I spent a lot of time testing A.I. Gigapixel with challenging lower-res digital images from the last decade and a half, but then I realized that scans of even older film-based images might see some benefit.
Hi,
Some interesting results on your blog. Also, the film scan originals showed significant improvement. Thanks for that.
It is important to note that there are different use-cases for AIG.
One is for upscaling small or cropped originals, in order to make decent size display output (on high-resolution displays), or another case is for smallish printed output from such small source images with higher quality. My first digital camera produced almost 4 MP images, and they were a challenge to print at a reasonable size larger than a postcard, while now I can print them as pretty good quality 8x10 inches.
In my professional photographic career, I've had to use reproductions of passport pictures as a basis for a somewhat decently sized portrait of deceased persons. AIG would have been a great help there. Also, smallish snaps in old photo albums (negatives are usually lost, and prints were expensive therefore kept small) can get revived in digitally printed photobooks at larger sizes, or small crops can be enlarged.
Another use-case is for producing large printed output that can be inspected from a short distance, from already large source images (but insufficient for the maximum native printer resolution).
It's also important to remember that AIG is not very well suited for Forensic types of image enlargement (although it could be interesting to try), because inventing non-existing data is a no-no there. While it improves the 'look' of an image, it's not necessarily accurate (if one compares with the actual source for the image).
I assume that facial recognition may improve with future updates of AIG, but it wouldn't hold up in court. But then, the technology also doesn't claim to be fit for that purpose, it's better suited for other uses.
Cheers,
Bart