Jobs done - Step by Step
Python 3.9.0
Imports ...
Parsing arguments ...
Using device: cuda:0
Loading model ...
Upscaling x2 ...
Saving output ...
Done
===========================================================================
Finished in 29m48s
===========================================================================
And thats the time, my computer comes in to play.
He gave me the Result Image from his SwinIR Upscale because his Computer stucked and ran out of Memory.
Now my Part:
- Try to Upscale the whole Image with SwinIR x4 but i ran out of memory aswell.
- Tried to Upscale one more time with ruDALL-E Real-ESRGAN x2 and works fine after 2 Minutes and 40 seconds.
The Result of the Upscale was Amazing at this tims - But we want more! Damn more Pixels!
We tried the last upscaled version to upscale it 2 more times. This took us round about a half hour and a lot of hoping the GPU will not crash and the computer stays cool and not burn.
After the half hour upscale work, all was fine and i put the results together in photoshop. Picture on top of this post. As u can see, AI currently doing a very very great Job in Upscaling and this by a commercial GPU.
Final Image - Scaled down to 10% of Original!!!
Sadly, Photoshop wont let me save the Big Image properly
Some Numbers!
- From 5,7MP to 365 Mega Pixels in under 1 Hour!
- The biggest Image contains 64 times the original image size.
- Biggest Image File Size: 9,40 MegaByte
It was realy fun to test this things out. Sitting in the Voice Chat and having a nice evening, exploring AI things try new stuff. Big Shout out to my Friends 
U can find the Original Image on Unsplash here to check out the earlier quality of the image:
Sailor Image by Brett Jordan on Unsplash
Whats your thoughts on this kind of mind blowing work in AI? Did u ever tried an upscaling in this sizes?
Lets discuss in the comments below.