CLIPTraVeLGAN for Semantically Robust Unpaired Image Translation

EasyChair Preprint 9933

Abstract

In this paper a novel approach for semantically robust unpaired image translation is presented. CLIPTraVeLGAN replaces the Siamese network in TraVeLGAN with a contrastively pretrained language-image model (CLIP) with frozen weights. This approach significantly simplifies the model selection and training process of TraVeLGAN, making it more robust and easier to use.

Keyphrases: CLIP, GAN, Transfer Knowledge, image-to-image translation

@booklet{EasyChair:9933,
  author    = {Yevgeniy Bodyanskiy and Nataliya Ryabova and Roman Lavrynenko},
  title     = {CLIPTraVeLGAN for Semantically Robust Unpaired Image Translation},
  howpublished = {EasyChair Preprint 9933},
  year      = {EasyChair, 2023}}