-reality/

You can see that the NN (neural network) is able to reproduce the most important features from the really low resolution input. Of course it doesn't match the ground truth perfectly -- there's simply not enough information in the input. Faces similar to the ones in the NN results were probably present in the training set.

(BTW, the article is wrong in claiming this makes the TV trope a reality -- these algorithms wouldn't allow you to, say, extract a license plate number from a low resolution image, they would simply dream up some random license plate number if the information in the source image is not sufficient to reproduce it.)

That said, temporal stability (= results should appear consistent from frame to frame in videos) is a huge problem with these algorithms, so I'm very skeptical how their "UHD upscaler" actually performs in practice.

TV requires that stuff to be realtime, and I don't think any of those guessing algos can run realtime on any modest hardware...

TmEE wrote:TV requires that stuff to be realtime, and I don't think any of those guessing algos can run realtime on any modest hardware...

With FPGAs or ASIC it's possible to run quite advanced algorithms in real time or almost real time (assuming you can accept some constant-timed lattency)

These algos still require access to a whole lot of "reference material" to work, moving it in and out of network connection isn't practical nor storing any huge amount inside the device itself... I will remain very skeptical for the time being.

I'd assume that if any algorithm could do anything close to that, in real time and on arbitrary images, it would have to rely on multiple frames of information. A jumble of pixels doesn't tell you much, but if you can watch it change over several frames you might be able to figure something out, and your time stability will be better too. You'd need to be able to detect pans and scene changes somehow, and the results would still be very inconsistent...

It'd look like shit. Through all the effort to do that sort of intensive upscaling, you might as well have just gotten a higher resolution video. It's solving a problem that doesn't even have to exist.