Subsequent Frame Prophecy

Abstract

Abstract: Future frame prediction project aims at predicting future frames of a video given previous frames. If ‘n’ frames are given into the model (n being 19 in our case) our model predicts the n+1th frame in the sequence (i.e., 20th frame). The model used for prediction is a deep learning model - GAN (Generative Adversarial Model). The Generative adversarial model has 2 components: the generator which generates the 20th frame and the adversary (Critic) which compares the outputs of the generator with real outputs. The purpose of this comparison is to train both the generator and the critic in a cyclic fashion to the point where the generator can create almost real-looking outputs. Complex systems like aiming machines,self-driving cars, etc require a good level of correct future prediction to make their outputs correct. We aim at creating a common prediction model which can be generalized for any task and can be used in any such machine.