Diverse Generation from a Single Video Made Possible

Supplementary Material

This page contains many videos, please give it a minute to load
If a video does not appear - please Refresh (F5)
Back to Main Page
PatchMatch vs. Exhaustive HP-VAE-GAN SinGAN-GIF

Comparing PatchMatch vs. Exhaustive Search

In the Supplementary PDF we show that there's no loss in quality of the generated samples by using PatchMatch instead of the exhaustive search used in GPNN.
All generated samples used for this evaluation can be found HERE

All Comparisons from Table 1 - Overview

Comparing to HP-VAE-GAN [Gur et al.]
  • Ours (VGPNN): 18 seconds runtime
  • HP-VAE-GAN: ~7.5 days of training per input video
  • Comparison on their 10 supplied 13-frame videos with resolution 144x256
  • Please also see comparison to Full-HD results in the main page

Comparing to SinGAN-GIF [Arora & Lee]
  • Ours (VGPNN): 10 seconds runtime
  • SinGAN-GIF: Training time per input video - Unpublished
  • Comparison on their 5 supplied videos (8-frame videos). No code is available
  • Note how in many outputs of SinGAN-GIF the colors are somewhat different (saturated) than that of the input video.
Comparison to HP-VAE-GAN [Gur et al.] Back to Top Back to Main Page
Input Video:
HP-VAE-GAN:
Ours (VGPNN):
Input Video:
HP-VAE-GAN:
Ours (VGPNN):
Input Video:
HP-VAE-GAN:
Ours (VGPNN):
Input Video:
HP-VAE-GAN:
Ours (VGPNN):
Input Video:
HP-VAE-GAN:
Ours (VGPNN):
Input Video:
HP-VAE-GAN:
Ours (VGPNN):
Input Video:
HP-VAE-GAN:
Ours (VGPNN):
Input Video:
HP-VAE-GAN:
Ours (VGPNN):
Input Video:
HP-VAE-GAN:
Ours (VGPNN):
Input Video:
HP-VAE-GAN:
Ours (VGPNN):



Comparison to SinGAN-GIF [Arora & Lee] Back to Top Back to Main Page
Input Video:
SinGAN-GIF:
Ours (VGPNN):
Input Video:
SinGAN-GIF:
Ours (VGPNN):
Input Video:
SinGAN-GIF:
Ours (VGPNN):
Input Video:
SinGAN-GIF:
Ours (VGPNN):
Input Video:
SinGAN-GIF:
Ours (VGPNN):
Back to Top Back to Main Page