Images are just data. That’s why, using machine learning, we can predict what an image represents by pushing its pixels into a sufficiently-trained neural network. In much the same way, given one frame from a video, we can predict what the next frame is likely to be, again using a sufficiently-trained neural network.
Damien Henry trained such a neural network using video shot from a moving train. He then generated the video below, starting with a single real frame, and then projecting each subsequent frame from the last using the neural network. As the video goes along, the neural network is replaced by a more advanced one every 20s, so the images become clearer over time. But to repeat: none of the video was filmed; after the first frame, all of the images are generated by the model. (via Gizmodo)
That’s all for this week. We’ll be back with more for the blog on Monday. In the meantime, enjoy your weekend!