Skip to content

asabuncuoglu13/sketch-embeddings

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Exploring Deep Generative Sketch Networks

All the code and resources here have one goal, satisfying our curiosity. I conduct field studies to explore the possible improvements in deep generative models. What do humans expect from a reconstruction task? What are the expectations from interpolation? ... If we can answer these questions, we can answer meaningful metrics for generative techniques. I believe that current sketch models focus too much on popular models produced in the image generation area. But, image is a more objective area. The end result is an output of a machine. Sketching is a unique process for each of us.

Let's first explore the recent generative models. We picked four generative models: Autoencoder, DCGAN, SketchRNN and Sketchformer. The first two take pixel values and the last two take stroke-based representations. We will use Quick, Draw! dataset in our training process. I think it is not an ideal dataset, but this is the best sketching dataset until someone comes with a better one. By the way, playing with Quick, Draw! dataset is extremely fun. I only think that it is not the most representative dataset to train generative models.

To understand the area, you can check the following work. I strongly recommend playing with the explorables to better understand the possibilities. You can also think of that as a fun EDA to explore Quick, Draw! dataset.

Explorables on Quick, Draw!

Explorables on Representation Visualization

What can you find in our repo?

About

Exploring Sketch Embeddings

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published