4 ms·
It's unclear to me why the wall would need to be "blank". Maybe simply unobstructed and flat or moderately consistent reflectance Lambertian? A dark black wall
by kurthr 5y ago
It's unclear to me why the wall would need to be "blank". Maybe simply unobstructed and flat or moderately consistent reflectance Lambertian? A dark black wall would be noticeably more difficult, while one painted with several bright colors could be corrected fairly easily. Even a wall with fixed fine black on white patterns (e.g. text) should be comparable, while a white sheet with fans blowing on it incredibly difficult.
- dahart 5y agoIt’s really a statement about what the researchers actually did for this paper, not a statement about what’s required or possible in general now or in the future, right? Because they qualify this is for blank walls both in the abstract and in the video, I’d first assume it’s possible they tried it and ran into substantial issues - even though it might logically seem easy. Maybe it will take a bit more math or engineering to handle non-blank walls, and it’s being saved for the next paper. Since non-blank walls will increase the dimensionality (and training time) of the problem, and likely come with reduced effectiveness in some cases, and since a result with blank walls is good enough for publication, maybe this is just a first step?
- hammock 5y ago>It's unclear to me why the wall would need to be "blank" The initial state of the wall (before people are introduced) should be known. With a solid color wall you can find shadows of people, objects, etc. But with a wall with objects on it, you need an image of it before the room was filled, in order to interpret the shadows. (From a machine standpoint. As humans we can process out the art or whatever is on the wall)