Hello all,
Just a quick update on what’s going on with LAS.
I’ve added a few neat things in my development build:
- RAW support (using rawpy)
- Basic denoising – useful when using RAW images
- Initial support for zoom lenses, i.e. lenses with multiple focal lengths
Adding these features was fairly straightforward, but when I looked at the scores coming from all the lenses in my dataset, I starting to see a strong correlation between focal length and score, and I’ve come to the conclusion that the Laplacian analysis is awarding higher scores to images with more edges, i.e. wide-angle lenses.
I need to take a few days to think about how to work around this issue. It’s still useful to compare lenses at the same or similar focal lengths, but it would be nice if the software were generalized enough to just toss in sample images of any sufficiently contrasty test pattern.
Don’t get me wrong, there are several ways to work around this issue, but any time your start normalizing scores, it becomes very important to keep your intention and end figures in mind, statistically speaking.
Thanks for your interest in LAS!
ER
Be First to Comment