3 ms·
I wonder if, with the right training data, this kind of thing could make use of detailed 3D/structural knowledge when doing the transforms. There is lots of in
by grandalf 11y ago
I wonder if, with the right training data, this kind of thing could make use of detailed 3D/structural knowledge when doing the transforms.
There is lots of information about the 3D world present in the light (at least for the bedrooms).
This could be very useful for spatial navigation or computer vision, I'd think.
- nl 11y agoYes. I know a guy doing his PhD on this.
- grandalf 11y agoInteresting! Any links or terms to google would be appreciated.
- sedachv 11y agohttps://en.wikipedia.org/wiki/Geometric_feature_learning https://en.wikipedia.org/wiki/Geometric_feature_learning
- nl 11y agoI believe it is part of this project: https://cs.adelaide.edu.au/~hengel/?/research/project-lego/ https://cs.adelaide.edu.au/~hengel/?/research/project-lego/ Paper: http://arxiv.org/pdf/1511.02570.pdf http://arxiv.org/pdf/1511.02570.pdf He's in that research group, anyway.