Loads of devices can preserve moments on camera, but what if you could capture situations that were about to happen? It’s not as far-fetched as you might think. MIT CSAIL researchers have crafted a deep learning algorithm that can create videos showing what it expects to happen in the future. After extensive training (2 million videos), the AI system generates footage by pitting two neural networks against each other. One creates the scene by determining which objects are moving in still frames. The other, meanwhile, serves as a quality check — it determines whether videos are real or simulated, and the artificial video is a success when the checker AI is fooled into thinking the footage is genuine.
The technology definitely has its limits. It can’t produce videos that go further than 1.5 seconds into the future, and the results aren’t mind-blowingly realistic: it isn’t aware that objects are still there when they move, and tends to exaggerate their sizes. However, it’s good enough to predict relatively complicated scenes like waves on the beach, or people walking on grass.
If CSAIL can extend predictions and make them more realistic, though, it could have a far-reaching impact. Self-driving cars could predict where vehicles and pedestrians are going, while security cameras could spot mismatches in footage based on what they expect to see. It could also be used for relatively everyday tasks like adding animation to still images or compressing videos (since you wouldn’t need every frame). And regardless of circumstances, predicting the future can help AI understand what’s going on right now — this could help with just about any instance where computer vision is important.