1. Take a look at this url to see how Dilation really works (image is 2/3rds of the way down the page): https://github.com/vdumoulin/conv_arithmetic/blob/master/README.md

  2. Use the "dilation" parameter in your 2D Convolutional filters: https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html

  3. I've found it works best when only applied to the last layer in my architecture. The intuition is that by increasing the receptive field of your convolutional filter, you enable it to take in more "context" from the surrounding pixels. This has given my images more "cohesion" and increased the resolution of the outputs.

  4. I've only gotten it to work well on the Discriminators, though... YMMV

Covers: implementation of Dilation
Questions this item addresses:
  • What does increasing the dilation of your convolutional filters do?
Author(s) / creator(s) / reference(s)
Pytorch and vdumoulin on Github
0 comment

Quick Tips To Tune Your Gans

Total time needed: ~20 minutes
Help you get started in developing and tuning GANs which generate high-resolution artwork
Potential Use Cases
image generation of high-resolution artwork
Who is This For ?
INTERMEDIATEPython developers with some experience with Deep Neural Nets and want to try their hands at generating artwork
Click on each of the following annotated items to see details.
WRITEUP 1. GAN - Ways to improve GAN performance
  • What are the common ways to tune GANs?
10 minutes
WRITEUP 2. A couple of good ways to deal with limited datasets....
  • What happens when I don't have enough data?
20 minutes
REPO 3. Use DeepSpeed
  • What are the ways that I can parallelize and speed up my GANs' training?
10 minutes
WRITEUP 4. Use Dilations on your Discriminator to increase the receptive field size of your convolutional filters.
  • What does increasing the dilation of your convolutional filters do?
10 minutes

Concepts Covered

0 comment