Get Started
Home
Topics
Search
Library
Research questionHow can vision models trained on one labeled domain generalize to unseen domains with difficult, semantically valid variations?With only one labeled source domain, ordinary augmentation may not expose a classifier to difficult but class-consistent visual changes. Broadening variation can also introduce samples whose semantics drift away from the intended class.
Computer Vision
Diffusion Models
Image Generation
Machine Learning
Latest papersRecent research connected to this question, newest first.PAPT++: Risk-Aware Adversarial Tuning and Generation for Single Domain GeneralizationThe paper studies this problem using a pretrained text-to-image diffusion model, image-text alignment, classifier-guided synthesis, and iterative updates from generated and source images. Evidence is reported on standard single-domain generalization benchmarks.research paper · Sep 4, 2026
Related questions
How can low-level vision generalist models adapt to previously unseen tasks from a single example?How can parking-space classifiers generalize across visual environments with limited labeled target-domain data?How can vision-language models maintain visual recognition when modalities are missing and source training data is unavailable?How can object detectors trained on uncamouflaged imagery generalize to unseen camouflage with little target-domain data?