LatentCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions

2021-10-06

ICCV 2021Proceedings of the IEEE International Conference on Computer Vision

LatentCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions

Oğuz Kaan Yüksel*Equal contribution·Enis Simsar*Equal contribution·Ezgi Gülperi Er·Pınar Yanardağ
October 2021
PaperCodeTalkWebsite

Research context

Open in graph

Addresses

Mechanistic interpretability

Uses

Generative adversarial network

Studies

Interpretable direction

Produces

AlgorithmContrastive learning

LatentCLR

A contrastive learning algorithm for discovering semantically meaningful directions in pretrained GAN latent spaces without manual direction labels.

Related publications