Self-learning computer predicts movements in photos
Researchers from the technical university MIT have a self-learning system has been developed that movement of objects in photos can predict. Therefore, it is possible in theory to get a photo to turn it into a short video.
To the neural network, researchers have more than two million videos downloaded from Flickr. Which were classified in four different environments on the images to see are: golf courses, beaches, train stations and hospitals. The images are then stabilized.
The system has the videos analyzed, and then the researchers photos to the database added. The computer predicted the movement on the basis of what in earlier videos.
The videos with the system are made for a few seconds, in a small size and in most cases out of focus. The technique can contribute in the future to voorspellingstechnieken for computers.
The computer predicted, for example, that trains at the photos of a drive through the picture would drive. Also knew the system to predict that someone on a golf course walk, and a dog swimming through the water.
For some videos, knew the system is not exactly which objects would move. Also see the running away from people is not realistic. According to the researchers, it looks like all might not be smooth, but the predicted movements are in most cases correctly.