New imaging system creates pictures by measuring time
phys.org
phys.org
I'm suspicious that their outputs may be solely the result of over-fitting from neural networks.
And that given a significantly structurally different scene, the output would be nonsense.
The article does not seem to show a broad variety of scenes, and physicists are frequently not creative photographers. I would not be completely surprised if the training images were rather predictable poses in a physics lab. And do not see how there can be enough information from the sensor to produce an image otherwise.