AI image enhancement

Gigapixel AI from Topaz Labs is a dedicated tool, performing image enlargement on one or more images (it does batch processing). Its promotional video claims that it is better than ordinary image editors such as Photoshop in enlarging smaller images. It makes use of  neural networks, a kind of artificial intelligence algorithm that learns based on examples, to recreate details.

The video clearly shows what it can do, and I think it would work great for most pictures. But I wanted to test it on some pictures that I’d consider as special cases.

I downloaded it, and immediately tested it on some smaller images I had on hand.

Screenshot from videos

The following is a screenshot from a video clip, which I took using my Pentax KS-2 with a 50mm 1.8 prime lens, and I filmed using the widest aperture and slowest exposure time possible (it went up to 30 for video). I think it looks pretty clear as it is, considering that it is taken in video mode. However it might benefit from some reconstruction if, for example, I wanted to print it on large media such as a tarp.

vlcsnap-2019-04-01-00h22m08s296

I cropped out one tiny portion to test on Gigapixel AI.

vlcsnap-2019-04-01-00h22m08s296_crop.png
470×264 pixels

I enlarged it with Photoshop.

vlcsnap-2019-04-01-00h22m08s296_crop photoshop
from 470×264 to 1881×1057 pixels in Photoshop, nothing else was done. Just enlargement.

Then with Gigapixel AI.

vlcsnap-2019-04-01-00h22m08s296_crop-edit
from 470×264 to 1880×1056 pixels in AI Gigapixel

I think the Gigapixel AI edit is visibly clearer, but not by much.

Old Photo of an old photo.

I downloaded a small image, an old picture of one of our heroes, from this website which includes his story.

here is the picture in the size I had downloaded it in, and I assume they were trying to save on space when it was uploaded:

macariosakay
440×329 pixels

Running it through AI Gigapixel produced this:

macariosakay-edit
2640×1974 pixels

The first thing I noticed was that the faces look wonky. I think this has something to do with the way Neural Networks work: it depends on the training data. If AI Gigapixel had been trained to work with a certain type of image, it might not know how to handle other kinds. Or maybe it has something to do with the dynamic range of the image? Whatever the cause may be, it looks like the software had a hard time figuring out this sort of picture.

To be fair, the clothes look much, much better than they do in the original photo.

Photo of a painting

I also tested it on a painting I made using the above image as a reference.

I took a photo of a painting a made, before turning it over to the person I made it for, but the camera I had at the time wasn’t a very good one. There is no way I could ask to see the painting so I could take a picture of it, since the place it was stored in has burned down. I cropped Macario Sakay’s face:

IMG_3248_v1
469×264 pixels

I enlarged it at 4x just like all the others.

IMG_3248_v1-edit (1)
1876×1056 pixels

This one turned out pretty nice, the pixellation is gone and I can even see some of the paper texture

The verdict

After the trial period, will I buy it? I might not be able to afford it at the already discounted $79 price tag, and to be honest I think it needs a bit more work, but if it were just a bit cheaper…like, priced-for-someone-living-in-a-third-world-country cheaper?

I really do like it, however. I would definitely recommend it whoever might need this sort of task. I’d warn them of the price but depending on what they need, it can be worth it.

 

Boredom

Last Sunday, my kid complained about how boring his day was and asked to be allowed to play on the X-box. I retorted, in the vernacular “don’t you have anything else to do?”

His answer was no, so I gave him something to do. Insert smug smile here.

IMGP2696_v1.JPG
Sunday last week

We had some sunflower seeds he himself had bought but had just planned to throw at some unoccupied space in the plant box at the back of our house. There they would grow like weeds in the dust-like, barren soil.

That never happened because I, the overbearing mom, had something else in mind, and had collected some drink-boxes specifically for this purpose (since we buy them anyway, as he really likes this drink). After the planting activity, which lasted for about an hour or so, maybe more…he then proceeded to resume complaining about how boring his day was.

IMGP2790_v1.JPG
Thursday
IMGP2806_v1.JPG
Friday

He would kind of check on his baby sunflowers once in a while, spray them with a bit of water. It’s saturday now, and later today I’m planning on suggesting that he stick some popsicle sticks into the soil, so the little plants can have something to lean on.

Boring is ok.

He definitely has a problem with boredom, he absolutely cannot stand it. I had a similar problem when I was younger; it might be hereditary.

When he asked to play on the X-box, I started with, “we had an agreement that we both only get to play during summer break, right?” Last summer I kind of got addicted to one of the Splinter Cell game modes, the one where you co-op with another player against hordes of enemy AI’s. I can control myself now, but years ago, I would sometimes skip school and head back home, sneak into the house and spend about half my day on the playstation. I generally managed to get by with ok grades, except for chemistry and some of the more advanced math courses. My son knows this, and I wouldn’t blame him if he called me out on my hypocrisy.

 

I’m a sort of prototype to today’s digital natives: I got to play Dig-dug on a computer at age 3 in the mid-80s, not exactly like today’s toddlers playing games on iPads, but close enough right? And when these things, computers, started invading us for real in the mid 90s, multimedia was at its infancy. I loved the games, I loved being able to mess around with music instead of just listening to it, but Encarta–the encyclopedia in a CD–could never compare to the school library’s Encyclopedia Brittannica. Today, this is certainly not the case anymore. Practically all of my shool work, and my son’s, are done digitally. I’ve seen the internet’s transition from being something you use very sparingly, something teachers might even prohibit you from using, to something you use pretty much all the time. Before, content was scarce and unreliable, and now it’s abundant and as long as you know what you’re doing, generally useful. I’m somewhere in between embracing technology–which I am undeniably fascinated with–and approaching it with caution.

My son and I have very different reading habits. He is very good at skimming, finding things quickly, mapping it out in his head, while I, his mom, often need more time in gathering, consolidating and using the information I need. I am not by nature a skimmer, having started out in life as a deep-reader. It took some effort on my part to switch to this way of life. I even print things out sometimes so I could read them better (which my son would roll his eyes at). My son’s projects seem to involve more interaction with classmates, and even teachers, than do mine. I’ve seen him sit in front of his laptop the whole day and produce nothing but a few bits of writing and a whole lot of visually appealing but very much empty layouts. It’s nothing new, I guess it’s the modern version of kids calling up their classmates while doodling on their notebooks.

We both write at our own pace, but usually finish writing things at about the same rate–if I finish, at all. That’s something he does better than me: he is able to complete more work as long.

While for him, boredom means not having anything interesting to do, for me boredom is doing something and getting tired of doing it. Sustaining the willpower to complete a task quickly–perhaps some artwork I’m expected to submit to school or gift to someone (paid jobs are, on the other hand, a different thing. Wink, wink) has been one of the biggest challenges for me.

I asked my uncle for some plywood that I can use to stretch watercolor paper, and he gave me enough for 4 of these stretchers. This means I can do 4 “parallel” paintings–when I get bored with one, I switch to another.

IMGP0505_v1.JPG
watercolor on master board, around 5.5 x 8 inches.
IMGP0506_v1
I got bored so I didn’t finish his face, but I kind of like this painting as it is.
IMG_3249.JPG
I gave this to one of my favorite professors, whom I like to call Sir Fabros the Fabulous. It took me a year to complete it, and it never gotr into the exhibit. I’m assuming its charred remains have by now been cleaned out of the building it was in, which burned down a few years ago.

I’m planning on doing 4 watercolor paintings within 2 weeks, I don’t know if that’s possible for me.

HDR Images on my Pentax KS-2

I’ve heard that you can do some neat tricks with exposure bracketing, and so I tested it on my Pentax KS-2.

IMGP2325_v1
original image, using 50mm prime lens at ISO100, F7.1, 1/100 exposure

 

IMGP2326_v1
HDR at +/- 1 steps
IMGP2324_v1
HDR at +/- 2 steps

 

For me, the +/-1 image works best, while the +/-2 seems a little bit too much.

I asked someone to take their pick and they chose the +/-2 as the best one because according to him it’s “stylish, it looks like a painting.”

He said pretty much the same thing about the following. He likes the high-contrast, seemingly photoshopped +/-2 HDR images.

IMGP2327_v1
No exposure bracketing. Using the 18-55mm kit lens at ISO100, F7.1/26.2 focal length, 1/100 exposure
IMGP2329_v1
HDR at +/- 1 steps
IMGP2328_v1
HDR at +/- 2 steps

I’d love to try this out on a landscape, or maybe I’ll just rearrange these potted plants and take more pictures. Meanwhile, my iced coffee, at +/- 3, +/- 2, +/- 1 and none:

IMGP2821_v1IMGP2822_v1IMGP2823_v1IMGP2824_v1

 

 

 

Design a site like this with WordPress.com
Get started