Adobe’s Project Indigo camera app seems pretty interesting. It is definitely a bunch of experimental features in a trench coat so don’t bank on any of the computational photography features working exactly as advertised. The one I was most excited about was reflection removal. That’s a feature that they added to Adobe Camera Raw but is kind of finicky. It’s a 410MB feature to download, and it can only work on DNGs you’ve taken with Project Indigo. It doesn’t always detect reflections correctly.
When you run it on an image you take you get three images. The original, the “reflection”, and the modified result. You’ll note that the surface reflections have not been removed, but what’s visible through the water has.
This one is a shop window, where any person could identify the reflection if you drew a grid over it for a CAPTCHA. Reflection removal does not correctly identify it and instead perceives an ambient haze to remove.
Woof.
The first time it worked. I guess it has to be an extreme, glancing angle for it to figure out what the reflection is. It extracted it without damaging the pastries. The strawberry pistachio danish at Nemesis is delicious, btw.
@joesteel Do we care that it looks kind of surreal after the reflection removal? Without any indication of glass, the image seems more AI generated to me?
@DavidAnson would you notice or care if there weren’t comparison photos? These are reflection removal tests so no one is supposed to post comparison images. There are practical ways to remove and reduce reflections with circular polarizers, but those are bulky.
@joesteel I’m not sure… Could you post it again without the comparison photos? :) That said, the uncanny valley is uncanny even when we can’t articulate exactly why it is.