2 Hits in 1.3 sec

CoMoGAN: continuous model-guided image-to-image translation [article]

Fabio Pizzati, Pietro Cerri, Raoul de Charette
2022 arXiv   pre-print
We rely on naive physics-inspired models to guide the training while allowing private model/translations features.  ...  CoMoGAN can be used with any GAN backbone and allows new types of image translation, such as cyclic image translation like timelapse generation, or detached linear translation.  ...  Our contributions are: • a novel model-guided setting for continuous i2i, • CoMoGAN: an unsupervised framework for disentanglement of continuously evolving features in generated images, using simple model  ... 
arXiv:2103.06879v3 fatcat:ljkxntuadnhzhcaolxnqrhfavm

ManiFest: Manifold Deformation for Few-shot Image Translation [article]

Fabio Pizzati, Jean-François Lalonde, Raoul de Charette
2022 arXiv   pre-print
Most image-to-image translation methods require a large number of training images, which restricts their applicability.  ...  In addition to the general few-shot translation task, our approach can alternatively be conditioned on a single exemplar image to reproduce its specific style.  ...  Extensions Few-shot continuous manifolds We investigate the use of ManiFest for performing continuous image translation as in CoMoGAN [40] , thus learning the transformation from S=day to A m =night  ... 
arXiv:2111.13681v3 fatcat:imw7sp3b5nbyjignhqtjgwzl5u