Hierarchical Video Generation from Orthogonal Information: Optical Flow and Texture (supplemental material)
This page presents animated gifs introduced in our paper. Although the original size of all videos are 64 $\times$ 64, we enlarge them in this paper for better visibility.
Experiments
TextureGAN: Video Generation Results from Ground Truth Optical Flow and $z_{tex}$
| Penn Action | SURREAL | ||||
|---|---|---|---|---|---|
| GT optical flow | Generated videos | GT videos | GT optical flow | Generated videos | GT videos |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
FTGAN: Video Generation Results from $z_{flow}$ and $z_{tex}$
| Penn Action | SURREAL | ||
|---|---|---|---|
| Generated optical flow | Generated videos | Generated optical flow | Generated videos |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Qualitative Comparison to Previous Study: Visualized Results
(1) Videos in Figure5.
| Penn Action | SURREAL |
|---|---|
| VGAN | |
 |
 |
| TextureGAN (ours) | |
 |
 |
| FTGAN (ours) | |
 |
 |
(2) Additional videos randomly sampled from generated results.
| Penn Action | SURREAL |
|---|---|
| VGAN | |
 |
 |
| TextureGAN (ours) | |
 |
 |
| FTGAN (ours) | |
 |
 |