A Survey on Video Future Frame Prediction

PhD Qualifying Examination


Title: "A Survey on Video Future Frame Prediction"

by

Mr. Jingyuan LIU


Abstract:

Future frame prediction has been extensively studied in recent years, 
especially with the flourish of deep generative models. Future frame prediction 
is a challenging problem as to generating high-resolution future frames and 
handling uncertainty when there are multiple possible future frames. Also, in 
order to generate reasonable future frames, the underlying tendencies of frame 
evolution from the existing image or video need to be captured. Many deep 
generative models have been made suitability for this task, yet there is still 
a large potential of improvements.

In this survey, we start with an introduction to the concept and background of 
the future frame prediction problem. Then, according to the history knowledge 
available for prediction, we classify existing methods into two categories: 
predicting from a single image and predicting from an image sequence. For each 
category, we discuss methods with respect to pixel-based, patch-based and 
motion feature-based. The survey ends with a discussion about the current 
challenges of future frame prediction and future research proposals.


Date:			Wednesday, 20 November 2019

Time:                  	10:30am - 12:30pm

Venue:                  Room 5501
                         Lifts 25/26

Committee Members:	Prof. Chiew-Lan Tai (Supervisor)
 			Dr. Xiaojuan Ma (Chairperson)
 			Prof. Long Quan
 			Dr. Pedro Sander


**** ALL are Welcome ****