Unsupervised Cross-Domain Image Generation
read the original abstract
We study the problem of transferring a sample in one domain to an analog sample in another domain. Given two related domains, S and T, we would like to learn a generative function G that maps an input sample from S to the domain T, such that the output of a given function f, which accepts inputs in either domains, would remain unchanged. Other than the function f, the training data is unsupervised and consist of a set of samples from each domain. The Domain Transfer Network (DTN) we present employs a compound loss function that includes a multiclass GAN loss, an f-constancy component, and a regularizing component that encourages G to map samples from T to themselves. We apply our method to visual domains including digits and face images and demonstrate its ability to generate convincing novel images of previously unseen entities, while preserving their identity.
This paper has not been read by Pith yet.
Forward citations
Cited by 3 Pith papers
-
Stabilizing In-Context Multi-Source Domain Adaptation for Biomedical Images Through Controls
Meta-learning with in-context control samples closes the domain gap for mechanism-of-action classification, raising accuracy on new batches from 0.862 to 0.935 on the JUMP-CP dataset.
-
Image-to-Image Translation Framework Embedded with Rotation Symmetry Priors
Rotation-equivariant convolutions and adaptive TL-Conv layers are added to I2I networks to preserve rotation symmetry and improve translation quality across domains.
-
Stabilizing In-Context Multi-Source Domain Adaptation for Biomedical Images Through Controls
CS-ARM-BN uses negative control samples to stabilize Batch Normalization statistics in a meta-learning framework, achieving robust MoA classification on new experimental batches under label shift and small sample sizes.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.