3 ms·
With all the work on making these ML models controllable it seems like they now have a decent concept of things like "tall person" or "wears glasses". I assume
by wongarsu 8y ago
With all the work on making these ML models controllable it seems like they now have a decent concept of things like "tall person" or "wears glasses". I assume it would be entirely possible to replace the last few layers and generate 3d models instead of images. Nobody is doing that right now because nobody has training data for that, and generating such training data is hard (or expensive). But it seems like it should be in reach of someone with a profitable use case.
- ben_w 8y agoI expect that data to grow quickly over the next decade, assuming things like the 3D selfie camera on iPhones end up on the main camera too. (I expect they will end up on all the cameras, just for the value they’d add to AR).