Long-term video motion temporal super resolution

This is an ongoing research, which is just half-done.

1. Introduction:

This research considered two tasks:

a.Long-term video temporal super resolution

Real World Ground Truth		Video with low temporal resolution		Recovered Video

b.Long-term video interpolation

Original Video		Input Video with missing frames		Recovered Video

2. Problem Formulation:

a.Long-term video temporal super resolution

b.Long-term video interpolation

3. Basic Model:

a.Long-term video temporal super resolution

b.Long-term video interpolation

3. Basic Model Result:

a.Long-term video temporal super resolution

temporal super resolution rate as 4(green frames is given, red frames are missing)

Methods	Running	boxing	clapping	waving
Ground Truth
Ours
Niklaus et al.

b.Long-term video interpolation

Given first 5 frames and last 5 frames, we can interpolate 10 frames in the middle
(green frames is given, red frames are missing)