@haltakov cool content. you can also mention the dice loss -- which is a differentiable version of intersection over union segmentation accuracy measure
Excited to share our recent work on unsupervised image segmentation:
Segmentation in Style: Unsupervised Semantic Image Segmentation with StyleGAN and CLIP
https://t.co/HskdAIkItF
We are able to discover semantic regions that coincide with some regions defined by human.
Segmentation in Style: Unsupervised Semantic Image Segmentation with Stylegan and CLIP
pdf: https://t.co/KSjm8txURN
abs: https://t.co/olELgzAMqI
a method that allows to automatically segment images into semantically meaningful regions without human supervision