Frame Prediction using cGAN

Interpolate and extrapolate video frames using cGAN

Authors: Hongyu Zhu, Xiaolong Wang and Siddha Ganju.

Network architecture

The GAN

### The generator

## Models and inputs

Models	Generator (G)	Discriminator (D)
Baseline	1st frame + noise	future frame
FlowGAN	1st frame + flow (128x128) + noise	flow (128x128) + future frame
FlowGAN-comp	1st frame + flow (128x128) + noise	1st frame + flow (128x128) + future frame
FlowGAN-sim	1st frame + flow (1x40) + noise	flow (1x40) + future frame

Dataset and experiments

Frame prediction -- UCF101

The 6^th frames

Every 5 frames recurrently

Action Recognition -- HMDB51

Static Image Editing -- MS COCO

GIFs with series of random flows