Threads
Update 07/05/2020: Made the interpolation GIF responsive, as well as added the YouTube video link.
This blog post will be short, as I have a few announcements:
Ph.D.
I’ve started a Ph.D. in Computer Science at the Autonomous University of Barcelona (UAB)! Specifically I will be doing my research at the Computer Vision Center, in the research group of Advanced Driver Assistance Systems or ADAS, under the guidance of Dr. Antonio López Peña. I am excited to be a part of both this research group and the CVC, and know that I will learn a lot from them!
Threads
Threads (2019) is a continuation of Latent Fabrics (2018) and as such, wishes to explore the similarities in the latent space of the threads of different cultures in Guatemala and Mexico. I focused now on the details of the different threads used in different regions, and over 3000 images where obtained from online and private collections.
However, I didn’t wish to simply resize each image, so I cut different square regions of each image, resulting in exactly 11852 square images which I then fed to a ProGAN. The image sizes varied, but some had to be resized and I decided to start with images of size $128\times128$.
A small interpolation video has been generated after ~10 days of training on a burrowed GTX 1080. In total, the Discriminator saw around 12 million images, which would not have been possible in the same timeframe with, e.g. DCGAN.
Interpolation video between random latent vectors.
You can find a longer version of this GIF in the following YouTube video:
This work couldn’t have been possible without me getting permission to use the collected images from the following:
- Karen Elwell’s Flickr collection
- Huipils.com by David and Sally Hamilton
- The Minneapolis Institute of Art
- The Museum of World Culture (to be used further on)
- The Burke Museum of Natural History and Culture
- The British Museum
- The National Museum of Ethnology (to be used further on)
As always, leave a like/comment and hopefully I’ll be able to continue writing more blog posts.
Cheers!