Differentiating physical and non-physical events

A non-transitory computer-readable medium comprising computer program instructions that when executed by a computer processor of a discrimination system causes the processor to perform steps comprising:

• receiving a set of starting content frames depicting a plurality of objects at a starting time point and behavior of the plurality of objects at the starting time point;
• applying a first model to generate a set of plausible ending content frames for each starting content frame, the first model trained to generate a subsequent content frame based on a starting content frame;
• applying a second model to generate a set of implausible ending content frames for each starting content frame, the second model trained to generate a subsequent content frame without using a starting content frame;
• and
• training a discriminator model using the set of starting content frames, the set of plausible ending content frames, and the set of implausible ending content, to predict a preferred ending frame conforming to plausible physical events from a first and second ending frame given a starting content frame, wherein the set of plausible ending content frames are a positive training set for plausible ending frames and the set of implausible ending frames are a negative training set for the set of plausible ending content frames.

A system comprising:

• a processor;
• and
• a non-transitory computer-readable medium comprising computer program instructions that when executed by the processor causes the processor to perform steps comprising:
• receiving a set of starting content frames depicting a plurality of objects at a starting time point and behavior of the plurality of objects at the starting time point;
• applying a first model to generate a set of plausible ending content frames for each starting content frame, the first model trained to generate a subsequent content frame based on a starting content frame;
• applying a second model to generate a set of implausible ending content frames for each starting content frame, the second model trained to generate a subsequent content frame without using a starting content frame;
• and
• training a discriminator model using the set of starting content frames, the set of plausible ending content frames, and the set of implausible ending content, to predict a preferred ending frame conforming to plausible physical events from a first and second ending frame given a starting content frame, wherein the set of plausible ending content frames are a positive training set for plausible ending frames and the set of implausible ending frames are a negative training set for the set of plausible ending content frames.